{"id":6240,"date":"2026-10-19T14:46:10","date_gmt":"2026-10-19T14:46:10","guid":{"rendered":"https:\/\/brandmentions.com\/blog\/?p=6240&#038;preview=true&#038;preview_id=6240"},"modified":"2026-10-19T14:50:38","modified_gmt":"2026-10-19T14:50:38","slug":"track-ai-citations-pr-campaign","status":"publish","type":"post","link":"https:\/\/brandmentions.com\/blog\/track-ai-citations-pr-campaign\/","title":{"rendered":"How to Track AI Citations After a PR Campaign (and Measure the Impact)"},"content":{"rendered":"<p><strong><em>To track AI citations after a PR campaign, freeze a set of buyer-intent prompts, record a multi-week baseline citation rate across ChatGPT, Perplexity, Gemini and Google AI Overviews, then re-run the same prompts on the same schedule after coverage goes live.<\/em><\/strong><\/p>\n<p>Verify each cited URL against your coverage register at URL level, not domain level, and compare the change against an untouched control topic. One run is an anecdote; a citation rate from repeated runs under fixed conditions is a number you can defend.<\/p>\n<p>The six steps to track AI Citations after a PR campaign in short:<\/p>\n<ol>\n<li>Establish your AI visibility baseline (a range, not a number).<\/li>\n<li>Choose relevant prompts and AI search platforms, then freeze them.<\/li>\n<li>Monitor earned media coverage and brand mentions in a coverage register.<\/li>\n<li>Identify AI answers that cite your PR coverage, and verify every citation.<\/li>\n<li>Compare brand mentions and citation rates after launch, with honest denominators.<\/li>\n<li>Compare results against competitors or a control group before claiming lift.<\/li>\n<\/ol>\n<h2 id=\"table-of-contents\">Table of Contents<\/h2>\n<ul>\n<li><a href=\"#key-takeaways-tracking-ai-citations-after-pr-coverage\">Key takeaways: tracking AI citations after PR coverage<\/a><\/li>\n<li><a href=\"#ai-citation-vs-brand-mention-vs-ranking-whats-the-difference\">AI citation vs brand mention vs ranking: what's the difference?<\/a><\/li>\n<li><a href=\"#can-pr-campaigns-influence-ai-search-citations\">Can PR campaigns influence AI search citations?<\/a><\/li>\n<li><a href=\"#why-cant-one-prompt-check-prove-ai-visibility\">Why can't one prompt check prove AI visibility?<\/a><\/li>\n<li><a href=\"#how-long-does-pr-coverage-take-to-become-eligible-for-ai-citations\">How long does PR coverage take to become eligible for AI citations?<\/a><\/li>\n<li><a href=\"#what-should-you-track-after-a-pr-campaign-the-10-core-kpis\">What should you track after a PR campaign? The 10 core KPIs<\/a><\/li>\n<li><a href=\"#which-ai-citation-measurement-approach-fits-your-situation\">Which AI citation measurement approach fits your situation?<\/a><\/li>\n<li><a href=\"#how-to-track-ai-citations-after-a-pr-campaign-6-steps\">How to track AI citations after a PR campaign: 6 steps<\/a>\n<ul>\n<li><a href=\"#step-1-establish-your-ai-visibility-baseline\">Step 1. Establish your AI visibility baseline<\/a><\/li>\n<li><a href=\"#step-2-choose-relevant-prompts-and-ai-search-platforms\">Step 2. Choose relevant prompts and AI search platforms<\/a><\/li>\n<li><a href=\"#step-3-monitor-earned-media-coverage-and-brand-mentions\">Step 3. Monitor earned media coverage and brand mentions<\/a><\/li>\n<li><a href=\"#step-4-identify-ai-answers-that-cite-your-pr-coverage\">Step 4. Identify AI answers that cite your PR coverage<\/a><\/li>\n<li><a href=\"#step-5-compare-brand-mentions-and-citation-rates-after-launch\">Step 5. Compare brand mentions and citation rates after launch<\/a><\/li>\n<li><a href=\"#step-6-compare-results-against-competitors-or-a-control-group\">Step 6. Compare results against competitors or a control group<\/a><\/li>\n<\/ul>\n<\/li>\n<li><a href=\"#how-to-calculate-pr-campaign-impact-on-ai-visibility-the-9-formulas\">How to Measure AI Citation Rate, Brand Mentions, and PR Campaign Impact<\/a><\/li>\n<li><a href=\"#what-can-you-claim-from-your-ai-citation-results-the-evidence-ladder\">What can you claim from your AI citation results? The evidence ladder<\/a><\/li>\n<li><a href=\"#what-should-an-ai-citation-tracking-dashboard-include\">What should an AI citation tracking dashboard include?<\/a>\n<ul>\n<li><a href=\"#ai-visibility-and-geo-tools-for-tracking-citations-after-pr\">AI visibility and GEO tools for tracking citations after PR<\/a><\/li>\n<\/ul>\n<\/li>\n<li><a href=\"#example-tracking-ai-citations-after-a-product-launch-pr-campaign\">Example: tracking AI citations after a product launch PR campaign<\/a><\/li>\n<li><a href=\"#why-isnt-my-pr-coverage-showing-up-in-ai-citations-8-causes\">Why isn't my PR coverage showing up in AI citations? 8 causes<\/a><\/li>\n<li><a href=\"#how-to-get-cited-by-ai-in-your-next-pr-campaign-5-levers\">How to get cited by AI in your next PR campaign: 5 levers<\/a><\/li>\n<li><a href=\"#frequently-asked-questions\">Frequently asked questions<\/a><\/li>\n<li><a href=\"#related-guides\">Related guides<\/a><\/li>\n<li><a href=\"#conclusion-measure-the-rate-not-the-moment\">Conclusion: measure the rate, not the moment<\/a><\/li>\n<\/ul>\n<p>A campaign closes, the coverage report looks healthy, and then someone senior asks whether any of it showed up in ChatGPT. Most teams answer by typing one prompt and taking a screenshot - a single draw from a distribution, presented as a result. This guide replaces that with a system you can operate: a frozen prompt portfolio, a real baseline, repeated runs, honest denominators, and a clear line between \"visibility went up\" and \"the campaign caused it.\" It is written for the marketer or comms lead who has to defend a number in a meeting, not for a data scientist.<\/p>\n<h2 id=\"key-takeaways-tracking-ai-citations-after-pr-coverage\">Key Takeaways: Tracking AI Citations After PR Coverage<\/h2>\n<ul>\n<li><strong>Earned coverage feeds AI answers, but its weight depends on the question.<\/strong> <a href=\"https:\/\/muckrack.com\/blog\/what-is-ai-reading-new-insights-may\" target=\"_blank\" rel=\"nofollow noopener noreferrer\">Muck Rack's \"What Is AI Reading?\" research<\/a>, built on more than 1 million cited links, found journalism cited in 46% of responses to industry-trend questions, more than twice the rate for how-to and comparative questions.<\/li>\n<li><strong>One check proves nothing, because AI citation sets churn constantly.<\/strong> <a href=\"https:\/\/www.yotpo.com\/blog\/how-long-does-an-ai-citation-last\/\" target=\"_blank\" rel=\"nofollow noopener\">Yotpo\u2019s citation stability study<\/a> found that even ChatGPT, the most consistent engine it measured, scored only 0.30 on week-to-week source stability. This indicates substantial week-to-week variation in the sources ChatGPT cites. The results reinforce why marketers should measure citation visibility across repeated observations rather than relying on a single answer.<\/li>\n<li><strong>There is no documented six-to-twelve-month \"eligibility\" waiting period.<\/strong> <a href=\"https:\/\/developers.google.com\/search\/docs\/fundamentals\/ai-optimization-guide\" target=\"_blank\" rel=\"noopener noreferrer\">Google's AI features guidance<\/a> specifies indexing, snippet eligibility and inclusion in Search's generative features, and states plainly that meeting the requirements does not guarantee inclusion. Start observing the day coverage goes live.<\/li>\n<li><strong>Off-site mentions correlate more strongly with AI Overview visibility than backlinks do.<\/strong> <a href=\"https:\/\/ahrefs.com\/blog\/ai-overview-brand-correlation\/\" target=\"_blank\" rel=\"nofollow noopener noreferrer\">Ahrefs' study of 75,000 brands<\/a> reported a 0.664 Spearman correlation for brand web mentions versus 0.218 for backlinks, and Ahrefs states explicitly that correlation is not causation.<\/li>\n<li><strong>Small samples produce wide intervals, so thresholds matter.<\/strong> <a href=\"https:\/\/www.itl.nist.gov\/div898\/software\/dataplot\/refman1\/auxillar\/propconf.htm\" target=\"_blank\" rel=\"noopener noreferrer\">NIST's published binomial example<\/a> shows 8 successes in 30 observations gives a 26.7% observed rate with a 95% Wilson interval of roughly 14.2% to 44.4%.<\/li>\n<li><strong>Google now reports AI Overview and AI Mode impressions to site owners.<\/strong> Search Console\u2019s generative AI performance reports launched June 3, 2026 and reached all sites worldwide on August 31, 2026, but they only cover your own pages. Pair them with prompt tracking and mention monitoring.<\/li>\n<\/ul>\n<h2 id=\"ai-citation-vs-brand-mention-vs-ranking-whats-the-difference\">AI Citation vs Brand Mention vs Ranking: What's the Difference?<\/h2>\n<p>Three things get collapsed into \"AI visibility,\" and the collapse is where most PR reports go wrong. Fix the vocabulary before you log a single row.<\/p>\n<p><strong>What is an AI citation?<\/strong> An AI citation is a source link that a generative engine displays alongside an answer, presented as a source or supporting reference for a statement in that answer.<\/p>\n<p><strong>What is an AI brand mention?<\/strong> An AI brand mention is your brand, product or executive named in the generated answer text, with or without any accompanying link.<\/p>\n<p><strong>What is AI search visibility?<\/strong> AI search visibility is the extent to which your brand appears - as a mention, a citation or a recommendation - inside the answers generative engines produce for a defined set of questions. \"LLM visibility\" describes the same outcome from the model's side: whether your brand and its supporting evidence are present in what the assistant generates and in the sources it shows.<\/p>\n<p><strong>What is generative engine optimization (GEO)?<\/strong> Generative engine optimization (GEO) is the practice of making your content, and the third-party evidence about your brand, retrievable, relevant and quotable for generative engines. <strong>AI visibility tracking is the measurement half of GEO:<\/strong> the repeated, documented observation of mentions and citations under fixed conditions. This playbook is the PR-specific version of that measurement discipline.<\/p>\n<table>\n<thead>\n<tr>\n<th>Outcome<\/th>\n<th>What you observe<\/th>\n<th>What it actually establishes<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Brand mention<\/td>\n<td>The answer names your brand<\/td>\n<td>Your brand appeared in that answer, that time<\/td>\n<\/tr>\n<tr>\n<td>Source citation<\/td>\n<td>The answer links to a page<\/td>\n<td>The platform presented that page as a source<\/td>\n<\/tr>\n<tr>\n<td>Recommendation<\/td>\n<td>The answer suggests your brand for a stated need<\/td>\n<td>The platform put you inside the advice, not just the context<\/td>\n<\/tr>\n<tr>\n<td>Position<\/td>\n<td>Where your brand appears among named alternatives<\/td>\n<td>Order within one answer, not a stable ranking<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>These move independently, and that is the point. AI brand mentions and source citations measure different types of visibility. A brand mention shows that an AI-generated answer references your business, while a citation identifies a source presented to support the response. Tracking both reveals whether your brand is appearing in answers and which published sources are contributing to that visibility. A campaign can lift one and leave the other flat, which is a diagnosis, not a failure.<\/p>\n<p>One distinction trips up almost every first report. A citation of your coverage is usually a citation of the publisher's domain, not yours. If a trade publication covers your launch and Perplexity cites that article, you earned a third-party citation that your own analytics will never see.<\/p>\n<p>Both belong in the log, tagged separately. The relationship between <a href=\"https:\/\/brandmentions.com\/blog\/brand-mentions-ai-visibility\/\">mentions and AI visibility<\/a> only becomes measurable when you stop forcing every kind of appearance into one \"ranking\" field.<\/p>\n<p><strong>For this playbook, \"brand-supporting citation\" means a verified source link that supports a statement about your brand.<\/strong> Ordinary source appearances get counted separately. That one definition prevents more inflated reports than any tool you could buy.<\/p>\n<h2 id=\"can-pr-campaigns-influence-ai-search-citations\">Can PR Campaigns Influence AI Search Citations?<\/h2>\n<p>Yes. Earned coverage creates publicly accessible material that AI search systems can discover, retrieve and reference. Without that coverage, there is nothing for an engine to cite about you except your own website and whatever already exists.<\/p>\n<p>The mechanism runs as a five-stage funnel, and your campaign can break at any stage. This is the canonical funnel used throughout this guide, and it is the same one you measure drop-off against in Step 4:<\/p>\n<ol>\n<li><strong>Published<\/strong> - the coverage is live and publicly accessible.<\/li>\n<li><strong>Discoverable<\/strong> - the page is reachable by the relevant crawler, with no block, hard paywall or login wall in the way.<\/li>\n<li><strong>Selected<\/strong> - the page is retrieved and selected as a source for one of the specific questions in your prompt portfolio.<\/li>\n<li><strong>Included<\/strong> - your brand, product or campaign claim appears in the passage that citation supports.<\/li>\n<li><strong>Displayed<\/strong> - the citation is shown in the answer, and in the best case can be associated with an identifiable visit.<\/li>\n<\/ol>\n<p>Ahrefs' analysis of 75,000 brands is the clearest quantification of the input side. Brand web mentions showed a 0.664 Spearman correlation with AI Overview brand visibility, ahead of branded anchors and brand search volume, with backlinks at 0.218. Ahrefs is explicit that this is correlation, not a causal mechanism, and that framing matters when you take it to leadership.<\/p>\n<p>Muck Rack's research describes the output side. Across more than 1 million cited links, non-paid sources dominated the citation mix, and journalism's share varied sharply by question type: 46% of responses to industry-trend questions versus less than half that for how-to and comparative questions. Read that as a targeting instruction. PR moves the needle hardest on the questions where journalism is the natural evidence.<\/p>\n<p>Now the honest part. <strong>Coverage does not guarantee citation.<\/strong> The engine still has to retrieve the page, judge it relevant to the specific wording of the question, and select it over everything else competing for that slot.<\/p>\n<p>There is also a widespread misreading worth correcting here. Cognizo's citation study found brand-owned domains accounted for roughly 12% of AI Overview citations in its sample of US AI-visibility software queries. That has been retold as \"88% of AI citations are journalism.\" It is not. The remaining share is everything outside the tracked brands' own domains, including forums, reviews, aggregators and documentation, with editorial and press pages a much smaller slice of the taxonomy.<\/p>\n<p>Understanding <a href=\"https:\/\/brandmentions.com\/blog\/how-ai-search-engines-recommend-brands\/\">how AI recommends brands<\/a> starts with accepting that your coverage competes with Reddit threads, review sites and vendor documentation, not only with other press.<\/p>\n<blockquote><p>PR coverage creates additional opportunities for your brand to appear in AI-generated answers, particularly when reputable publications provide relevant, accessible, and well-supported information. However, publication alone does not guarantee that AI search engines will retrieve or cite that coverage.<\/p><\/blockquote>\n<h2 id=\"why-cant-one-prompt-check-prove-ai-visibility\">Why Can't One Prompt Check Prove AI Visibility?<\/h2>\n<p>Because the answer you got today is one sample, and the sample-to-sample variation is larger than most campaign effects.<\/p>\n<p>Yotpo\u2019s stability data puts a number on it: ChatGPT\u2019s week-to-week citation overlap scored just 0.30, and on Google AI Mode only 11% of sources remained cited three weeks later. The source set is not stable, even on the most consistent engine.<\/p>\n<p>CiteLens tested the steadier case: 493 questions asked of Google's AI three times each, back to back. The cited source set was fully identical only 81% of the time, with average source-set similarity of 84%. That is the best-case scenario, under immediate repetition, on the most consistent surface.<\/p>\n<p><strong>The churn floor: if your observed change is smaller than your own baseline's week-to-week range, it is not lift yet.<\/strong><\/p>\n<p>The practical consequence is simple and it governs everything below. One run gives you an anecdote. Repeated runs give you a rate. Rates can be compared across time; anecdotes cannot.<\/p>\n<blockquote><p>Never report an AI citation from a single prompt run. Report a citation rate from repeated runs under fixed conditions, or report that you do not yet know.<\/p><\/blockquote>\n<h2 id=\"how-long-does-pr-coverage-take-to-become-eligible-for-ai-citations\">How Long Does PR Coverage Take to Become Eligible for AI Citations?<\/h2>\n<p>Shorter than the industry folklore suggests, and less predictable.<\/p>\n<p>Google's guidance for AI features specifies three conditions: the page must be indexed, eligible to appear with a Search snippet, and not excluded from Search's generative features. Google also states that meeting every requirement does not guarantee indexing or inclusion. There is no published minimum page age, brand age or waiting period.<\/p>\n<p>The commonly repeated \"six to twelve months before you are eligible\" is not a platform rule. What is real is cumulative mention density: brands that dominate category answers have usually been accumulating third-party coverage for years. That is a strategic timeline, not an eligibility gate, and conflating the two makes teams wait when they should be measuring.<\/p>\n<p>A second misreading needs retiring too. <a href=\"https:\/\/ahrefs.com\/blog\/do-ai-assistants-prefer-to-cite-fresh-content\/\" target=\"_blank\" rel=\"nofollow noopener noreferrer\">Ahrefs' study of 16.975 million cited URLs<\/a> found the average AI-cited page was 1,064 days old, with ChatGPT at 958 days and Google AI Overviews at 1,432 days. That measures <strong>the age of pages engines choose to cite<\/strong>. It does not measure how long your article waits before its first citation. Treat it as evidence that your three-week-old article competes against an established corpus, not as a prediction of delay.<\/p>\n<p>Separate three questions and you will stop arguing about timelines:<\/p>\n<ol>\n<li><strong>Can the platform access the material?<\/strong> Waiting does not fix a robots.txt block or a hard paywall.<\/li>\n<li><strong>Does it select that material for the questions you track?<\/strong> Relevance, not patience.<\/li>\n<li><strong>Does that selection persist across repeated observations?<\/strong> A first citation is not durable visibility.<\/li>\n<\/ol>\n<p>Fast pickup does happen. Otterly.ai documented a press release cited by Microsoft Copilot two days after publication, with other engines following within the week, and\u00a0<a href=\"https:\/\/scrunch.com\/blog\/half-life-of-ai-citations\" target=\"_blank\" rel=\"nofollow noopener\">Scrunch and Stacker\u2019s analysis of 3.5 million citation events<\/a> found that AI citation activity halves in about 4.5 weeks, while citations from domains in Stacker\u2019s news-publisher network last about twice as long as other domains.<\/p>\n<p>Record the date your tracker first sees a citation and label it <strong>first observed<\/strong>. Your collection schedule cannot establish the moment a platform discovered a page, and pretending otherwise invites a question you cannot answer.<\/p>\n<h2 id=\"what-should-you-track-after-a-pr-campaign-the-10-core-kpis\">What Should You Track After a PR Campaign? The 10 Core KPIs<\/h2>\n<p>Pick one primary outcome before launch. For most campaigns, use <strong>brand-supporting citation rate on unbranded buyer-intent prompts<\/strong>. Everything else is diagnostic, there to explain why the primary number moved.<\/p>\n<table>\n<thead>\n<tr>\n<th>KPI<\/th>\n<th>Working definition<\/th>\n<th>What it helps you decide<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Brand mention rate<\/td>\n<td>Valid runs naming your brand, as a percentage of all valid runs<\/td>\n<td>Whether the engine associates you with the topic at all<\/td>\n<\/tr>\n<tr>\n<td>Brand-supporting citation rate<\/td>\n<td>Valid runs containing a verified citation that supports a brand claim<\/td>\n<td>Whether visibility carries identifiable supporting evidence<\/td>\n<\/tr>\n<tr>\n<td>Campaign coverage citation rate<\/td>\n<td>Valid runs citing at least one verified campaign placement<\/td>\n<td>Whether campaign material entered the sampled answers<\/td>\n<\/tr>\n<tr>\n<td>AI share of voice<\/td>\n<td>Your brand-answer appearances divided by appearances for all brands in a fixed set<\/td>\n<td>Whether your relative position changed<\/td>\n<\/tr>\n<tr>\n<td>Citation share<\/td>\n<td>Qualifying citation instances for you, divided by all qualifying citation instances<\/td>\n<td>How much of the source mix you occupy<\/td>\n<\/tr>\n<tr>\n<td>Cited domains and source mix<\/td>\n<td>Publishers, owned pages, reviews, forums, documentation<\/td>\n<td>Where the next campaign should be placed<\/td>\n<\/tr>\n<tr>\n<td>Position and recommendation status<\/td>\n<td>First appearance among named alternatives, and whether the answer recommends you<\/td>\n<td>Whether exposure reaches the decision part of the answer<\/td>\n<\/tr>\n<tr>\n<td>Sentiment and accuracy<\/td>\n<td>Tone toward the brand, and correctness of material claims<\/td>\n<td>Whether more exposure is helping or hurting<\/td>\n<\/tr>\n<tr>\n<td>Referral traffic and outcomes<\/td>\n<td>Identifiable AI-referred visits, landing pages, subsequent actions<\/td>\n<td>Whether visibility connects to business activity<\/td>\n<\/tr>\n<tr>\n<td>Earned media volume and sentiment<\/td>\n<td>New articles, posts and threads mentioning the brand, with tone<\/td>\n<td>Whether the campaign created retrievable supply at all<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Two rules on this table. Never add AI share of voice to media share of voice; they use different populations and denominators. And never compress these into a single \"AI visibility score,\" because a composite hides which dimension broke, and the fix for a mention problem is nothing like the fix for a citation problem.<\/p>\n<p><strong>Earned-media monitoring vs AI-answer monitoring: the core difference.<\/strong> The first measures what the public web says about you. The second measures what appears inside a defined sample of AI answers. The first is the supply; the second is the outcome. Track only the outcome and you will never be able to explain a flat result. This distinction is the reason every step below runs two baselines on one calendar.<\/p>\n<h2 id=\"which-ai-citation-measurement-approach-fits-your-situation\">Which AI Citation Measurement Approach Fits Your Situation?<\/h2>\n<ul>\n<li><strong>Best for proving campaign contribution to a board:<\/strong> Exact coverage-to-citation matching layered on a frozen prompt portfolio. It connects a visible AI answer to a specific placement URL, which is the only evidence that survives someone re-running your prompt live in the meeting.<\/li>\n<li><strong>Best for separating real lift from platform noise:<\/strong> Repeated pre\/post measurement against an untouched comparison topic or market. Volatility in this channel is large enough to manufacture both fake wins and fake losses, so the comparison group is what makes the number defensible.<\/li>\n<li><strong>Best for a two-person team with zero budget:<\/strong> One commercially important topic, five frozen prompts, four platforms, run manually in a shared sheet. Narrow and repeated beats broad and sampled once.<\/li>\n<li><strong>Best for reputation-sensitive launches:<\/strong> Claim-level accuracy and sentiment review on every captured answer. An enthusiastic answer containing wrong pricing is a liability that a mention-rate chart will never surface.<\/li>\n<li><strong>Best for enterprise scale across markets:<\/strong> API or vendor collection with a documented capture method, plus human review of a sampled subset. Automation becomes unavoidable past roughly 50 prompts across 4 platforms with weekly repetition.<\/li>\n<li><strong>Best for diagnosing why good coverage produced nothing:<\/strong> The coverage-to-citation funnel in Step 4. It tells you whether the break is access, relevance, source selection or message inclusion, which are four different fixes with four different budgets.<\/li>\n<\/ul>\n<p><img decoding=\"async\" src=\"https:\/\/brandmentions.com\/blog\/wp-content\/uploads\/2026\/10\/image_02_ddd7a9cb.webp\" alt=\"Six-step workflow for tracking AI citations after a PR campaign.\" \/><\/p>\n<h2 id=\"how-to-track-ai-citations-after-a-pr-campaign-6-steps\">How to Track AI Citations After a PR Campaign: 6 Steps<\/h2>\n<ol>\n<li><a href=\"#step-1-establish-your-ai-visibility-baseline\">Establish your AI visibility baseline<\/a><\/li>\n<li><a href=\"#step-2-choose-relevant-prompts-and-ai-search-platforms\">Choose relevant prompts and AI search platforms<\/a><\/li>\n<li><a href=\"#step-3-monitor-earned-media-coverage-and-brand-mentions\">Monitor earned media coverage and brand mentions<\/a><\/li>\n<li><a href=\"#step-4-identify-ai-answers-that-cite-your-pr-coverage\">Identify AI answers that cite your PR coverage<\/a><\/li>\n<li><a href=\"#step-5-compare-brand-mentions-and-citation-rates-after-launch\">Compare brand mentions and citation rates after launch<\/a><\/li>\n<li><a href=\"#step-6-compare-results-against-competitors-or-a-control-group\">Compare results against competitors or a control group<\/a><\/li>\n<\/ol>\n<p>Each step gets the same treatment: what to do, the decision thresholds, the failure modes, and one thing that is not commonly discussed.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/brandmentions.com\/blog\/wp-content\/uploads\/2026\/10\/image_03_f546a7ae.webp\" alt=\"AI visibility baseline range tracked across repeated prompt runs.\" \/><\/p>\n<h3 id=\"step-1-establish-your-ai-visibility-baseline\">Step 1. Establish Your AI Visibility Baseline<\/h3>\n<p><strong>Do this: run your frozen prompt portfolio for four weeks before launch and record, for every prompt and platform, both the mean rate and the week-to-week range.<\/strong><\/p>\n<p>The <a href=\"https:\/\/www.iab.com\/wp-content\/uploads\/2026\/07\/IAB_Measuring_Visibility_in_the_AI_Era_August_2026-1.pdf\" target=\"_blank\" rel=\"nofollow noopener noreferrer\">IAB's \"Measuring Visibility in the AI Era\" framework<\/a>, published August 3, 2026, is the most useful external anchor for this step. It groups AI visibility into the 4 P\u2019s (Presence, Prominence, Portrayal and Persuasion) and separates decision-grade data from directional data. A baseline built on one-off checks is directional at best. Document your variability, disclose your methodology and define your reset rules from day one, so your numbers meet the decision-grade bar when someone challenges them.<\/p>\n<p><strong>Your baseline is not a number. It is a range.<\/strong> For each prompt and platform, you need the mean rate and the week-to-week spread, measured with the exact protocol you will use afterwards.<\/p>\n<h4 id=\"write-the-measurement-contract-first\">Write the measurement contract first<\/h4>\n<p>Before any collection starts, write down and circulate:<\/p>\n<ul>\n<li>The campaign's commercial topic and target audience.<\/li>\n<li>Brand, product names and accepted spelling variations.<\/li>\n<li>The fixed competitor set.<\/li>\n<li>The single primary metric.<\/li>\n<li>Platforms, countries, languages.<\/li>\n<li>The exact prompt wording and the collection schedule.<\/li>\n<li>What counts as campaign-related coverage.<\/li>\n<li><strong>The smallest improvement that would change your investment decision.<\/strong><\/li>\n<\/ul>\n<p>That last line is the one teams skip, and it is the one that saves you. A change can be measurable and still not worth reinvesting behind. Decide the threshold while you are still neutral about the answer.<\/p>\n<h4 id=\"size-the-sample-before-you-collect-it\">Size the sample before you collect it<\/h4>\n<p>This is where most published guides stop at \"run it a few times.\" Work through four inputs instead:<\/p>\n<table style=\"width: 100%;\">\n<thead>\n<tr>\n<th style=\"width: 16.875%;\">Input<\/th>\n<th style=\"width: 31.25%;\">What to establish<\/th>\n<th style=\"width: 51.625%;\">Why it changes your design<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td style=\"width: 16.875%;\">Baseline prevalence<\/td>\n<td style=\"width: 31.25%;\">Your current rate, roughly, from a short pilot<\/td>\n<td style=\"width: 51.625%;\">Rates near 0% or near 100% need far more runs to detect movement than mid-range rates. Smaller changes generally require larger samples.<\/td>\n<\/tr>\n<tr>\n<td style=\"width: 16.875%;\">Smallest useful change<\/td>\n<td style=\"width: 31.25%;\">The percentage-point move that changes a budget decision<\/td>\n<td style=\"width: 51.625%;\">A 3-point target demands a much larger sample than a 15-point target<\/td>\n<\/tr>\n<tr>\n<td style=\"width: 16.875%;\">Clustering<\/td>\n<td style=\"width: 31.25%;\">Repeated runs share prompts, days and conditions<\/td>\n<td style=\"width: 51.625%;\">Your effective sample is smaller than your raw run count, so do not treat 120 runs as 120 independent survey responses<\/td>\n<\/tr>\n<tr>\n<td style=\"width: 16.875%;\">Collection cost<\/td>\n<td style=\"width: 31.25%;\">Analyst minutes or vendor credits per run<\/td>\n<td style=\"width: 51.625%;\">This is what forces the real trade-off between number of topics and number of repetitions<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Two facts make this tractable without a statistician. First, interval width shrinks with the square root of the number of observations, so quadrupling your runs roughly halves the uncertainty band. Second, NIST's worked example shows what a small sample buys you: 8 successes in 30 observations is a 26.7% rate with a 95% interval spanning roughly 14.2% to 44.4%. If your target lift is smaller than that band, 30 runs will not settle the question.<\/p>\n<p><strong>Practical starting point:<\/strong> five fixed prompts per topic, each repeated four to six times weekly, per platform. That is 20 to 30 runs per topic per platform per week, and 120 per window over four weeks. Spread repetitions across different days. If resources are tight, cut the number of topics, never the repetition.<\/p>\n<h4 id=\"pre-register-the-comparison-or-you-will-find-a-win\">Pre-register the comparison, or you will find a win<\/h4>\n<p>Four platforms times five metrics times twelve weeks is 240 possible comparisons. Something in there will look like a breakthrough by chance alone.<\/p>\n<p>Decide in advance: one primary metric, one primary platform-weighted comparison, two scheduled decision dates. Everything else is exploratory, labelled as exploratory, and used to generate hypotheses for the next campaign rather than conclusions about this one.<\/p>\n<h4 id=\"define-your-reset-triggers-now\">Define your reset triggers now<\/h4>\n<p>Baselines break for reasons that have nothing to do with your campaign. A model retirement is the obvious one. OpenAI retired GPT-4o from ChatGPT on February 13, 2026, which is exactly the kind of event that breaks a trend line. Guides still recommending it as a tracking surface are measuring something that no longer exists.<\/p>\n<p>Write a short reset protocol covering model retirements, citation-interface changes, and vendor switches:<\/p>\n<ol>\n<li>Log the event, the date, and which series it affects.<\/li>\n<li>Run a bridge window where both the old and new configurations are collected in parallel, if the old one still exists.<\/li>\n<li>If parallel collection is impossible, close the old series, open a new one, and keep the historical version visible in the report rather than quietly splicing them.<\/li>\n<li>Never compare across a discontinuity without flagging it in the same chart.<\/li>\n<\/ol>\n<p>Logging the model version is not enough. Without a bridge, you have two series pretending to be one.<\/p>\n<h4 id=\"build-the-coverage-baseline-alongside-it\">Build the coverage baseline alongside it<\/h4>\n<p>Run the coverage baseline on the same calendar as the prompt baseline - coverage is the supply side described in <a href=\"#what-should-you-track-after-a-pr-campaign-the-10-core-kpis\">the 10 core KPIs<\/a>. The first thing you will be asked when a number moves is \"compared to what level of coverage?\"<\/p>\n<p><a href=\"https:\/\/brandmentions.com\/\">BrandMentions<\/a> fits this specific layer: continuous, web-wide mention baselining across news, blogs, forums, websites and social channels, with competitor share of voice and sentiment on the same timeline and historical lookback for reconstructing the period before you started tracking. That is the demand-and-coverage record that AI-prompt dashboards do not hold, and it is the same record that makes the link between <a href=\"https:\/\/brandmentions.com\/blog\/brand-mentions-ai-visibility\/\">brand mentions and AI visibility<\/a> measurable rather than anecdotal. Be realistic about its boundary too: public-web monitoring is a sample of the accessible web, not a census of every private conversation, and update frequency and historical depth vary by plan. State that limitation in your methodology note rather than letting someone else discover it.<\/p>\n<p><strong>If the campaign has already launched:<\/strong> do not ask an AI platform what it \"would have said\" before. That reconstructs nothing. Recover genuinely archived answers, backfill the media record where historical coverage exists, start prospective collection immediately, find a comparison topic, and disclose the gap in writing.<\/p>\n<h3 id=\"step-2-choose-relevant-prompts-and-ai-search-platforms\">Step 2. Choose Relevant Prompts and AI Search Platforms<\/h3>\n<p><strong>Do this: fix five buyer-intent prompts per topic and lock the platform, interaction mode, country, language and device you will reuse in every single later run.<\/strong><\/p>\n<p><a href=\"https:\/\/help.openai.com\/en\/articles\/9237897-searching-the-web-with-chatgpt\" target=\"_blank\" rel=\"noopener noreferrer\">OpenAI's documentation on searching the web with ChatGPT<\/a> notes that search can use location and saved memories. That single fact tells you why \"I ran it in incognito\" is an insufficient description of your collection conditions.<\/p>\n<p><strong>What is a buyer-intent prompt portfolio?<\/strong> A buyer-intent prompt portfolio is a fixed set of questions, phrased the way a prospect would ask an assistant, that spans the stages of a real buying decision and is run unchanged across every measurement cycle.<\/p>\n<h4 id=\"start-from-customer-decisions-not-from-prompts-designed-to-find-you\">Start from customer decisions, not from prompts designed to find you<\/h4>\n<p>Pull your questions from sales call recordings, support tickets, customer research interviews and existing search data. Then validate the portfolio honestly: pick ten recent sales conversations and check whether your prompt set contains the questions those buyers actually asked. If it does not, you built a portfolio that measures your assumptions.<\/p>\n<p>Keep four groups with separate denominators:<\/p>\n<table>\n<thead>\n<tr>\n<th>Group<\/th>\n<th>Example<\/th>\n<th>What it tests<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Unbranded discovery<\/td>\n<td>\"Which employee scheduling tools suit field-service companies with 50 to 200 staff?\"<\/td>\n<td>Whether you enter the consideration set at all<\/td>\n<\/tr>\n<tr>\n<td>Branded evaluation<\/td>\n<td>\"Is [your brand] reliable for shift-based teams?\"<\/td>\n<td>Accuracy and sentiment, not discovery<\/td>\n<\/tr>\n<tr>\n<td>Named comparison<\/td>\n<td>\"How does [your brand] compare with [competitor] for mobile workforces?\"<\/td>\n<td>Which criteria frame the evaluation<\/td>\n<\/tr>\n<tr>\n<td>Diagnostic<\/td>\n<td>\"What did [publication] report about [your launch]?\"<\/td>\n<td>Whether specific coverage is retrievable at all<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><strong>The frozen-portfolio rule: change the wording and you have started a new series, not continued the old one.<\/strong><\/p>\n<p>Add new questions to a separate exploratory cohort. Retire prompts only at a quarter boundary, and keep the retired series visible.<\/p>\n<h4 id=\"which-ai-platforms-should-you-track\">Which AI platforms should you track?<\/h4>\n<p>Track ChatGPT, Perplexity, Gemini and Google AI Overviews, and treat each interaction mode as its own surface.<\/p>\n<table>\n<thead>\n<tr>\n<th>Platform<\/th>\n<th>Collection rule<\/th>\n<th>Main scoring safeguard<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>ChatGPT<\/td>\n<td>Fix whether you test default behaviour or explicitly selected search<\/td>\n<td>Do not blend search-enabled answers with a different interaction mode<\/td>\n<\/tr>\n<tr>\n<td>Perplexity<\/td>\n<td>Keep search mode, model and source focus constant<\/td>\n<td>Perplexity's own help documentation distinguishes Standard from Pro Search in depth and sources; treat them as separate surfaces<\/td>\n<\/tr>\n<tr>\n<td>Gemini<\/td>\n<td>Keep app, model and account conditions fixed<\/td>\n<td>Google's Gemini help center distinguishes sources from related content and notes some responses carry no links; classify conservatively<\/td>\n<\/tr>\n<tr>\n<td>Google AI Overviews<\/td>\n<td>Same queries, same location and device<\/td>\n<td>Record whether an Overview appeared <strong>before<\/strong> scoring its contents<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>If you can only afford two, take ChatGPT and Google AI Overviews. They sit at opposite ends of the source-age spectrum, so together they bracket the likely range of outcomes. Add Perplexity when you want the earliest read, since it cites on effectively every answer.<\/p>\n<p>For a platform-specific setup before you go cross-platform, a <a href=\"https:\/\/brandmentions.com\/blog\/chatgpt-brand-visibility-audit\/\">ChatGPT visibility audit<\/a> is the sensible first pass.<\/p>\n<h4 id=\"standardise-the-run-and-record-what-you-standardised\">Standardise the run, and record what you standardised<\/h4>\n<ol>\n<li>Fresh conversation or search session every time.<\/li>\n<li>Logged-out access where the required experience supports it; a dedicated research account with personalisation disabled where it does not.<\/li>\n<li>Fixed country, language, device type.<\/li>\n<li>Record the model label and search mode on every row.<\/li>\n<li>No follow-up coaching, no rephrasing to get a better answer.<\/li>\n<li>Save the original answer before opening any diagnostic conversation.<\/li>\n<li>Record failures as failures. <strong>Never re-run a valid negative because the second attempt looks better.<\/strong><\/li>\n<\/ol>\n<p><strong>A useful constraint:<\/strong> if you force search on, you measure visibility under search-enabled conditions. You have not measured the probability that an ordinary user receives that same experience.<\/p>\n<h4 id=\"decide-how-you-will-collect-with-eyes-open-about-the-workload\">Decide how you will collect, with eyes open about the workload<\/h4>\n<table>\n<thead>\n<tr>\n<th>Method<\/th>\n<th>Requirements<\/th>\n<th>Execution complexity<\/th>\n<th>When it becomes decision-useful<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Manual collection<\/td>\n<td>Analyst, spreadsheet, saved answers and screenshots<\/td>\n<td>Low setup, heavy recurring labour<\/td>\n<td>After matched baseline and post-launch windows<\/td>\n<\/tr>\n<tr>\n<td>AI visibility platform<\/td>\n<td>Budget, prompt configuration, export access, human review of a sample<\/td>\n<td>Moderate setup, less collection work<\/td>\n<td>After you validate its capture method and complete matched windows<\/td>\n<\/tr>\n<tr>\n<td>API-based collection<\/td>\n<td>Technical support, logging, fixed configuration<\/td>\n<td>Higher setup, scales cleanly<\/td>\n<td>After you validate which surface it actually measures and build its own baseline<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>No collection method shortens time to business value. These choices change how you gather evidence, not how fast the campaign works.<\/p>\n<p>Before committing, get four answers from any vendor or internal build: what collection method is permitted and used, what the observed failure and retry rate is, how many analyst hours per week the review of flagged rows requires, and whether outputs come from a consumer interface or an API. If a provider collects API responses, those must be labelled as API responses in your report. They are not automatically equivalent to what a user sees.<\/p>\n<h4 id=\"the-entity-resolution-check-nobody-runs\">The entity-resolution check nobody runs<\/h4>\n<p>Before you trust an automated share-of-voice number, test the detector. Take 100 answers you have already scored by hand and compare them with the automated labels.<\/p>\n<p>Count two error types separately: brand mentions the tool missed, and matches that were not actually your brand (a parent company, an unrelated product with a similar name, a common-word brand). If either rate is material, your share of voice is reporting detector behaviour rather than market behaviour. Fix the alias list and the exclusions, then re-test before anything reaches a slide.<\/p>\n<h3 id=\"step-3-monitor-earned-media-coverage-and-brand-mentions\">Step 3. Monitor Earned Media Coverage and Brand Mentions<\/h3>\n<p><strong>Do this: log every placement in one coverage register that records URL, accessibility, crawler access and the campaign facts each piece carries, with syndicated copies grouped under their original story.<\/strong><\/p>\n<p>Muck Rack's citation research is the reason this register needs to be broader than a press clipping file: its source categories include corporate content, blogs and other non-paid material alongside journalism. Your coverage log should mirror the full range of sources an engine might actually reach for.<\/p>\n<p>Build one campaign coverage register with these fields:<\/p>\n<ul>\n<li>Publication date and first-seen date.<\/li>\n<li>Exact article URL and preferred canonical URL.<\/li>\n<li>Publication and domain.<\/li>\n<li>Type: original reporting, syndicated copy, press release, paid placement or owned material.<\/li>\n<li>Brand and product names included.<\/li>\n<li>Which campaign facts, quotes or research the piece carries.<\/li>\n<li>Sentiment toward your brand.<\/li>\n<li>Public accessibility (open, soft paywall, hard paywall, login required).<\/li>\n<li>Campaign relationship (pitched, organic pickup, unrelated).<\/li>\n<li>Original-story identifier.<\/li>\n<\/ul>\n<h4 id=\"count-distribution-and-independent-coverage-separately\">Count distribution and independent coverage separately<\/h4>\n<p>Preserve every URL, but group syndicated copies under the original story. That gives you two numbers instead of one inflated one:<\/p>\n<ul>\n<li><strong>Distribution footprint:<\/strong> how many accessible pages carry the story.<\/li>\n<li><strong>Independent editorial coverage:<\/strong> how many distinct reporting decisions produced it.<\/li>\n<\/ul>\n<p>A release republished across forty sites is one editorial decision with forty surfaces. Both facts are useful. Reporting the forty as independent endorsements is how campaign reports lose credibility in a single question.<\/p>\n<p>A grounding in <a href=\"https:\/\/brandmentions.com\/blog\/earned-vs-paid-vs-organic-media\/\">earned media explained<\/a> as distinct from paid and owned keeps these columns honest when a wire pickup sits next to a bylined feature in the same spreadsheet.<\/p>\n<h4 id=\"track-competitors-in-the-same-window\">Track competitors in the same window<\/h4>\n<p>A rival's launch, funding round or controversy can change your relative visibility while your campaign performs identically. Without their coverage curve beside yours, you will read their activity as your failure.<\/p>\n<p><strong>Decision thresholds.<\/strong> If coverage volume rose but none of those domains ever appear in your citation logs, the break is access or relevance, not PR output. If the domains appear but your brand does not, the articles mention you too weakly to be cited for anything. If coverage volume did not rise, there is nothing to attribute yet and the measurement question is premature.<\/p>\n<p><strong>Failure mode.<\/strong> Deduplicating syndication away entirely. You lose the accessibility picture, and accessible syndicated copies are sometimes the only crawlable version of a paywalled original.<\/p>\n<p><strong>The non-obvious move:<\/strong> log the publication date of every placement, then record the date of the first observed citation for each one, <strong>including the placements that are never cited at all<\/strong>. Over three or four campaigns that gives you a category-specific lag distribution with the failures still in the denominator. Analysing only the successful pickups produces a timeline that flatters you and misleads the next brief.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/brandmentions.com\/blog\/wp-content\/uploads\/2026\/10\/image_04_47056c47.webp\" alt=\"Coverage-to-citation funnel from published PR placement to displayed AI citation.\" \/><\/p>\n<h3 id=\"step-4-identify-ai-answers-that-cite-your-pr-coverage\">Step 4. Identify AI Answers That Cite Your PR Coverage<\/h3>\n<p><strong>Do this: open every cited link, match it to your coverage register at URL level, and read the passage it supports before you count it as a brand citation.<\/strong><\/p>\n<p><a href=\"https:\/\/developers.openai.com\/api\/docs\/bots\" target=\"_blank\" rel=\"noopener noreferrer\">OpenAI's crawler documentation<\/a> is the first thing to read here, because it separates two controls that get conflated constantly: OAI-SearchBot governs access for search features, GPTBot governs potential model-training use, and they are independent settings. A publisher blocking training is not necessarily blocking search, and vice versa. Google-Extended, similarly, governs specified Gemini training and grounding uses and is not a switch for inclusion or ranking in Google Search.<\/p>\n<p>Check this before you pitch, not after you fail to find a citation. It takes minutes and it prevents promising an outcome the platform has already ruled out.<\/p>\n<h4 id=\"verify-every-candidate-citation\">Verify every candidate citation<\/h4>\n<p><strong>A visible link is a candidate for verification, not proof that the source supports a claim about your brand.<\/strong><\/p>\n<p>For each captured answer:<\/p>\n<ol>\n<li>Open the cited page.<\/li>\n<li>Resolve redirects and identify the final article.<\/li>\n<li>Match it against the coverage register at <strong>URL level<\/strong>, not domain level.<\/li>\n<li>Read the passage the citation is attached to.<\/li>\n<li>Check whether the source actually supports that passage.<\/li>\n<li>Record whether the passage concerns your brand at all.<\/li>\n<li>Save the answer, the citation and the supporting excerpt.<\/li>\n<\/ol>\n<p>Domain matching is the shortcut most teams reach for, and it cannot tell you which article was used or whether the piece concerns you. A trade publication cited for a competitor's story is not your citation.<\/p>\n<table>\n<thead>\n<tr>\n<th>Evidence<\/th>\n<th>Record it as<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Exact campaign article supports a statement about your brand<\/td>\n<td>Verified campaign-linked brand citation<\/td>\n<\/tr>\n<tr>\n<td>Campaign article cited, but supports a general industry statement<\/td>\n<td>Campaign source exposure<\/td>\n<\/tr>\n<tr>\n<td>Brand named with no supporting link<\/td>\n<td>Uncited brand mention<\/td>\n<\/tr>\n<tr>\n<td>A campaign fact repeated with no source identified<\/td>\n<td>Possible message pickup, unattributed<\/td>\n<\/tr>\n<tr>\n<td>Related article in the source list with no claim connection<\/td>\n<td>Related source, not a brand-supporting citation<\/td>\n<\/tr>\n<tr>\n<td>Link broken, inaccessible, or does not support the claim<\/td>\n<td>Unverified citation<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><strong>The non-obvious distinction:<\/strong> a campaign can succeed at source exposure and fail at message delivery. Your article gets cited, and the answer uses it to support a market-size statistic while never mentioning you. Count both outcomes so one cannot hide inside the other.<\/p>\n<h4 id=\"measure-drop-off-in-the-coverage-to-citation-funnel\">Measure drop-off in the coverage-to-citation funnel<\/h4>\n<p>Score your placements against the same five stages defined in <a href=\"#can-pr-campaigns-influence-ai-search-citations\">Can PR campaigns influence AI search citations?<\/a> and count how many survive each one. Where the funnel breaks tells you which budget line to argue for.<\/p>\n<table>\n<thead>\n<tr>\n<th>Funnel stage<\/th>\n<th>What to count<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>1. Published<\/td>\n<td>Placements live and publicly accessible<\/td>\n<\/tr>\n<tr>\n<td>2. Discoverable<\/td>\n<td>Placements reachable by the relevant crawler - no block, hard paywall or login wall<\/td>\n<\/tr>\n<tr>\n<td>3. Selected<\/td>\n<td>Placements retrieved and selected as a source for a prompt in your portfolio<\/td>\n<\/tr>\n<tr>\n<td>4. Included<\/td>\n<td>Cited passages that actually carry your brand, product or campaign claim<\/td>\n<\/tr>\n<tr>\n<td>5. Displayed<\/td>\n<td>Citations shown in the answer, and where possible tied to an identifiable visit<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Most campaigns leak at stages 2 and 3. Those are a media-list problem and a story-angle problem respectively, and neither is solved by producing more coverage of the same kind.<\/p>\n<h4 id=\"score-it-so-a-second-person-gets-the-same-answer\">Score it so a second person gets the same answer<\/h4>\n<p>Use labels another reviewer can reproduce: presence (absent\/present), position (first appearance among named alternatives, or \"not ranked\" for narrative answers), recommendation (recommended \/ mentioned only \/ advised against), sentiment (positive \/ neutral \/ mixed \/ negative), accuracy (correct \/ partly incorrect \/ incorrect \/ unverifiable), citation support (supported \/ unsupported \/ unclear).<\/p>\n<p>Check material claims against an approved fact sheet holding current features, pricing, availability and research methodology. Then have a second reviewer blind-score a sample of 20 to 30 answers and compare. Resolve the disagreements before you scale the process, because an unreliable scoring rubric produces unreliable trends no sample size can rescue.<\/p>\n<blockquote><p>A neutral, accurate recommendation is commercially useful. A glowing answer with the wrong pricing is a problem the mention-rate chart will never show you.<\/p><\/blockquote>\n<h3 id=\"step-5-compare-brand-mentions-and-citation-rates-after-launch\">Step 5. Compare Brand Mentions and Citation Rates After Launch<\/h3>\n<p><strong>Do this: re-run the identical portfolio on the identical platforms and schedule, then report rates by platform with honest denominators and collection failures shown separately.<\/strong><\/p>\n<p><a href=\"https:\/\/www.meltwater.com\/en\/blog\/ai-search-visibility-report-september-2026\" target=\"_blank\" rel=\"nofollow noopener noreferrer\">Meltwater's GenAI Lens report for September 2026<\/a>, published October 15, analysed more than 4,000 prompts across eight AI platforms and recorded Earned and News citation share rising from 29.4% to 35.6%, a 6.2-percentage-point change. Use that as context for what the source environment is doing, not as evidence that your PR improved. Their prompt mix, categories and platform configuration are not yours.<\/p>\n<p>Compare like with like: same portfolio, same platform and interaction mode, same country and language, same number of scheduled runs, same scoring definitions, same window length.<\/p>\n<p>Report by platform first, then by topic and intent. If you publish one blended figure, fix the weights in advance, or a platform that simply produces more citations will dominate the aggregate without mattering more to your customers.<\/p>\n<h4 id=\"keep-the-denominator-honest\">Keep the denominator honest<\/h4>\n<ul>\n<li>Include completed answers that omit your brand.<\/li>\n<li>Include Google searches that complete with no AI Overview as zero exposure.<\/li>\n<li>Exclude genuine collection failures from valid runs, and report the failure rate separately.<\/li>\n<li>Count each brand once per answer, and each distinct cited URL once per answer.<\/li>\n<li>Deduplicate tracking parameters and equivalent URL versions.<\/li>\n<\/ul>\n<p>For Google AI Overviews, always show two rates:<\/p>\n<ol>\n<li>Citation rate across all valid searches (exposure opportunity).<\/li>\n<li>Citation rate among searches where an Overview actually appeared (selection within that opportunity).<\/li>\n<\/ol>\n<p><strong>The exposure\/selection split: more AI answers can raise your citation count without the engine choosing you any more often.<\/strong><\/p>\n<p>That decomposition has saved more than one team from claiming credit for a Google interface change.<\/p>\n<h3 id=\"step-6-compare-results-against-competitors-or-a-control-group\">Step 6. Compare Results Against Competitors or a Control Group<\/h3>\n<p><strong>Do this: measure an untouched control topic or market over the same window and report control-adjusted change instead of raw lift.<\/strong><\/p>\n<p><a href=\"https:\/\/blogs.worldbank.org\/en\/impactevaluations\/revisiting-difference-differences-parallel-trends-assumption-part-i-pre-trend\" target=\"_blank\" rel=\"nofollow noopener noreferrer\">The World Bank's guidance on the parallel trends assumption<\/a> sets the bar for this step: a before-and-after comparison only supports a causal reading if your comparison group credibly represents what would have happened without the campaign. Most PR \"controls\" do not clear that bar, and saying so is what makes the rest of your report trustworthy.<\/p>\n<p>Candidate controls:<\/p>\n<ul>\n<li>A related topic the campaign did not target.<\/li>\n<li>A comparable product line with no launch activity.<\/li>\n<li>A market the campaign coverage did not reach.<\/li>\n<li>A competitor with no overlapping campaign activity.<\/li>\n<\/ul>\n<p><strong>Competitor benchmark vs true control: the core difference.<\/strong> A benchmark tells you whether you gained relative position. A control tells you what would have happened anyway. Teams report the first and claim the second. And the trap inside the trap: a rival named inside your own coverage is not an untouched comparison, because a comparison article covering your launch alongside a competitor can lift their visibility too. Keep those rivals for competitive reporting. Drop them as a causal control.<\/p>\n<h4 id=\"three-checks-before-any-inference\">Three checks before any inference<\/h4>\n<ol>\n<li><strong>Persistence.<\/strong> Does the improvement recur across subsequent collection windows, or did it appear once?<\/li>\n<li><strong>Breadth.<\/strong> Does it show across several relevant prompts, or one favourable question?<\/li>\n<li><strong>Comparison.<\/strong> Did the target improve more than the control over the same window?<\/li>\n<\/ol>\n<p>These strengthen a decision. They do not replace uncertainty analysis. For a consequential budget call, have someone calculate the interval around the <strong>change<\/strong>, accounting for repeated prompts and collection days, and keep the inference scoped to your portfolio.<\/p>\n<p><strong>The non-obvious use of competitor data:<\/strong> your competitor set is simultaneously your noise estimate and your share-of-voice denominator. If a rival launches during your window, their rise is a confound and a free read on which of their tactics the engines rewarded. Record it as both.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/brandmentions.com\/blog\/wp-content\/uploads\/2026\/10\/image_05_fae7b2ef.webp\" alt=\"Scorecard for calculating PR impact on AI visibility metrics.\" \/><\/p>\n<h2 id=\"how-to-calculate-pr-campaign-impact-on-ai-visibility-the-9-formulas\">How to\u00a0Measure AI Citation Rate, Brand Mentions, and PR Campaign Impact<\/h2>\n<p>The formulas are short. The discipline lives in the denominators and the labels.<\/p>\n<p><strong>The nine formulas for measuring PR impact on AI visibility are:<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Metric<\/th>\n<th>Formula<\/th>\n<th>Denominator \/ caveat<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Brand mention rate<\/td>\n<td>Valid runs mentioning your brand \u00f7 all valid runs \u00d7 100<\/td>\n<td>Denominator is all valid runs, including answers that never name you<\/td>\n<\/tr>\n<tr>\n<td>Brand-supporting citation rate<\/td>\n<td>Valid runs with at least one verified brand-supporting citation \u00f7 all valid runs \u00d7 100<\/td>\n<td>Counts qualifying sources whether they are yours or a third party's<\/td>\n<\/tr>\n<tr>\n<td>Campaign coverage citation rate<\/td>\n<td>Valid runs citing at least one verified campaign URL \u00f7 all valid runs \u00d7 100<\/td>\n<td>Measures use of campaign material; it does not mean the answer recommended you<\/td>\n<\/tr>\n<tr>\n<td>AI share of voice<\/td>\n<td>Answers mentioning your brand \u00f7 sum of answer appearances for every brand in the fixed set \u00d7 100<\/td>\n<td>Unbranded and comparison prompts only; adding a competitor changes the denominator and breaks the trend<\/td>\n<\/tr>\n<tr>\n<td>Citation share<\/td>\n<td>Your qualifying citation instances \u00f7 all qualifying citation instances \u00d7 100<\/td>\n<td>Measures composition, not frequency; not interchangeable with citation rate<\/td>\n<\/tr>\n<tr>\n<td>Absolute lift<\/td>\n<td>Post-campaign rate \u2212 baseline rate<\/td>\n<td>Expressed in percentage points; report alongside relative lift<\/td>\n<\/tr>\n<tr>\n<td>Relative lift<\/td>\n<td>(Post \u2212 baseline) \u00f7 baseline \u00d7 100<\/td>\n<td>Undefined when the baseline is zero; report percentage points and raw counts instead<\/td>\n<\/tr>\n<tr>\n<td>Control-adjusted change<\/td>\n<td>(Target post \u2212 target pre) \u2212 (control post \u2212 control pre)<\/td>\n<td>Call it control-adjusted change, not \"PR-caused lift\"<\/td>\n<\/tr>\n<tr>\n<td>AI Overview citation rate (decomposed)<\/td>\n<td>Overview appearance rate \u00d7 citation rate within displayed Overviews<\/td>\n<td>Using proportions, 0.50 \u00d7 0.20 = 0.10, or 10%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><strong>Brand mention rate<\/strong> = valid runs mentioning your brand \u00f7 all valid runs \u00d7 100.<\/p>\n<p><strong>Brand-supporting citation rate<\/strong> = valid runs with at least one verified brand-supporting citation \u00f7 all valid runs \u00d7 100. Count qualifying sources whether they are yours or a third party's.<\/p>\n<p><strong>Campaign coverage citation rate<\/strong> = valid runs citing at least one verified campaign URL \u00f7 all valid runs \u00d7 100. This measures use of campaign material. It does not mean the answer recommended you.<\/p>\n<p><strong>AI share of voice<\/strong> = answers mentioning your brand \u00f7 the sum of answer appearances for every brand in the fixed set \u00d7 100. Compute it on unbranded and comparison prompts only; branded prompts inflate it mechanically. Keep the competitor set unchanged, because adding a competitor changes the denominator and breaks the trend.<\/p>\n<p><strong>Citation share<\/strong> = your qualifying citation instances \u00f7 all qualifying citation instances \u00d7 100. This measures composition. Citation rate measures the proportion of answers containing a qualifying citation. They are different questions and cannot be swapped.<\/p>\n<p><strong>Absolute lift<\/strong> = post-campaign rate minus baseline rate, in percentage points.<br \/>\n<strong>Relative lift<\/strong> = (post minus baseline) \u00f7 baseline \u00d7 100.<\/p>\n<p>Report both, labelled. A move from 20% to 30% is +10 percentage points and +50% relative. If the baseline is zero, report percentage points and raw counts, because relative lift has no defined value.<\/p>\n<p><strong>Control-adjusted change<\/strong> = (target post minus target pre) minus (control post minus control pre).<\/p>\n<p>Call it control-adjusted change. Not \"PR-caused lift.\" The causal reading still depends on whether the control was credible and whether anything else changed at the same time.<\/p>\n<p><strong>For AI Overviews:<\/strong> citation rate across all valid searches = Overview appearance rate \u00d7 citation rate within displayed Overviews. Using proportions, 0.50 \u00d7 0.20 = 0.10, or 10%.<\/p>\n<h3 id=\"key-terms-the-ai-citation-measurement-vocabulary\">Key terms: the AI citation measurement vocabulary<\/h3>\n<ul>\n<li><strong>Citation rate is<\/strong> the percentage of valid runs in which a qualifying citation appears at all.<\/li>\n<li><strong>Citation share is<\/strong> the percentage of all qualifying citation instances in a sample that belong to you, which describes composition rather than frequency.<\/li>\n<li><strong>AI share of voice is<\/strong> your brand's answer appearances divided by the answer appearances of every brand in a fixed competitor set, expressed as a percentage.<\/li>\n<li><strong>Control-adjusted change is<\/strong> your target topic's change minus the untouched control topic's change over the same window, in percentage points.<\/li>\n<li><strong>A brand-supporting citation is<\/strong> a verified source link that supports a statement about your brand, as opposed to an ordinary source appearance.<\/li>\n<li><strong>A valid run is<\/strong> a completed observation under your documented conditions, including completed answers that never mention you; genuine collection failures are excluded and reported separately.<\/li>\n<li><strong>First observed is<\/strong> the date your tracker first saw a citation, which is not the date the platform discovered the page.<\/li>\n<\/ul>\n<h3 id=\"measurable-improvement-vs-causal-attribution\">Measurable improvement vs causal attribution<\/h3>\n<p>Here is the line, stated plainly, because almost every competing guide blurs it.<\/p>\n<p><strong>You can measure<\/strong> that your rate rose, that the rise exceeded your documented variability range, that it persisted across windows, that campaign URLs appear in the cited sources, and that your control did not move.<\/p>\n<p><strong>You cannot prove<\/strong> that the campaign caused it from prompt-based measurement alone. Retrieval and selection are proprietary. A model update, a competitor's coverage going stale, a change in source weighting or a shift in the information environment around your category can all produce the same pattern.<\/p>\n<p>What you can do is make the strongest claim the evidence supports, and say so in those words.<\/p>\n<h4 id=\"copy-ready-reporting-template-what-to-tell-leadership\">Copy-ready reporting template: what to tell leadership<\/h4>\n<blockquote><p>Following the [campaign name] campaign, brand-supporting citation rate on [platform] rose from a [length] baseline of [baseline %] (weekly range [range low] to [range high]) to a [length] post-campaign average of [post %]. Verified campaign URLs appeared in [n] of [total] post-launch runs. The untouched comparison topic rose [control change] points over the same window. Collection conditions were unchanged. The timing, magnitude and comparison are consistent with a campaign contribution, and the result does not establish causation on its own.<\/p><\/blockquote>\n<p><strong>Worked example of the same template, filled in (illustrative figures):<\/strong> \"Following the launch campaign, brand-supporting citation rate on ChatGPT rose from a four-week baseline of 20% (weekly range 16% to 24%) to a four-week post-campaign average of 40%. Verified campaign URLs appeared in 24 of 120 post-launch runs. The untouched comparison topic rose 5 points over the same window. Collection conditions were unchanged. The timing, magnitude and comparison are consistent with a campaign contribution, and the result does not establish causation on its own.\"<\/p>\n<p>That paragraph survives scrutiny. The screenshot does not.<\/p>\n<h2 id=\"what-can-you-claim-from-your-ai-citation-results-the-evidence-ladder\">What Can You Claim From Your AI Citation Results? The Evidence Ladder<\/h2>\n<p>This is the asset to pin above your desk. It maps what you observed to what you are entitled to say, and it is deliberately conservative.<\/p>\n<table>\n<thead>\n<tr>\n<th>Evidence you hold<\/th>\n<th>Defensible statement<\/th>\n<th>Next action<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>One answer or screenshot<\/td>\n<td>\"We observed a citation\"<\/td>\n<td>Repeat under fixed conditions before reporting anything<\/td>\n<\/tr>\n<tr>\n<td>Repeated pre\/post improvement in a frozen portfolio<\/td>\n<td>\"Visibility increased in our tracked sample\"<\/td>\n<td>Check persistence, controls and whether collection changed<\/td>\n<\/tr>\n<tr>\n<td>Verified campaign URLs cited in relevant answers<\/td>\n<td>\"AI answers cited our campaign coverage\"<\/td>\n<td>Check message accuracy and breadth across prompts<\/td>\n<\/tr>\n<tr>\n<td>Persistent improvement beyond a credible control<\/td>\n<td>\"The results support a campaign contribution\"<\/td>\n<td>Review remaining alternative explanations before expanding spend<\/td>\n<\/tr>\n<tr>\n<td>A suitable randomised or independently validated design<\/td>\n<td>A bounded causal claim, scoped to that design<\/td>\n<td>Report the design, the uncertainty and the scope<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>A source match is strong evidence of <strong>material use<\/strong>. It does not tell you what the answer would have looked like without that material. And imperfect attribution does not make the evidence worthless; it determines the size of the decision you are entitled to make from it.<\/p>\n<p>Most teams sit on rung two and present rung four. Moving honestly from rung two to rung three costs almost nothing: a handful of control prompts and a URL-matching column.<\/p>\n<h2 id=\"what-should-an-ai-citation-tracking-dashboard-include\">What Should an AI Citation Tracking Dashboard Include?<\/h2>\n<p>One page. Five panels. Each one answers a question leadership will actually ask.<\/p>\n<ol>\n<li><strong>Campaign activity:<\/strong> publication dates, independent stories, distribution footprint.<\/li>\n<li><strong>AI outcomes:<\/strong> mention rate, brand-supporting citation rate and campaign coverage citation rate, by platform, with the baseline range shaded.<\/li>\n<li><strong>Competitive context:<\/strong> fixed-set AI share of voice, plus competitor coverage volume on the same timeline.<\/li>\n<li><strong>Quality:<\/strong> supported claims, inaccurate claims, sentiment distribution.<\/li>\n<li><strong>Business evidence:<\/strong> identifiable AI referrals, qualified actions, relevant customer feedback.<\/li>\n<\/ol>\n<p><strong>The Monday morning test: if a panel would not change a pitch, a media list or a budget line, remove it.<\/strong><\/p>\n<p>Keep the coverage and conversation baseline running between campaigns, not just during them, so the next launch starts with a current record instead of a reconstruction. BrandMentions is the part of the stack that holds that continuous web, news, blog, forum and social mention history with competitor share of voice and sentiment attached, which is what lets you join a citation back to the coverage that produced it. Applying consistent <a href=\"https:\/\/brandmentions.com\/blog\/media-monitoring\/\">media monitoring basics<\/a> across both datasets, stable queries, documented exclusions, consistent duplicate handling, is what keeps them joinable.<\/p>\n<h3 id=\"ai-visibility-and-geo-tools-for-tracking-citations-after-pr\">AI Visibility and GEO Tools for Tracking Citations After PR<\/h3>\n<p>Three categories of tools cover the job, alongside a manual spreadsheet for teams starting out. Most teams need two at most.<\/p>\n<table>\n<thead>\n<tr>\n<th>Tool or category<\/th>\n<th>What it covers<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>AI prompt trackers (Profound, Peec AI, Otterly.ai)<\/td>\n<td>Scheduled prompt runs and reporting on the answer side of the equation; validate the capture method and whether outputs come from a consumer interface or an API before you trust a trend.<\/td>\n<\/tr>\n<tr>\n<td>Google Search Console generative AI performance insights<\/td>\n<td>Impressions from AI Overviews and AI Mode by page, country, date and device, for your own verified property only - first-party data on your pages, not on publisher coverage or answer text.<\/td>\n<\/tr>\n<tr>\n<td><a href=\"https:\/\/brandmentions.com\/\">BrandMentions<\/a><\/td>\n<td>The mention baseline: continuous web-wide monitoring across news, blogs, forums, websites and social, with competitor share of voice, sentiment and historical lookback, which is the coverage record a prompt tracker does not hold.<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Prompt trackers report the answer side of LLM visibility well; the mention record explains what fed it. Neither replaces URL-level verification by a human.<\/p>\n<h3 id=\"use-googles-first-party-reporting-and-know-its-boundary\">Use Google's first-party reporting, and know its boundary<\/h3>\n<p>Google's generative AI performance insights in Search Console, introduced June 3, 2026 and rolled out worldwide by August 31, 2026, give you impressions, appearing pages and countries for your own verified property.<\/p>\n<p>Use them to complement your prompt sample. Then remember the boundary: your property data cannot show mentions of your brand inside a publisher's article or inside an AI answer. Third-party coverage tracking stays in place.<\/p>\n<h3 id=\"keep-citations-separate-from-visits\">Keep citations separate from visits<\/h3>\n<p><a href=\"https:\/\/www.pewresearch.org\/short-reads\/2025\/07\/22\/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results\/\" target=\"_blank\" rel=\"noopener noreferrer\">Pew Research Center's study of 900 US adults<\/a> found users clicked a link within an AI summary on 1% of visits to search pages containing a summary, in its March 2025 browsing data. That is an observed result from that study and that period, not a universal click-through rate.<\/p>\n<p>The practical consequences for your report:<\/p>\n<ul>\n<li>Track identifiable AI referrals and their landing pages.<\/li>\n<li>Preserve campaign parameters on links you control.<\/li>\n<li>Treat self-reported discovery (\"I found you through ChatGPT\") as supporting evidence.<\/li>\n<li>Do not assign direct-traffic growth to AI by default.<\/li>\n<li>Accept that a citation to a publisher sends the visit to the publisher.<\/li>\n<\/ul>\n<p>A citation can contribute to awareness without producing a visit you can see. Do not invent the missing attribution to fill the gap in the slide.<\/p>\n<h3 id=\"join-the-numbers-to-money-or-the-reinvestment-question-stays-open\">Join the numbers to money, or the reinvestment question stays open<\/h3>\n<p>Citation lift alone does not establish financial return. Before the reinvestment conversation, put four columns side by side: campaign spend (agency, wire, content, events), measurement cost (tools and analyst hours), verified exposure (independent stories, accessible placements, verified campaign citations), and qualified outcomes (AI-referred qualified actions, self-reported discovery, pipeline tagged to the launch).<\/p>\n<p>Then be explicit about what you cannot separate. A product launch usually ships earned PR alongside new product pages, a research page, paid distribution and sales enablement. Verified use of your campaign article does not tell you which component deserves the next budget increment. If that question matters, stagger the components across markets or time windows deliberately, so the next campaign can answer it.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/brandmentions.com\/blog\/wp-content\/uploads\/2026\/10\/image_06_c0d49eca.webp\" alt=\"Illustrative citation-rate lift by AI platform after a product launch PR campaign.\" \/><\/p>\n<h2 id=\"example-tracking-ai-citations-after-a-product-launch-pr-campaign\">Example: Tracking AI Citations After a Product Launch PR Campaign<\/h2>\n<p><strong>This example is illustrative. The company, prompts and numbers are constructed to demonstrate the method, not to report observed results.<\/strong><\/p>\n<p>A fictional company, CrewHarbor, launches scheduling software for field-service teams. The campaign targets one buyer topic and generates six independently produced articles: two trade features, one business daily article, and three additional editorial stories. A separate wire release is syndicated across eighteen websites. The campaign therefore produces six independent editorial stories, one press release, and eighteen syndicated placements.<\/p>\n<p>The team freezes five prompts:<\/p>\n<ol>\n<li>\"Which employee scheduling tools suit US field-service companies with 50 to 200 staff?\"<\/li>\n<li>\"What scheduling software works well for teams moving between customer sites?\"<\/li>\n<li>\"Which workforce scheduling tools support shift swaps and mobile access?\"<\/li>\n<li>\"How should a growing field-service company compare scheduling platforms?\"<\/li>\n<li>\"What should we check before replacing our employee scheduling software?\"<\/li>\n<\/ol>\n<p>Each prompt runs six times weekly on each platform, spread across days. That is 30 runs per platform per week, and 120 per platform in each four-week window.<\/p>\n<h3 id=\"baseline-and-post-campaign-results\">Baseline and post-campaign results<\/h3>\n<table>\n<thead>\n<tr>\n<th>Platform<\/th>\n<th align=\"right\">Brand mentions, before \u2192 after<\/th>\n<th align=\"right\">Brand-supporting citations, before \u2192 after<\/th>\n<th align=\"right\">Citation rate change<\/th>\n<th align=\"right\">Post-launch runs citing campaign coverage<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>ChatGPT<\/td>\n<td align=\"right\">42 \u2192 72<\/td>\n<td align=\"right\">24 \u2192 48<\/td>\n<td align=\"right\">20% \u2192 40%, <strong>+20 pp<\/strong><\/td>\n<td align=\"right\">24<\/td>\n<\/tr>\n<tr>\n<td>Perplexity<\/td>\n<td align=\"right\">48 \u2192 78<\/td>\n<td align=\"right\">30 \u2192 54<\/td>\n<td align=\"right\">25% \u2192 45%, <strong>+20 pp<\/strong><\/td>\n<td align=\"right\">30<\/td>\n<\/tr>\n<tr>\n<td>Gemini<\/td>\n<td align=\"right\">30 \u2192 36<\/td>\n<td align=\"right\">12 \u2192 18<\/td>\n<td align=\"right\">10% \u2192 15%, <strong>+5 pp<\/strong><\/td>\n<td align=\"right\">6<\/td>\n<\/tr>\n<tr>\n<td>Google AI Overviews<\/td>\n<td align=\"right\">24 \u2192 36<\/td>\n<td align=\"right\">12 \u2192 18<\/td>\n<td align=\"right\">10% \u2192 15%, <strong>+5 pp<\/strong><\/td>\n<td align=\"right\">6<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Every cell uses 120 valid runs. The campaign-coverage column overlaps with the citation column, so the two cannot be added together.<\/p>\n<h3 id=\"how-to-read-it\">How to read it<\/h3>\n<p>ChatGPT and Perplexity show the largest observed improvement, and verified campaign URLs establish that the campaign's material entered sampled answers. URL-level matching shows the two trade features and the business daily account for most of those citations, while the eighteen wire surfaces contribute a small share, concentrated on one high-authority financial domain.<\/p>\n<p>Assume the untouched comparison topic on ChatGPT moved from 10% to 15%. The control-adjusted change is (40% \u2212 20%) \u2212 (15% \u2212 10%) = <strong>15 percentage points<\/strong>, subject to the control's assumptions holding.<\/p>\n<p>Google needs a different reading. Suppose AI Overviews appeared in 60 baseline searches and 90 post-campaign searches. The conditional citation rate is 12 \u00f7 60 = 20% before, and 18 \u00f7 90 = 20% after. Exposure increased. Selection inside displayed Overviews did not.<\/p>\n<p>The statement to leadership: \"Campaign coverage appeared in sampled AI answers, and citation rates rose on all four platforms. ChatGPT and Perplexity showed the largest changes, and the control-adjusted change on ChatGPT was 15 percentage points. Google's increase reflects more AI Overviews appearing rather than a higher citation rate within them. Next campaign: more original data, same three publishers, and keep measuring Google for another quarter.\"<\/p>\n<p>That is a reinvestment decision. \"AI visibility increased\" is not.<\/p>\n<h2 id=\"why-isnt-my-pr-coverage-showing-up-in-ai-citations-8-causes\">Why Isn't My PR Coverage Showing Up in AI Citations? 8 Causes<\/h2>\n<p>Work the list in order. The early causes are cheaper to check and more often responsible.<\/p>\n<table>\n<thead>\n<tr>\n<th>Symptom<\/th>\n<th>What to investigate<\/th>\n<th>Practical response<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Coverage exists, no relevant sources appear anywhere<\/td>\n<td>Topic fit between the article and the tested questions<\/td>\n<td>Compare the article's actual contents with the prompts, word by word<\/td>\n<\/tr>\n<tr>\n<td>The page is hard to reach<\/td>\n<td>Login walls, hard paywalls, crawler access<\/td>\n<td>Ask the publisher; check whether an accessible syndicated version exists<\/td>\n<\/tr>\n<tr>\n<td>Google does not show the supporting page<\/td>\n<td>Indexing, snippet eligibility, generative-feature inclusion<\/td>\n<td>Work through Google's three documented conditions<\/td>\n<\/tr>\n<tr>\n<td>ChatGPT never surfaces your owned material<\/td>\n<td>Search-crawler access, specifically OAI-SearchBot<\/td>\n<td>Review with your web team; training controls are a separate setting<\/td>\n<\/tr>\n<tr>\n<td>The brand appears but the campaign claim does not<\/td>\n<td>Weak attribution inside the article<\/td>\n<td>Publish verifiable supporting material the answer can lean on<\/td>\n<\/tr>\n<tr>\n<td>Only branded prompts improve<\/td>\n<td>Prompt dependence<\/td>\n<td>Report unbranded discovery separately and treat it as the real score<\/td>\n<\/tr>\n<tr>\n<td>Everything shifts at once, in all directions<\/td>\n<td>Collection or platform change<\/td>\n<td>Check the event log before attributing anything to PR<\/td>\n<\/tr>\n<tr>\n<td>A campaign article is cited with no useful brand statement<\/td>\n<td>Source exposure without message delivery<\/td>\n<td>Inspect which passage the answer actually used<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Two causes deserve expanding, because they are the ones teams misdiagnose.<\/p>\n<p><strong>Freshness competition, read correctly.<\/strong> Your three-week-old article competes against a corpus where the average cited page runs over a thousand days old on some surfaces. That is a competitive statement about what engines select, not a waiting period you can sit out. The fix is relevance and evidence, not patience.<\/p>\n<p><strong>Insufficient sampling masquerading as absence.<\/strong> If your coverage is cited on 30% of runs and you ran the prompt twice, two misses mean nothing. A brand appearing on 3 of 8 runs has a real 37.5% presence rate. A brand appearing on 0 of 2 runs is unmeasured, not absent. This is the one failure mode the protocol itself cures.<\/p>\n<p>For a structured walk-through of each cause in diagnostic order, the guide to a <a href=\"https:\/\/brandmentions.com\/blog\/brand-not-showing-up-ai-search\/\">brand missing from AI<\/a> answers covers the sequence of access, relevance, selection and message inclusion in more depth.<\/p>\n<h2 id=\"how-to-get-cited-by-ai-in-your-next-pr-campaign-5-levers\">How to Get Cited by AI in Your Next PR Campaign: 5 Levers<\/h2>\n<p>Five levers, each analysed the same way so you can compare them honestly: what it is, why it works, and what the evidence does and does not support.<\/p>\n<h3 id=\"lead-with-original-data\">Lead with original data<\/h3>\n<p><strong>What it is:<\/strong> a survey, benchmark, index or dataset published with its method, sample, dates, definitions and limitations.<\/p>\n<p><strong>Why it works:<\/strong> engines cite sources to support specific claims. A press release announcing a launch gives an answer nothing to attribute. A finding like \"a survey of 500 operations leaders found 67% still schedule shifts manually\" gives it a claim with an owner.<\/p>\n<p><strong>What the evidence supports:<\/strong> a <a href=\"https:\/\/arxiv.org\/pdf\/2609.35077\" target=\"_blank\" rel=\"noopener\">2026 observational study from Discovered Labs<\/a> of roughly 2 million citations across ChatGPT, Claude, Google AI Overviews and Gemini found that domain-level authority outweighed the strongest page-level feature by about six times. Where your finding is published matters more than how the page is formatted. This is a preprint on B2B SaaS data, not a universal formula.<\/p>\n<p><strong>What it does not support:<\/strong> restructuring a narrative feature into bullet points as a citation tactic. CXL's analysis of 100 AI Overview citations found 55% of cited passages sat in the top 30% of the source article, and CXL itself notes placement is not a guaranteed ranking factor. Question fit comes first. Formatting is downstream of having something worth citing.<\/p>\n<h3 id=\"supply-expert-commentary-with-named-attribution\">Supply expert commentary with named attribution<\/h3>\n<p><strong>What it is:<\/strong> a quote from a named person with a stated role, answering a specific buyer uncertainty rather than praising the product.<\/p>\n<p><strong>Why it works:<\/strong> a quote attributed to a named person at a named company connects a claim to your brand entity in the source text. \"A spokesperson said\" connects it to nothing.<\/p>\n<p><strong>What the evidence supports:<\/strong> Meltwater's GenAI Lens research found roughly 75% of the LinkedIn citations in its sample came from individual member profiles rather than company pages. Read that as a signal about expert-led, personally attributed content in the source mix, not as a guaranteed channel ranking or a reason to flood LinkedIn.<\/p>\n<p><strong>Execution constraint:<\/strong> check the published piece. If attribution came out vague, ask for a correction. It matters more now than when only humans read the article.<\/p>\n<h3 id=\"target-publications-the-engines-already-cite-in-your-category\">Target publications the engines already cite in your category<\/h3>\n<p><strong>What it is:<\/strong> building the media list partly from your own citation logs rather than entirely from reach and circulation.<\/p>\n<p><strong>Why it works:<\/strong> Ahrefs' 75,000-brand study found off-site brand mentions tracking AI Overview visibility more closely than backlinks did. Where those mentions sit is a targeting variable you control.<\/p>\n<p><strong>How to execute it:<\/strong> from your Step 1 baseline, list every domain cited for your prompt portfolio. Cross-reference against crawler access. The overlap is your AI-priority media list. It will not match your reach list, and that is fine. Run both, labelled, and report against both. Pair it with a conventional <a href=\"https:\/\/brandmentions.com\/wiki\/Share_of_Voice\">share of voice<\/a> read and a <a href=\"https:\/\/help.brandmentions.com\/en\/articles\/12597399-how-can-brandmentions-help-me-analyze-my-competitors\">competitor analysis in BrandMentions<\/a>, so you can see which outlets carry your rivals' coverage as well as your own before the next pitch list is built.<\/p>\n<p><strong>What it does not support:<\/strong> abandoning reach-based targeting. Human audiences and retrieval pools are different objectives with different lists, and a placement can be excellent for one and irrelevant to the other.<\/p>\n<h3 id=\"keep-attribution-intact-end-to-end\">Keep attribution intact end to end<\/h3>\n<p><strong>What it is:<\/strong> consistent brand naming, consistent product naming, and a traceable chain from the finding to the source page across every placement.<\/p>\n<p><strong>Why it works:<\/strong> entity extraction fragments when you are \"CrewHarbor\" in one article, \"Crew Harbor\" in another, and an unnamed \"the company\" in a third. The same fragmentation that confuses an engine also breaks your own automated detection, which is why the entity-resolution check in Step 2 exists.<\/p>\n<p><strong>Execution constraint:<\/strong> this is a boilerplate and briefing discipline, not a content tactic. Lock the naming conventions into the press kit and the spokesperson prep.<\/p>\n<h3 id=\"publish-accessible-supporting-material-on-your-own-domain\">Publish accessible supporting material on your own domain<\/h3>\n<p><strong>What it is:<\/strong> a readable evidence page carrying the method, the data, and the full findings, in text, on a stable URL.<\/p>\n<p><strong>Why it works:<\/strong> journalists link to it, engines can retrieve it, and owned-domain citations are the only ones that can produce referral traffic you can verify. Google's guidance is clear that important content should be in textual form rather than buried in images or documents.<\/p>\n<p><strong>What the evidence supports:<\/strong> maintaining it. Updating the page with new numbers annually while keeping the URL stable keeps one citable surface alive across campaigns rather than creating a new orphan each quarter.<\/p>\n<p>This is the practical core of <a href=\"https:\/\/brandmentions.com\/blog\/what-is-answer-engine-optimization-aeo\/\">answer engine optimization<\/a> - and of generative engine optimization - for a comms team: give the engine a page that answers the buyer's question, on a domain you control, and let the earned coverage point at it.<\/p>\n<h2 id=\"frequently-asked-questions\">Frequently Asked Questions<\/h2>\n<h3 id=\"what-is-the-difference-between-an-ai-citation-and-an-ai-brand-mention\">What is the difference between an AI citation and an AI brand mention?<\/h3>\n<p>An AI citation is a source link the engine displays alongside an answer as support for a statement in that answer. An AI brand mention is your brand, product or executive named in the answer text, with or without a link. Brand mentions and AI citations measure different outcomes. Mentions show whether an AI-generated answer references your brand, while citations identify the sources presented to support that answer. Tracking both helps distinguish brand visibility from source visibility.<\/p>\n<h3 id=\"how-long-does-it-take-for-pr-coverage-to-show-up-in-ai-answers\">How long does it take for PR coverage to show up in AI answers?<\/h3>\n<p>There is no documented eligibility window. Google's AI features guidance specifies indexing, snippet eligibility and inclusion in Search's generative features, states that meeting the requirements does not guarantee inclusion, and publishes no minimum page age or waiting period, so start observing the day coverage goes live. Fast pickup does happen - Otterly.ai documented a press release cited by Microsoft Copilot two days after publication - but that is a counterexample, not an expected return. Record the date your tracker first sees a citation and label it \"first observed,\" because your collection schedule cannot establish when a platform discovered the page. The commonly repeated \"six to twelve months before you are eligible\" is not a platform rule.<\/p>\n<h3 id=\"what-is-a-good-ai-citation-rate-after-a-pr-campaign\">What is a good AI citation rate after a PR campaign?<\/h3>\n<p>There is no industry benchmark worth reporting against, because rates depend entirely on your prompt portfolio, category, platform mix and scoring definitions. Your benchmark is your own baseline range: if the observed change is smaller than your baseline's documented week-to-week range, it is not lift yet. Judge the size of the change against the uncertainty your sample can support - NIST's binomial example shows 8 successes in 30 observations gives a 26.7% rate with a 95% interval of roughly 14.2% to 44.4% - and do not import headline lift figures from vendor studies into your own forecast.<\/p>\n<h3 id=\"how-do-you-calculate-ai-share-of-voice\">How do you calculate AI share of voice?<\/h3>\n<p>AI share of voice = answers mentioning your brand \u00f7 the sum of answer appearances for every brand in your fixed competitor set \u00d7 100. Compute it on unbranded and comparison prompts only, because branded prompts inflate it mechanically, and keep the competitor set unchanged so the denominator stays stable. Never add AI share of voice to media share of voice; they use different populations and denominators.<\/p>\n<h3 id=\"which-ai-platforms-should-you-track-after-a-pr-campaign\">Which AI platforms should you track after a PR campaign?<\/h3>\n<p>Track ChatGPT, Perplexity, Gemini and Google AI Overviews, keeping each interaction mode fixed and recorded as its own surface. If you can only afford two, choose ChatGPT and Google AI Overviews, because they sit at opposite ends of the source-age spectrum and together bracket the likely range of outcomes. Add Perplexity when you want the earliest read, since it cites on effectively every answer.<\/p>\n<h3 id=\"why-isnt-my-pr-coverage-showing-up-in-chatgpt\">Why isn't my PR coverage showing up in ChatGPT?<\/h3>\n<p>Check search-crawler access first: OAI-SearchBot governs access for search features and GPTBot governs potential model-training use, and they are independent settings. Then check topic fit between the article's actual contents and the questions you test, and whether your brand is named strongly enough in the piece to be cited for anything. Finally, check your sample size, because a brand appearing on 0 of 2 runs is unmeasured, not absent.<\/p>\n<h3 id=\"how-many-times-should-i-repeat-each-prompt-to-measure-ai-citations-reliably\">How many times should I repeat each prompt to measure AI citations reliably?<\/h3>\n<p>Start with five fixed prompts per topic, each repeated four to six times weekly on each platform, which gives 20 to 30 runs per topic per platform per week. Spread the repetitions across different days. Treat this as an operating starting point rather than a statistical guarantee: NIST's binomial example shows 8 successes in 30 observations carries a 95% interval of roughly 14.2% to 44.4%, so small samples cannot settle small differences.<\/p>\n<h3 id=\"can-i-measure-pr-impact-if-i-never-collected-a-pre-campaign-ai-baseline\">Can I measure PR impact if I never collected a pre-campaign AI baseline?<\/h3>\n<p>Partly. You can measure current visibility, verify citations to campaign coverage, and start prospective tracking immediately. You cannot reconstruct missing historical answers by asking a platform what it previously knew, and you should say so in the report. Use genuinely archived answers, backfill the media record where historical coverage exists, and find a comparison topic or market to stand in for the missing baseline.<\/p>\n<h3 id=\"does-a-citation-to-a-news-article-count-if-it-does-not-link-to-my-website\">Does a citation to a news article count if it does not link to my website?<\/h3>\n<p>Yes, when you have verified the article's campaign connection, count it as campaign coverage exposure. Count it as a brand-supporting citation only when the cited source actually supports a statement about your brand. Keep owned-domain citations in a separate column, because those are the only ones capable of producing referral traffic you can see in your own analytics.<\/p>\n<h3 id=\"how-do-i-tell-leadership-whether-the-pr-campaign-caused-the-increase\">How do I tell leadership whether the PR campaign caused the increase?<\/h3>\n<p>Say that a before-and-after comparison cannot establish causation, then present the strongest supported claim: the size of the change, the baseline variability range it exceeded, whether it persisted across windows, how many runs cited verified campaign URLs, and what the comparison group did. When all of those line up, the defensible phrase is \"the results support a campaign contribution,\" and the next action is to review remaining explanations before expanding spend.<\/p>\n<h3 id=\"what-tools-track-ai-citations-after-a-pr-campaign\">What tools track AI citations after a PR campaign?<\/h3>\n<p>A manual spreadsheet run on a fixed schedule is enough to start. Beyond that: AI prompt trackers such as Profound, Peec AI and Otterly.ai for the answer side, Google Search Console\u2019s generative AI performance reports for AI Overviews and AI Mode impressions on your own verified property, and BrandMentions for the continuous web, news, blog, forum and social mention baseline that explains what fed the answers. No tool removes the need for URL-level human verification of each citation.<\/p>\n<h2 id=\"conclusion-measure-the-rate-not-the-moment\">Conclusion: Measure the Rate, Not the Moment<\/h2>\n<p>The hard part of this work is not technical. It is giving up the screenshot.<\/p>\n<p>A single ChatGPT answer naming your brand feels like proof, and trading it for a citation rate with a control group feels like more effort for a duller number. But the screenshot evaporates on the next run, and everyone in the room who re-checks it learns that in about four seconds. The duller number is the one that holds.<\/p>\n<p>The teams that will own this conversation through 2027 are treating AI visibility the way finance treats revenue: a defined period, a documented baseline, a stated comparison, and an explicit note on what is attributable and what is not. None of it requires expensive software. It requires freezing a prompt portfolio, running it on a schedule, keeping the denominators honest, and reporting what the data says rather than what you hoped.<\/p>\n<p>So here is the next step, and it is deliberately small.<\/p>\n<p>Start by selecting five buyer-intent prompts and monitoring how your brand appears across AI search platforms. At the same time, use BrandMentions to track earned media coverage, online brand mentions, competitor visibility, and sentiment.<\/p>\n<p>Combining AI citation tracking with ongoing media monitoring helps you identify which PR placements are referenced in AI-generated answers, measure visibility changes over time, and make better-informed decisions about future campaigns.<\/p>\n<p>Do that and you will walk into the post-campaign review with something nobody else in your category has: a number you can defend when someone opens ChatGPT in the meeting and tries to break it. Start the baseline this week. The campaign can wait a month. The missing history cannot be rebuilt afterwards.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>A step-by-step measurement system for proving whether a PR campaign actually moved your brand&#8217;s citations across ChatGPT, Perplexity, Gemini and Google AI Overviews using baselines, frozen prompt portfolios and repeated runs.<\/p>\n","protected":false},"author":4,"featured_media":6234,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_ef_editorial_meta_date_first-draft-date":"","_ef_editorial_meta_paragraph_assignment":"","_ef_editorial_meta_checkbox_needs-photo":"","_ef_editorial_meta_number_word-count":"","footnotes":""},"categories":[224],"tags":[352,121,68],"class_list":["post-6240","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-visibility-seo","tag-ai-citations","tag-ai-search-visibility","tag-pr-measurement"],"acf":[],"_links":{"self":[{"href":"https:\/\/brandmentions.com\/blog\/wp-json\/wp\/v2\/posts\/6240","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/brandmentions.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/brandmentions.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/brandmentions.com\/blog\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/brandmentions.com\/blog\/wp-json\/wp\/v2\/comments?post=6240"}],"version-history":[{"count":11,"href":"https:\/\/brandmentions.com\/blog\/wp-json\/wp\/v2\/posts\/6240\/revisions"}],"predecessor-version":[{"id":6251,"href":"https:\/\/brandmentions.com\/blog\/wp-json\/wp\/v2\/posts\/6240\/revisions\/6251"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/brandmentions.com\/blog\/wp-json\/wp\/v2\/media\/6234"}],"wp:attachment":[{"href":"https:\/\/brandmentions.com\/blog\/wp-json\/wp\/v2\/media?parent=6240"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/brandmentions.com\/blog\/wp-json\/wp\/v2\/categories?post=6240"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/brandmentions.com\/blog\/wp-json\/wp\/v2\/tags?post=6240"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}