How to Run an AI Visibility Audit: A Step-by-Step Process
Contents
An AI visibility audit is a repeatable process that tests how often your brand appears, gets cited, and is described accurately when real customer questions are asked on ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews. You don't need a paid tool to run one: define your prompts, test them across platforms, log the results, check the technical foundation behind your top-cited pages, and repeat on a set schedule.
That last part matters more than it sounds. AI-referred traffic to U.S. retail sites grew 138% year over year in May 2026 (source: Adobe Digital Insights, reported by Digital Commerce 360, June 17, 2026), which means the channel this audit measures is no longer a rounding error. The six steps below walk through the full process end to end, with a copy-paste prompt formula and a tracking table you can reuse every cycle.
What Is an AI Visibility Audit, and Why Doesn't Ranking on Google Guarantee It?
An AI visibility audit is a structured check of whether your brand gets mentioned, cited, or accurately described when AI platforms answer questions your buyers actually ask, as opposed to a traditional SEO audit, which checks how your pages rank in a list of blue links. Ranking well in classic Google search does not automatically carry over. Roughly 48% of tracked queries triggered an AI Overview in February 2026, up from about 30% a year earlier (industry research across millions of tracked queries, February 2026), and only around 17% of the sources cited inside those AI Overviews also appear in the top 10 organic results for the same query, a gap that has stayed roughly stable for months (industry research, February 2026). A page can hold position 3 on Google and still be completely invisible inside the generated answer sitting above it. Once you know a page is being cited, the next question is whether that citation actually sends anyone to your site, which is a separate thing to measure and covered in this guide to tracking traffic and CTR from AI Overviews.
That gap is exactly why an AI visibility audit exists as its own exercise. It isn't a rebrand of rank tracking, it's a separate measurement of a separate surface, one where citation, not position, is the unit that matters. If the underlying concept is still fuzzy before you run the steps below, this plain-English rundown of what AI visibility actually means covers it in more depth.
The Difference Between a Mention and a Citation
A mention is your brand named inside an AI-generated answer, with or without a link attached. A citation is a sourced, attributable reference the reader can click through to verify or learn more, the AI equivalent of a normal search result. An audit that doesn't separate the two will tell you that you exist inside AI answers without ever telling you whether anyone can actually reach your site from them, which is a meaningfully different problem to fix. Once you've run the audit below and want to turn these findings into a single number you can track over time, see this breakdown of how to calculate an AI visibility score.
Step 1: Define Your Scope and Choose Which AI Platforms to Test
Start by choosing which AI platforms you'll test consistently every cycle, since a partial or shifting platform list is the single easiest way to make an audit's results incomparable from one run to the next. For most brands, that means ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews at minimum, since together they now cover the large majority of everyday consumer AI search behavior. Adoption at that scale is no longer marginal: 44% of American adults say they use ChatGPT, according to a Pew Research Center survey fielded February 17-23, 2026 and published June 17, 2026 (source: Pew Research Center, "Americans and AI 2026").
Scope also means deciding, before you run a single prompt, whether this audit covers your whole brand or one specific product line, service, or location. A broad first pass across the whole brand is usually the right starting point; a narrower, product-specific audit becomes more useful once you already have a baseline and want to dig into a particular gap.
Which Platforms Actually Matter for a Small or Midsize Brand
A brand without a dedicated SEO team doesn't need to test every AI surface that exists, it needs the handful that its actual buyers use before making a purchase decision. ChatGPT and Google AI Overviews cover the broadest share of everyday research behavior for most B2B and consumer categories, Perplexity tends to matter more for research-heavy or technical purchases where buyers actively compare sources, and Gemini's weight depends heavily on how much of your audience already lives inside Google's ecosystem through Workspace or Android. Testing all four consistently, rather than picking a favorite and skipping the rest, is what prevents a false sense of security built on just one platform's results.
Step 2: Build a Prompt Set From Real Customer Questions, Not Guessed Keywords
The single biggest quality factor in an AI visibility audit isn't the platforms you test, it's whether your prompts sound like something a real buyer would actually type into a chatbot. Pull them from your own sales call notes, support tickets, live chat transcripts, and review sites instead of guessing at keyword-style phrases, since AI models respond to conversational, buyer-intent prompts very differently from short SEO keywords. A prompt set built from guessed keywords tends to test whether AI models understand your industry in the abstract; a prompt set built from real customer language tests whether they'd actually recommend you to someone in the exact position your buyers are in.
Mix branded prompts, questions that already name your brand or product, with non-branded prompts, the kind a buyer would ask before they've ever heard of you. Branded prompts mostly confirm that AI models recognize and describe you correctly. Non-branded, buyer-intent prompts test something more commercially urgent: whether you get recommended to someone who hasn't found you yet, which is where most of an audit's real value sits.
A Reusable Prompt Formula You Can Copy
A dependable starting formula is: "What's the best [category] for [specific constraint or use case]?" It forces the AI platform to make an actual recommendation rather than return a generic definition, which is what makes brand mentions and citations show up in the first place. Two examples built on that template: "What's the best project management tool for a 10-person marketing agency on a tight budget?" and "What's the best accounting software for a solo freelancer who invoices in multiple currencies?" Write 15 to 20 of these, evenly split between branded and non-branded, and reuse the exact same wording every audit cycle so a change in results reflects a real change in visibility, not a change in your methodology.
Step 3: Run Every Prompt Across Each Platform and Log the Results
Run your full prompt set against every platform in scope, one prompt at a time, and log the outcome in a shared table rather than trusting memory or scattered screenshots. A simple five-column log is enough to make the results usable later: whether your brand was mentioned, whether it was cited with a clickable link, how accurately it was described, whether competitors also showed up in the same answer, and which page, if any, got cited.
| Prompt | Platform | Mention? | Citation with link? | Description accuracy | Competitors mentioned? | Page cited |
|---|---|---|---|---|---|---|
| "What's the best accounting software for a solo freelancer who invoices in multiple currencies?" | Perplexity | Yes | Yes | Correct | Yes | /blog/multi-currency-invoicing-guide |
What to Track Beyond a Simple Yes or No
A plain yes-or-no count flattens results that actually behave very differently in practice. Description accuracy catches the case where you're mentioned but described with an outdated feature set or the wrong pricing tier, a problem no simple mention count would ever surface. Tracking which page gets cited, when a citation exists at all, tells you exactly which piece of content is doing the work, so you know what to protect and what to model future content on, rather than treating "visibility" as one undifferentiated blob of good news.
This article, Blue could have written it for you: content optimized for Google + AI, without you writing a single word.
Try mentionLABStep 4: Analyze Which Pages Get Cited, and Check the Technical Foundation Behind Them
Once the log is filled in, look for the pattern behind your citations rather than treating each one as a one-off. In practice, the pages that get cited most often across a brand's results tend to share three technical traits: clean structured data, open access for AI crawlers, and a clear, self-contained written structure, far more consistently than they share raw traffic volume. A high-traffic page with none of those three foundations can go completely uncited while a lower-traffic page with all three gets pulled into answer after answer.
Fix Crawler Access Before Anything Else
None of the content or structure work below matters if the AI platforms you're auditing can't actually reach your pages in the first place, so check crawler access before anything else on this list. This is a two-minute check, not a project: review your robots.txt file for accidental blocks on the crawlers behind ChatGPT, Perplexity, Claude, and Gemini before assuming a citation gap is a content problem. For the specific crawler names and a copy-paste-ready robots.txt setup, see this rundown of which AI crawlers to allow.
Add Structured Data So AI Platforms Can Parse the Page
Structured data gives an AI platform an explicit, machine-readable description of what a page is about, instead of forcing it to infer that from prose alone, which measurably speeds up how confidently a system can decide whether to cite you. If your top pages don't yet carry Article, Organization, or FAQ schema, that's frequently the fastest fix on this entire list relative to the effort involved. For the specific schema types worth prioritizing first, see this guide to schema markup for AI.
Give AI Crawlers a Clear Entry Point With an llms.txt File
An llms.txt file gives AI crawlers a curated, plain-text map of your most important pages instead of forcing them to guess at site structure from a full crawl. It won't fix a citation gap caused by thin or outdated content on its own, but it removes one small, low-effort barrier to being found and read in the first place. For the exact format and how to publish one, see this walkthrough of how to create an llms.txt file.
Step 5: Benchmark Your Results Against Competitors
Run the identical prompt set again, substituting two or three close competitors' names in for yours, and compare not just who shows up but how each brand is positioned relative to the others in the same answer. Raw mention counts alone can be misleading in isolation: a brand mentioned in every answer but always listed third or fourth, behind two rivals, has a real AI share of voice problem that a simple mention tally won't reveal. This is also where E-E-A-T signals, experience, expertise, authoritativeness, and trustworthiness, tend to separate winners from also-rans, since AI systems lean on those same signals when deciding which brand to lead with in a comparison.
A Citation Isn't Always Good News (Sentiment Matters)
Being cited is not automatically a win if the surrounding language undercuts the recommendation. A hedged answer, "Brand X is an option, though some users report it's harder to set up than competitors," is a citation and a mention, but it's not the outcome you were auditing for. Read the actual sentence around every mention you log, not just its presence, and flag anything hedged, comparative, or lukewarm separately from a clean, confident recommendation.
Step 6: Turn Your Findings Into a Prioritized Action Plan
Sort every gap the audit surfaced into one of three buckets before you touch anything: missing content the AI platforms have nothing to cite, technical barriers like blocked crawlers or missing schema, and inconsistent positioning or messaging that confuses how you're described. Each bucket calls for a different fix, and treating them as one undifferentiated "improve AI visibility" to-do list is how audits stall without producing any real action.
Close the Content Gaps AI Platforms Keep Skipping
When a buyer-intent prompt returns no mention of your brand at all, the likely cause is that nothing on your site directly and clearly answers that exact question, not that AI platforms are ignoring you. Map each unmentioned prompt back to a piece of content that should exist to answer it, and prioritize the prompts with the clearest commercial intent first. For the fuller set of tactics that earn citations once that content exists, see this breakdown of how to get cited by AI.
How Often Should You Repeat an AI Visibility Audit Without an In-House SEO Team?
Run a full AI visibility audit quarterly, with a lighter monthly check limited to your 5 to 10 highest-commercial-intent prompts, rather than trying to match the weekly or monthly cadence agencies typically recommend to clients paying for that level of attention. That heavier cadence assumes a dedicated team or a paid monitoring tool running in the background, which most small and midsize brands simply don't have, and chasing it usually means the audit quietly stops happening at all after the second or third month.
A quarterly full pass is enough to catch real, durable shifts in how you're described, while a monthly spot-check on your highest-value prompts catches anything urgent, a factual error, a dropped mention, a new competitor suddenly appearing, without demanding a full day of manual work every few weeks. Tools built for this kind of repeatable content and technical work, MentionLab included, run an automated GEO audit that checks schema, AI crawler access, and metadata on a standing basis, which is one way to keep this cadence going without dedicating a full day to it every quarter. If you're tracking the audit's results over time alongside other metrics, this list of SEO KPIs for AI search covers what else is worth watching in parallel.
Frequently Asked Questions
Do I need a paid tool to run an AI visibility audit? No. A basic audit only requires access to ChatGPT, Perplexity, Claude, and Gemini's standard chat interfaces, plus a spreadsheet to log results. Paid platforms mostly add automation and historical tracking, not a capability you can't reproduce manually with the process above.
How many prompts should I test in a first AI visibility audit? 15 to 20 prompts, split evenly between branded and non-branded, is enough for a first pass across four or five platforms. That's roughly 60 to 100 individual tests, manageable in an afternoon, and gives you a large enough sample to spot a real pattern rather than one or two noisy results.
What's the difference between an AI visibility audit and an AI visibility score? An AI visibility audit is the full process, defining prompts, testing platforms, logging results, analyzing citations, benchmarking competitors, described in this article. An AI visibility score is a single number, typically a percentage, that summarizes one cycle of that audit's results so you can track it over time. Run the audit first; calculate the score from what it produces.
Can a small brand actually compete with larger competitors in AI-generated answers? Yes, particularly on non-branded, buyer-intent prompts tied to a specific niche. AI platforms tend to reward depth of coverage on a narrow topic and clear, well-structured content over company size or ad spend on its own, which gives a smaller, more specialized brand a real path to outscoring a larger, more generic competitor on the exact prompts its own buyers actually use.
Does blocking AI crawlers in robots.txt affect my AI visibility audit results? Yes, directly. If your robots.txt file blocks the crawlers behind ChatGPT, Perplexity, Claude, or Gemini, those platforms cannot read your pages at all, regardless of how good your content or schema is, and your audit will show a citation gap that has nothing to do with content quality. Checking crawler access, covered in Step 4 above, should always come before assuming a missing mention is a content problem.
Blue handles your SEO and your GEO. On autopilot.
You approve, she produces content optimized for Google + AI.
Join the Lab · 5-day trial