An AI visibility audit shows where your brand gets recommended, mentioned, misdescribed or skipped when buyers ask AI about your category. This is the process we run at VisibleIQ, compressed into 7 steps you can run yourself in an afternoon.
- Test real buyer questions: around 100 AI prompts plus a Google keyword set, roughly 85% unbranded, weighted toward decision-stage questions.
- Run every prompt more than once, since AI answers shift from one run to the next.
- Classify every appearance into 4 tiers: Recommended, Cited, Mentioned, Absent.
- Check that AI crawlers can reach your key pages, then check the reviews, listicles and comparisons the engines quote.
- Score recommendation rate, accuracy, sentiment and share of voice against named competitors, then turn the gaps into a revenue model and a 90-day plan.
An AI visibility audit, sometimes sold as a GEO audit or an AEO audit, is the first thing we run in the growth strategy plan we build before an engagement: what the AI engines say about a company’s category, who they recommend, who they skip, and why. Our full version goes deeper on volume, scoring and revenue modeling, but the bones below are the same, and a marketing lead can run this lean pass in an afternoon without buying anything. The query list decides your result, so read how to find the gaps in your AI search visibility alongside this.
What is an AI visibility audit?
An AI visibility audit is a structured test of how AI engines describe your brand when buyers ask about your category. It measures 6 things across a fixed set of buyer questions: mentions, recommendations, citations, answer accuracy, sentiment, and share of voice against named competitors. You come out with a scored gap list and a plan, not a screenshot folder.
Notice what’s missing from that list. Traffic. Rankings sit in the inputs and revenue sits in the outputs, but AI search is an influence channel, so the audit itself measures influence: what the engines say about you when a buyer with budget asks a buying question. A brand can be mentioned warmly and never cited, cited once and recommended nowhere, or described with pricing that’s 2 years out of date. Each of those is a different problem with a different fix, which is why lumping them into one “visibility score” hides more than it shows.
Why your Google rankings don’t tell you what AI says
A traditional SEO audit scores how you rank for keywords. An AI visibility audit scores how the engines talk about you, and the 2 can disagree completely. A site can hold page 1 positions on Google and still be missing from the AI answers in its category, because the engines build answers from sources a rank tracker never looks at.
Our own citation study showed why. Across the other engines we tested, 79.0% of product-query citations point to third-party websites: the reviews, listicles and comparisons written about you. ChatGPT flipped the pattern and pulls 74.6% of its product-query citations from vendors’ own sites, going straight to pricing and product pages.
So there are 2 layers to audit, and they fail independently. Your site can be in great shape while the third-party layer says nothing about you, and the third-party layer can glow while your own pricing page is unreadable to the crawlers. It’s just how the game works now, and the audit has to cover both.
A 30-minute call where we map what AI says about you and build a tailored roadmap.
How to run an AI visibility audit in 7 steps
The full version runs about 500 Google keywords and 100 AI prompts through every major engine, then classifies every appearance. The lean version follows the same 7 steps with 30 to 50 prompts, a spreadsheet and an afternoon. The order below matches our internal process, trimmed to what you can do yourself.
Step 1: Build the query sets
Our full audit tests around 600 queries: 100 AI prompts and about 500 Google keywords. The prompts are the star, because they’re what buyers type into the engines. The keyword set is there on purpose: the engines retrieve from Google’s index, and Google’s AI Overviews trigger on keywords, so your Google footprint sits upstream of your AI answers. Audit the prompts without the keywords and you can see that you’re absent without seeing why.
Keep both sets roughly 85% unbranded and 15% branded, and weight the prompts toward the decision stage: the “best X for Y”, “X vs Y” and “is X worth it” questions where vendors get picked. A buyer asking “what is desktop virtualization” is reading. A buyer asking “best VDI provider for a regulated company” is choosing.
Pull the phrasing straight from sales calls and transcripts, because buyers’ words beat keyword-tool words: the problem in their language, the constraints they name, the integrations they care about, the competitors they compare you against. The 50-prompt set behind our case study came from that method, transcripts first and keyword tools second.
Step 2: Run every prompt on every engine, more than once
Run each prompt on every AI engine your buyers use, and run the keyword set against Google with its AI Overviews. Same queries, same week, logged in one sheet. Then run each prompt 2 or 3 more times. AI answers vary between runs, so a single pass can show you on the list one minute and off it the next. What you’re measuring is how often you appear across runs, and a one-off result is noise.
Step 3: Classify every appearance into 4 tiers
For every answer that comes back, classify how your brand shows up:
| Tier | What it means |
|---|---|
| Recommended | The engine explicitly recommends you |
| Cited | Your URL appears in the citations |
| Mentioned | Your name appears in the text, without a citation |
| Absent | You’re not in the answer at all |
Log who else appears on every prompt, because the competitive picture matters as much as your own. Prompts where nobody dominates are open territory, and those are often the cheapest wins on the whole map.
Step 4: Check AI crawlers can reach your key pages
Before you judge the answers, make sure the engines can read you. It’s the cheapest step and the one teams skip. Open your robots.txt and check the AI crawlers by name. OpenAI runs separate bots for training (GPTBot) and for search (OAI-SearchBot), so blocking GPTBot doesn’t take you out of ChatGPT search, and blocking OAI-SearchBot does. Perplexity crawls as PerplexityBot and Anthropic as ClaudeBot. Google-Extended controls whether Google can use your content for Gemini training; it doesn’t change whether you appear in Google Search or its AI Overviews.
Then load your pricing, product and comparison pages with JavaScript switched off. If the prices, plans or feature lists vanish, the crawlers ChatGPT sends to your site may be reading an empty page.
A note on schema and llms.txt, since every audit checklist now includes them: we treat both as nice add-ons. Neither is the game. The engines read well-structured pages fine, and no markup rescues a page that has nothing quotable on it. Spend the hours on the evidence layer instead.
Step 5: Check the evidence layer
The engines lean on third-party sources, so audit the sources themselves. Which review platform does AI quote in your category? For the client in our case study, G2 appeared in the brand’s AI mentions more than any other review site, which told us exactly where the review work should go. Check whether the listicles and comparison pages in your category include you, who wrote them, and whether the best comparison of you was written by a competitor.
Step 6: Read branded and unbranded separately
Unbranded results answer “can buyers discover you?” Branded results answer “what do buyers find when they check you out?” They’re different business problems with different fixes, so don’t blend them into one score. Watch for competitors appearing inside your own branded answers. When a buyer asks an engine about you and the answer pitches somebody else, that prompt cluster goes to the top of the fix list.
Step 7: Model the revenue gap and build the 90-day plan
For each gap, estimate what it’s worth: monthly searches, times the click rate at the position you’d earn, times your visitor-to-lead rate, times your close rate, times your deal size. Present it as a range with a break-even, as in “you need 2 deals at your ACV for this to pay for itself”, because ranges survive scrutiny and single numbers don’t. Group the gaps by page type (comparisons, use cases, buyer research, education), then map the work. The AI answers are the goal, and SEO and content, review management and digital PR are the levers that move them.
What your audit findings mean
Most audits surface one of 3 patterns, and each points at a different fix. Ranking on Google while AI skips you is an evidence problem. Being named but misdescribed is a source problem. A competitor inside your branded answers is a positioning problem. Read the pattern before you build anything, because the fixes pull in different directions.
| What you found | What it usually means | Where the fix starts |
|---|---|---|
| You rank on Google but AI skips you | Thin third-party evidence: few reviews, no listicle placements, no independent comparisons | Reviews, roundups and digital PR in the sources the engines quote |
| You’re named but described wrongly | Outdated or inconsistent facts on your site or the pages that describe you | Fix the source page, then the third-party pages repeating it |
| A competitor shows up in your branded answers | Comparison pages about you were written by someone else | Publish your own comparisons and earn independent ones |
You rank on Google but AI skips you
This is the most common finding, and it’s rarely a content problem on your own site. The engines want corroboration from other people’s pages, and without it they hedge. Our breakdown of why B2B SaaS brands don’t show up in ChatGPT walks through the 6 causes behind it.
You’re named but described wrongly
Wrong pricing, a retired feature or last year’s positioning usually traces back to one page the engines trust. Find it, fix it, and the answer moves, often within weeks. We’ve written up why AI describes products wrong and how to trace the source.
A competitor shows up in your branded answers
When a buyer asks about you and hears about a rival, somebody else wrote the comparison the engine is reading. That’s the prompt cluster to fix first, since it’s where you lose buyers who were already looking for you.
Which tools can run an AI search audit?
You don’t need a paid tool for the first audit: the engines are free to query and a spreadsheet holds the results. Paid AI visibility trackers add scale, repeat runs and ongoing monitoring. SEO suites such as Ahrefs or Semrush cover the Google keyword half. What no tool does is the fixing, which is where the budget ends up.
Monitoring tools price by prompt volume, engines covered and competitors tracked. Before you buy one, check 3 things: whether it lets you bring your own prompt list, whether it runs each prompt more than once, and whether it separates recommendations from plain mentions. A tool that blends those into a single score won’t tell you much you can act on. If you’re weighing tools against services, our guide to what AI search optimization costs lays out the price bands.
What the scores mean (and what they don’t)
A visibility score only means something if you can say what it measures. Ours blends brand visibility, recommendation rate, share of voice, answer accuracy, sentiment and platform coverage, reported against named competitors on a defined prompt set. If a tool hands you a score and can’t tell you the denominator, you have a number, and a number isn’t a measurement.
AI search measurement is still young, so the discipline has to come from the method: a fixed prompt set, a defined classification, competitors named upfront, and the same test re-run the same way every month. Citations get treated as the whole scoreboard far too often. They’re the receipt. The metrics that move pipeline are recommendation rate, answer accuracy and share of voice, because those decide whether a buyer hears your name and hears it right.
What a real result looks like
The clearest proof of what this audit produces is the engagement it kicked off: a $10M B2B SaaS client added $1.1M in ARR from organic and AI search in 8 months. AI referrals made up 4.2% of organic sessions and drove 23% of the revenue, which is why we track answers and pipeline together.
The chain ran the way the steps above suggest. Buyer questions became a 50-prompt set, the prompt set exposed the gaps, and the gaps became decision-stage pages, a review campaign and PR placements. Then the answers moved. The case study publishes the funnel, the split and the traffic decline we inherited along the way, because a result you can interrogate beats a result you have to take on faith.
Run it yourself or bring us in
Run it yourself if you have an afternoon, a spreadsheet and a writer who can act on what you find. The lean version works, and running it internally teaches your team how the engines see your category, which is worth the afternoon on its own. Bring in help once the findings span reviews, PR and content at the same time.
At that point it’s a program, and programs need owners who’ve done it before. It obviously comes down to budget, right? If you’re a lean team pre-Series A, run it yourself every quarter and fix the top 3 gaps. If pipeline depends on organic and the gaps span 3 service lines, that’s what an engagement is for.
If you want help acting on the findings, our AI Search Optimization service covers the source corrections, content, authority work, and ongoing measurement.
How often to re-run it
Track your core prompts every month and re-run the full audit each quarter. AI answers move when models update, when competitors publish and when review pages shift, so a set-and-forget audit goes stale inside 90 days.
The audit is the snapshot; the tracking is the movie. Watching the same 50 prompts month over month is how you catch an engine changing its mind about you while there’s still time to do something about it.
Frequently asked questions
The questions founders and marketing leads ask us most about auditing AI visibility, answered the way we answer them on calls: how many prompts you need, whether it costs anything, how it differs from an SEO audit, what a GEO audit is, how to treat AI crawlers, and how long the whole thing takes.
How many prompts do you need for an AI visibility audit?
30 to 50 gets a lean audit that finds your biggest gaps, and around 100 covers a category properly. Volume matters less than sourcing: prompts phrased the way real buyers talk, weighted toward decision-stage questions, mostly unbranded.
Is there a free way to run an AI search visibility audit?
Yes. The engines are free to query, and the 7 steps above need nothing more than a spreadsheet. Paid tools add scale and ongoing tracking, which matter once you’re acting on the findings, but that first audit doesn’t cost more than an afternoon.
What’s the difference between an SEO audit and an AI visibility audit?
An SEO audit scores how you rank: keywords, technical health, backlinks. An AI visibility audit scores how engines answer: mentions, recommendations, accuracy, sentiment, share of voice. They overlap on crawlability and authority, and the second one starts where the rank tracker stops.
Is a GEO audit the same as an AI visibility audit?
In practice, yes. GEO (generative engine optimization) and AEO (answer engine optimization) are other names for the same discipline, so a GEO audit or AEO audit covers the same ground: how AI engines answer questions about your category and where your brand sits in those answers.
Should you allow AI crawlers in robots.txt?
For most B2B SaaS companies, yes, at least the search crawlers. Blocking OAI-SearchBot or PerplexityBot removes you from the answers those engines build from live pages. Blocking training crawlers such as GPTBot is a separate call about model training, and it doesn’t take you out of ChatGPT search.
How long does an AI visibility audit take?
The lean version takes an afternoon. The full version, with 2 query sets, every engine, competitor classification, scoring and a 90-day roadmap, takes 1 to 2 weeks including the write-up.
