AI tools are answering your audience's questions right now, whether or not your organization is part of the answer. This guide walks through a one-afternoon, no-cost audit of how ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews describe, recommend, and cite you — and how to turn what you find into a fix list.
By Matt·Updated July 11, 2026
Quick answer
What is an AI visibility audit?
An AI visibility audit is a systematic, one-time check of how AI tools — ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews — describe, recommend, and cite your organization. You run it by asking each tool the real questions your audience asks, grading every answer for presence, accuracy, and citations, and turning the results into a prioritized list of fixes. No paid software is required.
It differs from a traditional SEO audit, which checks whether search engines rank your pages. An AI audit checks whether AI tools understand your organization well enough to describe it correctly and confidently. The audit is the baseline; the monthly measurement routine is what keeps it current.
Why audit this now
A growing share of your audience's first impressions are formed inside an AI answer, not on your website. A 2025 Pew Research Center study of real browsing behavior found that when a Google search produced an AI summary, people clicked through to a traditional result only 8% of the time — about half the rate of searches without one. When someone asks an AI tool "is the downtown clinic still doing walk-in vaccinations?" or "which accounting firms in town work with nonprofits?", the answer they read may be the only impression they ever get — and nothing in your analytics will show it happened.
You can't fix what you haven't seen. Most organizations that run this audit for the first time find at least one surprise: a discontinued program still being recommended, a peer organization answering questions they should own, or a whole section of their site that AI tools can't read at all. The audit turns those unknowns into a to-do list.
Three behaviors the audit accounts for
One question becomes many. Google's AI search breaks a single question into multiple smaller searches — Google calls it "query fan-out" — so your content is competing to answer specific small questions, not broad topics.
Mentions without links count. AI tools learn about you from any text that mentions you — news coverage, directories, forum threads — not just pages that link to your site. The audit looks at your whole footprint, not just your domain.
Answers vary run to run. The same question can produce different answers and different sources each time. That's how these systems work, so the audit looks for patterns across many questions rather than treating any single answer as a verdict.
How to run the audit, step by step
Five steps, roughly an afternoon of work, ending in a prioritized fix list.
1
Write your question list — 10 to 15 real questions
Phrase them the way your audience actually talks, and cover three kinds: topic questions your organization should be an answer to ("where can seniors get help with taxes in [county]?"), service questions about what you offer ("does the historical society do school group tours?"), and branded questions about you by name ("is [organization] legitimate?"). Write them down and freeze the wording — you'll reuse this exact list for monthly measurement later.
2
Run every question in every engine
Free tiers cover all of them; use fresh or logged-out sessions where possible so your own history doesn't color the results. We have a walkthrough for each: ChatGPT, Google AI Overviews, Gemini, Claude, and Perplexity. Record each answer as you go — screenshots or pasted text, plus the date.
3
Grade every answer, not just scan for your name
For each answer record four things in a spreadsheet: were you mentioned (yes/no), was the information accurate (yes/partly/no), was your website cited as a source (yes/no), and which other organizations were named. Then apply the quality checks in the "AI lens" section below — how you're described matters as much as whether you appear. The AI visibility scorecard gives you a ready-made structure for all of this.
4
Check whether your own website is part of the problem
If engines get you wrong or miss you entirely, the cause is often technical or structural: a robots.txt rule blocking AI crawlers, key information locked inside PDFs or scripts that crawlers can't read, or pages with no structured data to confirm basic facts. Run the website readiness check as a companion to this step.
5
Turn findings into a prioritized fix list
Sort what you found into three piles, in priority order: accuracy errors (an AI tool is saying something wrong about you — fix the source pages immediately), absence (questions where you should appear but don't — see finding missing organizational information), and losing to others (a peer gets recommended instead — start with analyzing your competitors to learn why). One afternoon of auditing typically produces a quarter's worth of well-targeted work.
What to audit for in each engine
The major tools build answers from different places, so each deserves its own focus. Behavior described as of July 2026 — these products change often.
Dimension
ChatGPT
Claude
Gemini
Perplexity
Where answers come from
Its training plus live web search, drawing heavily on established reference sites and directories.
Its training plus live web search; responds well to pages that state facts in clear prose.
Google's own index and Knowledge Graph — the same infrastructure as Google Search.
Searches the live web for every answer and cites its sources directly.
Whether your own pages state who you are and what you do plainly, without marketing language.
Your structured data, consistent organization details across the web, and traditional search health.
Whether current, crawlable pages — yours and others' — mention you on the topics you care about.
Look first at
Your third-party listings
Your own site's clarity
Your schema & Google presence
Recent off-site mentions
ChatGPT
Where answers come from
Training plus live web search; leans on reference sites and directories.
Audit focus
How third-party sites describe you.
Look first at
Your third-party listings
Claude
Where answers come from
Training plus live web search; favors clear, factual prose.
Audit focus
Whether your own pages state facts plainly, without marketing language.
Look first at
Your own site's clarity
Gemini
Where answers come from
Google's own index and Knowledge Graph.
Audit focus
Structured data, consistent org details, traditional search health.
Look first at
Your schema & Google presence
Perplexity
Where answers come from
Live web search for every answer, with direct citations.
Audit focus
Current, crawlable mentions of you — on your site and others'.
Look first at
Recent off-site mentions
The "AI lens": grading the quality of an answer
Appearing in an answer isn't the finish line. When you grade answers in step 3, look at how you're presented.
The company you keep. When an AI tool lists you alongside other organizations, are they the right ones? A community health clinic grouped with urgent-care chains, or a state association grouped with for-profit training companies, tells you the engines have misfiled what you are — usually because your own pages never say it plainly.
Footer sources vs. inline citations. There's a difference between your URL appearing in a list of sources at the bottom of an answer and a citation attached directly to a specific claim. An inline citation means the tool treated your page as the answer to that particular question — the strongest signal you can earn. (What makes a page earn it: what makes content citation-worthy.)
Hedging language. Watch the qualifiers. Phrases like "however, some reports suggest…" or "though information may be outdated…" mean the tool found conflicting or stale information about you somewhere. Note the hedge and hunt down its source — often an old news story or an out-of-date directory listing.
What an audit turns up in practice
Two representative scenarios — one absence problem, one accuracy problem.
The library whose events didn't exist
What the audit found
A county library system asks "what's happening at the library this weekend?" across the engines and gets either generic advice or a neighboring county's events. The cause surfaces in step 4: every event lives inside a calendar widget that loads via script, so AI crawlers see an empty page.
What it changed
The library adds a plain-HTML "This month at the library" page — dated listings, one paragraph each, Event structured data — updated monthly. Two audits later, Perplexity and AI Overviews both surface its events for weekend-activity questions.
The lesson: the audit located where the failure was — technical invisibility, not content quality. Without it, the instinct would have been "write more blog posts," which would have fixed nothing.
The food bank with last year's schedule
What the audit found
A regional food bank asks "how do I get food assistance in [county]?" and finds engines naming it — but describing a distribution schedule discontinued a year ago, pulled from an old news story and a stale 211 directory listing. Its own site is never cited.
What it changed
The team rewrites its distribution page to lead with the current schedule and a "last updated" date, corrects the 211 listing and two partner-site listings, and adds Organization schema. Follow-up checks show the engines quoting the current schedule and citing the food bank's own page.
The lesson: accuracy errors are the audit's most urgent finding — a wrong answer can send someone to a closed distribution site. And the fix often lives off your website, in listings you don't control but can correct.
Run your audit this week
Five steps from zero to a baseline you can act on.
One evidence note for the skeptics on your team: the fixes this audit points you toward aren't guesswork. The original academic research on generative engine optimization — Aggarwal et al., presented at KDD 2024 — found that changes like adding citations, quotations, and statistics improved a source's visibility in AI-generated answers by up to 40% in benchmark testing. And if you'd rather have the whole audit run for you, tell me about your situation.
Common questions
If we rank #1 on Google, will AI tools automatically recommend us?
Not necessarily. Strong Google rankings are a useful foundation, but AI tools regularly pass over top-ranked pages in favor of lower-ranked ones that answer the question more directly or are backed up by other sources. The only way to know how AI tools actually treat your organization is to ask them your audience's real questions and read the answers.
Should we block AI crawlers in our robots.txt?
Only if you have content you genuinely need to keep out of AI systems. Blocking crawlers like GPTBot also blocks the live web lookups AI tools use to answer current questions — which, for most organizations, means disappearing from AI answers entirely. If your goal is to be found and recommended, blocking is usually self-defeating.
Do we need a paid tool to run an AI visibility audit?
No. A one-time audit needs free accounts on the major AI tools, a written list of real questions, and a spreadsheet. Paid monitoring platforms exist for large organizations tracking hundreds of queries continuously — they are a later upgrade, not a requirement to start.
How long does the audit take?
Plan for about an afternoon. Running 10–15 questions through four or five AI tools takes two to three hours, plus another hour to grade the answers and write up what you found. The fixes the audit surfaces take longer — but the audit itself is what tells you which fixes are worth your limited time.
How often should we repeat it?
The full audit is a baseline: run it once, then again after major changes to your site, programs, or leadership — or roughly once a year. Between audits, a lighter monthly check of the same questions is enough to track whether things are improving.