AI Visibility Audit Checklist: 8 Steps for Every AI Engine

TL;DR

An AI visibility audit tests how often each AI engine mentions, cites and recommends you for real buyer prompts, then checks why.
Run every prompt several times on each engine and score each engine separately. Answers change from run to run, and the engines cite different sources.
Check crawler tokens one at a time. OpenAI and Anthropic each run separate bots for search, training and user requests, and blocking one does not block the others.
Map which sites get cited instead of you. Vendor studies point to listicles, comparison pages, review sites and YouTube.
Treat schema and llms.txt as cheap hygiene. Current vendor data shows little or no citation lift from either.

What is an AI visibility audit, and how do I run one?

An AI visibility audit is a repeatable test of how often AI assistants mention, cite and recommend your business for the questions buyers ask. It also checks why. To run one, list 20 to 30 buyer prompts and run each one several times on every engine. Log mentions and sources, then check crawler access, indexes and your business facts.

Most guides stop at "ask ChatGPT about your brand." That is only the first of eight steps. The other seven explain why you show up or not, and those are the parts you can actually fix. If you are new to this, start with what AI search visibility is, then come back.

What does an AI visibility audit measure?

It measures six things for each prompt and each engine: whether you are mentioned, whether you are recommended, whether your site is cited, whether the facts are right, the tone, and which other sources the engine used.

Mentions and recommendations are not the same thing. As one checklist puts it, a long-list mention should not receive the same weight as a first-choice recommendation. Keep them in separate columns.

Signal	What it means	How to log it
Mention	Your name appears	Yes or no, per run
Recommendation	You are a top pick	Position in the list
Citation	Your URL is a source	Which page was linked
Accuracy	Facts about you are right	Note each wrong fact
Sentiment	Tone of the description	Positive, neutral, negative
Sources	Who got cited instead	List of domains

Step 1: How do I pick the right prompts?

Use the words your buyers actually use, not your internal jargon. Aim for 20 to 30 prompts spread across four types of questions. Keep the wording fixed so later runs compare fairly.

One useful approach is to choose a small initial set of questions spanning discovery, comparison, verification, and purchase constraints. In practice that looks like this:

Discovery: "best [service] for [type of customer]" or "who should I hire for [problem]."
Comparison: "[you] vs [competitor]" and "alternatives to [competitor]."
Verification: "is [you] legit," "[you] reviews," "what does [you] charge."
Constraints: "[service] open on weekends," "[service] under $500," "[service] that takes insurance."

Pull the wording from customer emails, sales calls, reviews and your Google Search Console queries. Then use the exact same wording during the baseline, and identify each question with a stable ID, such as A01 or A02. Add two or three branded prompts like "What is [your business]?" These catch wrong facts fast.

Step 2: How do I run the prompts on each engine?

Run every prompt three or more times on each engine, in a clean session, and log the date, the engine and the mode. Score each engine on its own. They pull from different sources, so a single combined score hides the real problems.

A common tip is to use a logged-out or incognito session, since signed-in chat history skews answers toward brands you already engage with. Repeat runs matter just as much. One guide recommends running each prompt three to five times each . Variance is real. A BrightLocal study (a vendor study, relayed on X) found only 20 to 33% of results overlapped on a repeated search. Our own desk rule is to report only snapshots from logged re-runs, never a single screenshot.

Scoring per engine matters because the engines barely agree. One cited study found only 11% of sources overlap between ChatGPT and Perplexity. Here is what to watch on each engine:

Engine	Check first
ChatGPT	OAI-SearchBot allowed, Bing indexing
Claude	Claude-SearchBot and Claude-User allowed, Brave index
Gemini	Google index, Search Console health
AI Overviews and AI Mode	Indexed with a snippet, Search Console AI setting
Perplexity	PerplexityBot allowed, cited URLs per answer
Grok	Your X profile and posts, plus web presence
Copilot	Bing Webmaster Tools, Bing Places

Engine-by-engine guides: ChatGPT, Claude, Gemini, Grok, Perplexity and Google AI Overviews. For a quick first look, see how to check if AI recommends your business.

Step 3: How do I score the results?

Work out a mention rate for each engine and each prompt type: runs that mention you, divided by total runs. Do the same for recommendations and citations. Then compare those rates with your top three competitors, not with an arbitrary target.

Keep the signals apart at first. One analyst's advice is to not combine those outcomes into one score before reviewing them separately. A blended score can look fine while Claude never mentions you at all.

Some checklists set pass marks. One, for example, uses 20%+ on category-defining prompts . That is a fair rule of thumb, but no study backs a universal threshold. Your competitors are the honest benchmark. If they show up in 60% of runs and you show up in 10%, that gap is your finding.

Flag every factual error: wrong hours, old prices, services you dropped, a wrong owner name. Wrong facts often trace back to an old directory listing. That is Step 6.

Step 4: Can AI crawlers reach your site?

Check robots.txt for each vendor's search bot by name, not just "AI bots" as a group. OpenAI and Anthropic each run separate bots for search, training and user requests. Blocking one does not block the others.

OpenAI's docs say a webmaster can allow OAI-SearchBot in order to appear in search results while disallowing GPTBot to indicate that crawled content should not be used for training . They also warn that sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links. After you change robots.txt, it can take ~24 hours from a site's robots.txt update for our systems to adjust.

Anthropic's crawler page lists ClaudeBot (training data collection), Claude-User (fetching pages when Claude users ask questions), and Claude-SearchBot (indexing content for search results) as separate bots . Anthropic says blocking Claude-SearchBot "may reduce your site's visibility and accuracy in user search results." One difference is worth knowing: Anthropic says all three of its bots honor robots.txt, including Claude-User, while OpenAI and Perplexity warn that robots.txt rules may not apply to ChatGPT-User and generally don't apply to Perplexity-User.

Company	Search and answer bots	Training bot
OpenAI	OAI-SearchBot, ChatGPT-User	GPTBot
Anthropic	Claude-SearchBot, Claude-User	ClaudeBot
Perplexity	PerplexityBot, Perplexity-User	None listed separately
Google	Googlebot	Google-Extended

robots.txt is only half of this check. Also confirm that your CDN and firewall settings let the AI search crawlers read your pages. Bot protection can block crawlers even when robots.txt allows them. And make sure your key content is in the page HTML, not loaded only by JavaScript. The full walkthrough is in how to allow AI crawlers in robots.txt.

Step 5: Are you in the indexes these engines search?

AI engines mostly cite pages they can find in a search index. So check Google, Bing and Brave. If a page is missing from the index an engine uses, that engine has a hard time citing it.

Google. Google's documentation says "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." A page must be indexed and eligible to appear in Google Search with a snippet. Google also says both features may use a "query fan-out" technique, issuing multiple related searches across subtopics and data sources . That means pages covering subtopics (pricing, comparisons, specific use cases) give you more chances to be pulled in.

There is one new item to check. Google has started testing a Search Console control where website owners can decide if they want their site to appear in and help ground responses in our generative AI Search features . Make sure nobody on your team turned that off. Target Internet also reports that in June 2026 Google added dedicated Search generative AI performance reports in Search Console . If you can see them, use them as your Google baseline. More detail is in Google AI Mode for local businesses.

Bing. Verify your site in Bing Webmaster Tools, submit your sitemap and turn on IndexNow. Microsoft Copilot runs on Bing, and many practitioners believe ChatGPT search leans on it too. Be careful with the popular claim that "87% of ChatGPT citations match Bing." The post that spread it gives no source, and I have not seen OpenAI confirm any figure. If you are a local business, claim Bing Places as well.

Brave. In our desk's testing notes, Claude's web results seemed to lean on Brave Search. Treat that as an observation, not a documented rule. Search site:yourdomain.com on Brave. If key pages are missing, that is a gap worth closing.

Step 6: Are your business facts the same everywhere?

Put your name, phone, hours, categories and one-line description side by side across your site and every major profile. They should match exactly. When they conflict, AI engines may repeat the wrong version or skip you.

Check your site, Google Business Profile, Bing Places, Apple Business Connect, Yelp, your industry directories, LinkedIn and X. Some data points suggest Google is not the only source that matters. A BrightLocal vendor study, as relayed on X, found Yelp was the second biggest local source and that ChatGPT barely used Google Business Profiles. In our own desk notes, Claude and Gemini cited directories and listicles like Clutch, DesignRush and The Manifest for "best agency" prompts.

Get real reviews on the platforms your buyers use. Never buy or fake them. The FTC has a final rule banning fake reviews and testimonials, and fake reviews are not worth the risk.

Step 7: Which sources get cited instead of you?

For every prompt, list the domains each engine cited. Group them into types: listicles, comparison pages, review sites, directories, Reddit, YouTube, competitor sites and your own site. The pattern shows you where to earn a presence.

This is the most useful part of the audit, and it is where vendor studies help, as long as you treat them as vendor studies. They are single snapshots run by companies that sell tools:

Listicles: Ahrefs reports that "best X" listicles make up 43.8% of the page types ChatGPT cites.
Comparison and review pages: AirOps found ChatGPT, Perplexity and Claude send 52 to 72% of citations to off-site product and comparison pages and 16 to 21% to review platforms.
Reddit and YouTube: The same AirOps data puts Reddit in 39 to 52% of AI Overview and AI Mode answers and YouTube in 38 to 47%, but says Claude rarely cites either one.
YouTube correlation: Ahrefs found YouTube mentions had the highest correlation (0.737) with AI brand visibility. That is a correlation, not proof that YouTube causes anything.
Google rank is not everything: Ahrefs says 28.3% of ChatGPT's most-cited pages have no Google visibility.

Now compare this with your own source list. If every engine cites the same three roundups and you are not in them, that tells you more than any score. If a competitor keeps winning, here is how to work out why.

Step 8: Does your own site answer each prompt?

For each prompt, find the page on your site that answers it directly. If no such page exists, that is a content gap. Engines can only cite you for an answer you have actually published.

There is some evidence that your own pages matter more than the third-party numbers suggest. In one single-client case covering about 15,000 answers, 85% of citations went to third-party pages, but 87% of answers that mentioned the client cited the client's own site. For local searches, BrightLocal reported that business websites got 42% of citations across about 200,000 AI searches (a vendor study).

The academic side points the same way. The original GEO paper found that GEO can boost visibility by up to 40% in GE responses . Its strongest methods added cited sources, quotations and statistics to the page. Its authors also note that GEO is more helpful if optimization methods are made for target domain , so what works in one industry may not work in yours. See generative engine optimization explained.

For each page, check that:

The answer comes in the first two sentences under a heading that matches the question.
Specific numbers (prices, time frames, counts) are on the page in plain text.
Comparisons use a real HTML table, not an image.
There is a visible "updated" date, and it is accurate.
Claims link to their sources.

Those are practitioner advice, not proven levers. They are cheap to do and they help human readers too.

What should you not spend audit time on?

Do not spend much time on schema tweaks or llms.txt. Both are cheap hygiene, but current data shows little or no citation lift from either. Spend the time on Steps 4 through 8.

Schema. Ahrefs found that schema had no meaningful impact on citations. Accurate schema still helps with rich results and with stating your facts clearly. Keep it correct, and keep it matched to what the page shows. More in local business schema for AI search.

llms.txt. Ahrefs log data across 137,000 domains, relayed on X, showed 97% of valid llms.txt files got zero requests in May. A widely shared experiment called the evidence "GEO astrology". On the other side, one developer firm reported Claude fetching it. It takes a few minutes to add, so add one if you like, then move on. See llms.txt for small business.

The full AI visibility audit checklist

Use this table as your scorecard. Each row is a pass or fail you can check yourself.

Check	Pass looks like	Red flag
Prompt list	20 to 30 buyer prompts, four types	Only branded prompts
Engine coverage	All major engines, scored apart	ChatGPT only
Repeat runs	3+ runs per prompt, dated	One screenshot
Mentions vs picks	Logged separately	One blended score
Accuracy	No wrong facts	Old hours or prices
Search bots	OAI-SearchBot, Claude-SearchBot, PerplexityBot allowed	Old blanket block
Firewall and CDN	Bots get pages, not challenges	Bot protection blocks all
Google	Key pages indexed, AI setting on	Pages not indexed
Bing and Brave	Key pages found	Missing from site: search
Entity facts	Identical on every profile	Conflicting phone or name
Source map	Cited domains grouped by type	Not recorded
Page fit	One page per key prompt	No page answers it

How do I turn the audit into a fix list?

Fix things in this order: blockers first, then wrong facts, then missing pages, then off-site presence. Then set a baseline and measure again. Each step depends on the one before it.

Remove blockers. Crawler rules, firewall settings, index problems and the Search Console AI setting.
Fix wrong facts. Correct every profile that disagrees with your site.
Build or fix answer pages. One clear page for each high-intent prompt you are missing.
Earn a place in the sources engines cite. Honest outreach to the roundups, directories and review sites from your source map, plus useful YouTube or Reddit content where your buyers really are.
Set a baseline and re-run. Use the same prompts, the same engines, and dated logs.

Another published checklist lands on a similar order. It starts with remove accidental crawl and index blockers , then fixes product facts, then improves high-intent pages.

A word on expectations. An audit tells you where you stand. It cannot make any engine pick you. Nobody controls those answers, so I can't promise any placement, citation or timeline, and you should be wary of anyone who does. If you want help with the work, here is how we work, what our services cover, what AI search optimization costs, and how to choose an agency.

How often should you re-run the audit?

Run your top 10 to 20 prompts monthly and the full audit quarterly. Re-run right away after any site migration or template change.

One guide suggests that monthly, you run your 10 to 15 priority money prompts manually across ChatGPT, Perplexity, and Google AI Overviews to catch significant changes between full audits. It also advises an immediate re-audit after any major CMS migration, URL structure change, or template revision . Those are the moments when crawl and index problems slip in quietly.

How long the first pass takes depends on your approach. One guide says you can do it in 6 to 10 hours with no paid tools. A tool vendor argues you should invest at least 14 days to scan the environment . Both can be right. A manual pass gives you a snapshot, and a tool shows you the trend over time.

How I put this together

On October 22, 2026, I searched "ai visibility audit" in the US and read the top three articles, skipping service and product pages:

Yotpo: AI Visibility Audit, Step-by-Step. It is written for ecommerce. It covers four stages: target queries, manual prompting and sentiment, mapping citations, and automated execution, plus KPIs. Its final stage leads into its own product. It does not cover crawler tokens, indexes or business facts.
Mentionlytics: AI Visibility Audit, Steps, Metrics, How to Do It for Free. It gives a clear definition and a useful comparison with a traditional SEO audit, and it handles sentiment and accuracy well. It leans on its own tool and says little about technical access or what to fix first.
Outreach Bloom: The Free 7-Step Checklist. This is the most thorough of the three, with a robots.txt reference and a pass/fail table. It recommends rolling out FAQ schema as a fix, which current vendor data does not support. It also sets fixed pass marks without a source and skips Grok.

Here is what this guide adds that none of the three cover together:

Grok and Copilot coverage.
Crawler facts checked against OpenAI's documentation and coverage of Anthropic's crawler page, including the robots.txt difference between Claude-User and Perplexity-User.
Google's new Search Console AI control, and a Bing and Brave index check.
Entity checks for local businesses.
Vendor evidence on what gets cited, labeled as vendor evidence, including why schema and llms.txt sit low on the list.

Statistics from X posts are attributed to whoever ran the study. None of them are settled fact. For background on our approach, see Get Found with AI, GEO vs SEO and ChatGPT vs Google for local search. Industry examples are in our guides for accountants and veterinarians. More posts are on the blog.

Frequently asked questions

How long does an AI visibility audit take?

A first manual pass takes several hours. One published guide estimates 6 to 10 hours with no paid tools. That covers 20 to 30 prompts, run a few times each on every engine, plus the crawler, index and entity checks. Tool-based audits take longer to collect data, because they watch answers change over days or weeks.

Which AI engines should an AI visibility audit cover?

Cover ChatGPT, Claude, Gemini, Grok, Perplexity, and Google AI Overviews and AI Mode. Add Microsoft Copilot if your buyers use Microsoft products. Score each engine separately, because they cite different sources. One cited study found only 11% source overlap between ChatGPT and Perplexity.

Does schema markup or llms.txt improve AI citations?

The current evidence says not much. An Ahrefs vendor study found schema had no meaningful impact on citations, and Ahrefs log data showed 97% of valid llms.txt files got zero requests in a month. Google says AI Overviews and AI Mode need no special optimizations. Keep accurate schema for rich results and clarity, but do not count on either one to win citations.

Why does ChatGPT give a different answer every time I check?

AI answers are probabilistic, and they pull from live search results that change. A BrightLocal vendor study found only 20 to 33% of results overlapped on a repeated search. Run each prompt several times, log every run with a date, and report rates, not single screenshots.

Should I block GPTBot in robots.txt?

That is your call. OpenAI says its bot settings are independent. You can disallow GPTBot to opt out of model training and still allow OAI-SearchBot so you can appear in ChatGPT search answers. Blocking OAI-SearchBot removes you from ChatGPT search answers, though navigational links can still appear.

Can an AI visibility audit guarantee my business gets recommended?

No. An audit tells you where you stand and what is blocking you. Nobody controls what an AI engine recommends, so I can't promise any placement, citation or timeline. What you can do is remove blockers, fix wrong facts, publish pages that answer buyer questions, and measure again.

How often should I re-run an AI visibility audit?

Run your top 10 to 20 prompts monthly and the full audit quarterly. Also re-run right after a site migration, a URL change or a template change, because those can quietly break crawling, indexing or structured data.