AEO audit checklist: what to check and in what order
8 min read
By Jake Bauman
ai-search / aeo

An AEO audit checklist tests three things: whether AI answer engines can reach a page, whether they can pull a clean answer out of it, and whether they trust it enough to cite it. Check in this order. Start with crawler access and the answer in the raw HTML. Then check sections that stand alone and schema with entity signals. Finish with real citation tests in ChatGPT, Perplexity and Google. The full checklist is a plain file further down. Run it on your ten most important pages before you rewrite anything.
What an AEO audit measures
An AEO audit measures whether your content gets cited inside AI answers, not where it ranks. AirOps frames the audit around four areas: technical access for AI crawlers, content structure, E-E-A-T and brand authority, and citation tracking (AirOps, March 2026). That is a sensible frame. The difference from a classic SEO audit is the output. A traditional SEO audit gives you a list of ranking problems. An AEO audit should give you a log of questions, engines and whether your page showed up as a source.
Ranking and citation overlap less than most teams assume. AirOps reports that roughly 60% of AI Overview citations come from pages outside the top 20 organic results (AirOps). Treat that as a reason to audit citations directly, not as proof that rankings stopped mattering.
Crawler access comes first
Crawler access is the first check because a blocked bot makes every other fix pointless. Open your robots.txt and confirm it does not disallow GPTBot, ClaudeBot, PerplexityBot, Bingbot or Googlebot. The audits from RevvGrowth and Stackmatix both start with this list. A blanket Disallow: / under User-agent: * blocks all of them at once, and that line is easy to miss on an old site.
Google-Extended needs its own note. It is a robots.txt token that controls whether Google can use your content for Gemini models. It is not a separate crawler. Google AI Overviews come from Googlebot and normal Search indexing, so blocking Google-Extended does not remove you from AI Overviews. Blocking Googlebot does.
Robots.txt only tells you what is allowed. Server logs tell you what happened. AirOps recommends confirming in server logs that AI bots actually visit (AirOps). If GPTBot has never requested a page, ChatGPT is not citing it from a fresh crawl.
The answer must be in the raw HTML, near the top
A page passes this check when the direct answer to its main question appears in the server-delivered HTML within the first paragraph. Retrieval systems do not reliably run JavaScript. If your answer loads client-side, sits behind a tab, or opens on click, a bot may read an empty shell. RevvGrowth lists hidden tabs and expand-to-view answers as failures for the same reason (RevvGrowth).
The test is one command. Fetch the page with curl and search the output for the first sentence of your answer. If the sentence is missing, the page fails, however good it looks in a browser. Fix rendering before you touch copy.
Placement matters as much as presence. AI Labs Audit tells you to answer the main question in the first 100 words (AI Labs Audit). RevvGrowth goes further and wants a 40 to 60 word answer directly under every H2 (RevvGrowth). I would hold every priority page to the first rule and every H2 to the second.
Each section must work when lifted out alone
A section passes when someone can read it with the rest of the page deleted and still get a complete answer. AI search scores relevance at the passage level. Aleyda Solís put it plainly in the AirOps piece: "With AI search this happens at a passage or chunk level of relevance" (AirOps). A chunk that opens with "As mentioned above" or "This approach" is useless once it is cut away from the page above it.
Three checks catch most failures:
- The H2 names the topic of the section in plain words.
- The first sentence under the H2 answers it.
- No pronoun or phrase points back to a different section.
Heading order is part of the same check. Use one H1, then H2s, then H3s. Do not skip levels for styling. AirOps reports that pages with clean, sequential heading hierarchies show 2.8 times higher citation likelihood (AirOps). I would fix headings anyway, because the change is cheap and it helps human readers too.
Schema and entity signals must agree with each other
Schema passes when it validates and matches what the page and the rest of the web say about you. The basic set is Organization on the homepage and Article or BlogPosting on posts. Add Person for the author, plus datePublished and dateModified. RevvGrowth and AI Labs Audit both recommend this set and validating in Google's Rich Results Test (RevvGrowth, AI Labs Audit). Use sameAs links in Organization schema to point at the profiles you control.
Consistency is the harder check. AI systems build their picture of a company from many sources. A homepage that says "CRM" and a LinkedIn page that says "data platform" give a model two answers. Compare your one-line description across your site, LinkedIn and the directories your buyers use. Do the same for your name, address and phone number if you are local. Foster's checklist for law firms and medical practices asks the right question here: does AI consistently identify your primary specialty (Foster Web Marketing). If the honest answer is no, fix the profiles before the page copy.
Citation tests need several engines and repeated runs
A citation test passes when your page appears as a source for a question your buyers really ask, in more than one engine, across more than one run. Write down your real questions first. Stackmatix suggests 10 to 20 priority questions per platform (Stackmatix). RevvGrowth suggests 20 to 30 high-intent queries (RevvGrowth). Ten is enough for a first audit.
Run each question more than once. AirOps found that only about 30% of brands stay visible from one answer to the next. Only 20% stay visible across five consecutive runs (AirOps). A single screenshot of ChatGPT citing you tells you almost nothing. Three runs per question per engine, logged with the date, give you a pattern you can act on.
Also record whether Google shows an AI Overview at all. On October 9, 2026, Google showed no AI Overview for the query "aeo audit checklist" on US English desktop. On queries without an Overview, the Google part of the citation test is just the organic result. Do not count a missing Overview as a lost citation.
Published AEO statistics conflict, so trust your own log
Published AEO numbers disagree with each other, so treat them as hypotheses and let your citation log decide. AirOps says roughly 60% of AI Overview citations come from pages outside the top 20 organic results (AirOps). AI Labs Audit, updated May 2026, says 97% come from the top 20 (AI Labs Audit). Those figures cannot both describe the same queries. Methodology, query mix and timing probably explain the gap.
Freshness advice splits the same way. AI Labs Audit says to update every 30 days (AI Labs Audit). RevvGrowth says quarterly for top pages (RevvGrowth). My position is to update when a fact changes and to show an honest dateModified. Bumping dates without changing content is noise.
The checklist file
Save this as aeo-audit.md. Make one copy per page, replace example.com with your domain, and work top to bottom. Stop at the first section with an unchecked box and fix that before moving on.
# aeo-audit.md
Page URL:
Main question this page answers:
Audit date:
## 1. Access
[ ] robots.txt does not block GPTBot, ClaudeBot, PerplexityBot, Bingbot, Googlebot
curl -s https://example.com/robots.txt
[ ] No blanket "Disallow: /" under "User-agent: *"
[ ] Server logs show AI bot visits in the last 30 days
grep -E "GPTBot|ClaudeBot|PerplexityBot" /var/log/nginx/access.log | tail -20
[ ] Page returns 200, no redirect chain, no noindex
curl -sIL https://example.com/page | grep -iE "^HTTP/|^location|x-robots-tag"
[ ] No meta robots noindex tag in the HTML
## 2. Answer in raw HTML
[ ] First sentence of the answer appears in raw HTML
curl -s https://example.com/page | grep -i "paste first 6 words of the answer"
[ ] Answer sits in the first 100 words under the H1
[ ] No core content behind tabs, accordions or click-to-load
## 3. Standalone sections
[ ] One H1, then H2s, then H3s, no skipped levels
[ ] Each H2 names its topic in plain words
[ ] First sentence under each H2 answers that H2
[ ] No "as mentioned above", "this", "they" pointing to another section
[ ] Every statistic links to its source
## 4. Schema and entity
[ ] Organization schema on homepage with sameAs links
[ ] Article or BlogPosting schema on this page
[ ] Person schema for the author, with a real bio page
[ ] datePublished and dateModified present and honest
[ ] Passes Google Rich Results Test with no errors
[ ] One-line company description matches site, LinkedIn, directories
[ ] Name, address, phone match everywhere (local businesses)
## 5. Citation log (3 runs per engine)
| Question | Engine | Date | Run | Cited? | Cited URL | Competitor cited |
|----------|--------|------|-----|--------|-----------|------------------|
| | ChatGPT | | 1 | | | |
| | Perplexity | | 1 | | | |
| | Google (AI Overview shown? y/n) | | 1 | | | |
## First fix
Section with first unchecked box:
Change to make:
Re-test date:
If you want more plain files to start from, the free Grok Bot AEO files and the AI Search and SEO Growth Suite kit are both built as plain files you can edit.
What I would leave off a first AEO audit
A first AEO audit should skip anything with weak evidence or high effort. I would leave out these three items until the core checks pass:
- An llms.txt file. AI Labs Audit states that no measurable impact on AI visibility has been proven and calls it optional (AI Labs Audit). RevvGrowth still lists it as a check (RevvGrowth). It takes ten minutes, so add it later if you like. Do not count it as a fix.
- A FAQ block on every page. A FAQ helps where buyers really ask follow-up questions. Bolted onto every page, it repeats the body and dilutes the answer.
- Voice assistant testing. Stackmatix suggests querying Google Assistant, Alexa and Siri (Stackmatix). For most sites, ChatGPT, Perplexity and Google cover the questions that bring buyers, and those three are enough to start.
I run NorthSignal, where I work on search visibility and AI citations for owner-led businesses. A longer version of this process is in my AI search visibility audit.
Copy aeo-audit.md, fill it in for your single most important page today, and fix the first unchecked box in section 1.