GEO, AEO and AI Visibility
GEO audit checklist: what to check, and what to fix first
A GEO audit checks seven areas: AI crawler access, indexing and snippets, brand facts, pages that answer buyers’ questions, evidence worth quoting, mentions on other sites, and a baseline on a fixed set of prompts. Every failed check gets one of four priority labels, so the audit ends as an ordered plan, not a list of problems.
By QueryRev ResearchPublished Updated9 min read
Key takeaways
- A GEO audit checks seven areas: crawler access, indexing and snippets, brand facts, pages that answer the question, evidence worth quoting, mentions on other sites and a prompt baseline.
- Run the checks in order, retrievable first and worth citing second: a failure early in the list can make later checks meaningless.
- Test what each crawler actually receives, in the server logs and at the CDN or firewall, not what the robots.txt file seems to say.
- Order every failed check by how serious it is, how close it sits to a sale, how much work the fix takes and how surely you can measure the change.
- An audit ends with an order of work, not a list of problems: each Fix first finding gets an owner and a place in the next two weeks.
How to use this checklist
Run the checks in order. They follow the two halves of generative engine optimization: first be retrievable, then be worth citing, and finally know where you stand. A failure early in the list can make later checks meaningless: a page that robots.txt keeps from OAI-SearchBot will not be shown in ChatGPT search answers, however well it is written[1].
Each check below has a test and a label for when it fails. Order the fixes by four things: how serious the problem is, how close it sits to a sale, how much work the fix takes and how surely you can measure the change.
Fix first
- What it means
- It blocks crawlers or customers today
- Example
- robots.txt shuts out an AI search crawler you want
Fix next
- What it means
- Worth doing as soon as the first fixes are in
- Example
- Pricing questions answered only deep in a long page
Look closer
- What it means
- Needs a closer look before anyone decides
- Example
- Citations fell on one engine and the cause is unclear
Keep an eye on
- What it means
- No action yet; watch it
- Example
- A new crawler appears in the logs
| Label | What it means | Example |
|---|---|---|
| Fix first | It blocks crawlers or customers today | robots.txt shuts out an AI search crawler you want |
| Fix next | Worth doing as soon as the first fixes are in | Pricing questions answered only deep in a long page |
| Look closer | Needs a closer look before anyone decides | Citations fell on one engine and the cause is unclear |
| Keep an eye on | No action yet; watch it | A new crawler appears in the logs |
Area 1: crawler access
Test what each crawler actually receives, not what the robots.txt file seems to say. A crawler names itself by its user agent, such as OAI-SearchBot or PerplexityBot, and the file, the firewall and the CDN can each treat it differently. Most access failures sit in a layer nobody on the content team looks at.
Search crawlers allowed
Every host covered
- How to test it
- Fetch /robots.txt on each subdomain and protocol you serve: each is read from the top of its own scheme and host[3]
- If it fails
- Fix first
robots.txt answers
- How to test it
- Check its status code: a server error on the file tells crawlers to assume the whole site is disallowed[3]
- If it fails
- Fix first
CDN and firewall
Logs agree
- How to test it
- Search the server logs for each crawler and the status it got: a 200 means it was let in, a 403 that it was refused
- If it fails
- Fix first
Verified by IP
Training rules chosen
- How to test it
- Check that rules for GPTBot and other training crawlers are a deliberate choice; OpenAI keeps each setting independent[1]
- If it fails
- Look closer
| Check | How to test it | If it fails |
|---|---|---|
| Search crawlers allowed | Read the robots.txt groups that apply to OAI-SearchBot and PerplexityBot; both vendors recommend allowing them for a site that wants to appear in their search results[1][2] | Fix first |
| Every host covered | Fetch /robots.txt on each subdomain and protocol you serve: each is read from the top of its own scheme and host[3] | Fix first |
| robots.txt answers | Check its status code: a server error on the file tells crawlers to assume the whole site is disallowed[3] | Fix first |
| CDN and firewall | Check the bot rules of the CDN, firewall and host; Google and Perplexity both name this layer[4][2] | Fix first |
| Logs agree | Search the server logs for each crawler and the status it got: a 200 means it was let in, a 403 that it was refused | Fix first |
| Verified by IP | Match the requests to the IP addresses each vendor publishes, so that a fake user agent is not taken for the real crawler[1][2] | Look closer |
| Training rules chosen | Check that rules for GPTBot and other training crawlers are a deliberate choice; OpenAI keeps each setting independent[1] | Look closer |
For context: in the October 2026 edition of QueryRev Research’s State of AI crawler access study, 14.2% of the 6,294 top sites read shut at least one AI crawler out of the whole site by name, and 2.7% shut out OAI-SearchBot[5]. Which crawlers to allow is a decision of its own. This area only confirms that what you decided is what crawlers get.
An llms.txt file is not on the list as a requirement, because Google Search ignores it[6]. If you keep one, check that it answers, and that it lists only pages the crawlers you serve may fetch. What the file is for explains when it is worth keeping.
Area 2: indexing and snippets
Google’s AI features, AI Overviews and AI Mode, can only link a page that its classic results could show. Take the pages behind your target prompts and test each one.
Indexed
- How to test it
- Run URL Inspection in Search Console; a page must be indexed and eligible for a snippet to be a supporting link in AI Overviews or AI Mode[4]
- If it fails
- Fix first
Snippet allowed
- How to test it
- Look for nosnippet, data-nosnippet, max-snippet and noindex; each limits what Google can show from the page[4]
- If it fails
- Fix first
AI features included
- How to test it
- Open Settings, then Search generative AI, in Search Console: the site must be included, which is the default[7]
- If it fails
- Fix first
Content in text
- How to test it
- Check that the key facts are in the HTML text, not only in images, tabs, expandable menus or PDFs[8]
- If it fails
- Fix next
Rendered as seen
- How to test it
- Compare the HTML Googlebot received, in URL Inspection, with the page a visitor sees[4]
- If it fails
- Fix next
One page per question
- How to test it
- Check that each target question has one page that answers it, not two near-copies competing for it
- If it fails
- Fix next
Bing has the page
- How to test it
- Check the same pages in Bing Webmaster Tools, where citations in Copilot and Bing’s AI summaries are reported[9]
- If it fails
- Fix next
| Check | How to test it | If it fails |
|---|---|---|
| Indexed | Run URL Inspection in Search Console; a page must be indexed and eligible for a snippet to be a supporting link in AI Overviews or AI Mode[4] | Fix first |
| Snippet allowed | Look for nosnippet, data-nosnippet, max-snippet and noindex; each limits what Google can show from the page[4] | Fix first |
| AI features included | Open Settings, then Search generative AI, in Search Console: the site must be included, which is the default[7] | Fix first |
| Content in text | Check that the key facts are in the HTML text, not only in images, tabs, expandable menus or PDFs[8] | Fix next |
| Rendered as seen | Compare the HTML Googlebot received, in URL Inspection, with the page a visitor sees[4] | Fix next |
| One page per question | Check that each target question has one page that answers it, not two near-copies competing for it | Fix next |
| Bing has the page | Check the same pages in Bing Webmaster Tools, where citations in Copilot and Bing’s AI summaries are reported[9] | Fix next |
Where near-copies exist, choose one page and point the others to its canonical URL, or merge them, so that an engine finds one clear answer instead of two partial ones.
Area 3: brand and entity facts
An answer can only describe your brand well if the facts agree wherever it reads about you: on your pages, in your structured data and on other sites.
Core facts written once
- How to test it
- Write what you sell, to whom, where, at what price and under which policies, then compare that wording with the home, about, pricing and product pages
- If it fails
- Fix next
Organization markup
- How to test it
- Look for Organization structured data on the home page; Google uses it to tell an organization apart from others[10]
- If it fails
- Fix next
sameAs is right
- How to test it
- Each sameAs URL is a page that unambiguously indicates who you are, such as an official website or a Wikidata entry[11]
- If it fails
- Fix next
Markup matches the page
- How to test it
- Check that the structured data states what the visible text states[4]
- If it fails
- Fix first
Listings current
- If it fails
- Fix next
Wrong facts in answers
- How to test it
- From the baseline, list every answer that says something false or out of date about you, and the page or source it came from
- If it fails
- Fix first
| Check | How to test it | If it fails |
|---|---|---|
| Core facts written once | Write what you sell, to whom, where, at what price and under which policies, then compare that wording with the home, about, pricing and product pages | Fix next |
| Organization markup | Look for Organization structured data on the home page; Google uses it to tell an organization apart from others[10] | Fix next |
| sameAs is right | Each sameAs URL is a page that unambiguously indicates who you are, such as an official website or a Wikidata entry[11] | Fix next |
| Markup matches the page | Check that the structured data states what the visible text states[4] | Fix first |
| Listings current | Check Google Business Profile and Merchant Center[6], and Bing Places for Business[9] | Fix next |
| Wrong facts in answers | From the baseline, list every answer that says something false or out of date about you, and the page or source it came from | Fix first |
Area 4: pages that answer the question
Read the pages behind your target prompts the way an answer engine does: passage by passage. Each page should match the search intent behind its prompt; a buying question needs a page that helps someone buy, not an article about the category.
Answer first
- How to test it
- The question sits in a heading and the first one or two sentences under it answer it
- If it fails
- Fix next
Self-contained
- How to test it
- Each answer makes sense lifted out alone, with its subject named and its condition stated[8]
- If it fails
- Fix next
Titles agree
- How to test it
- The title, H1 and description say what the page delivers, in the words a buyer uses[8]
- If it fails
- Keep an eye on
Next questions covered
- How to test it
- The page answers the obvious follow-up questions that one search can fan out into[6]
- If it fails
- Fix next
No page per variant
- How to test it
- No set of near-identical pages written for each variant of a question; Google treats that as scaled content abuse when it is done to manipulate results[6]
- If it fails
- Fix first
Structure helps readers
- How to test it
- Headings, lists and tables where the content has steps or comparisons; Google says no special chunking is needed[6]
- If it fails
- Keep an eye on
| Check | How to test it | If it fails |
|---|---|---|
| Answer first | The question sits in a heading and the first one or two sentences under it answer it | Fix next |
| Self-contained | Each answer makes sense lifted out alone, with its subject named and its condition stated[8] | Fix next |
| Titles agree | The title, H1 and description say what the page delivers, in the words a buyer uses[8] | Keep an eye on |
| Next questions covered | The page answers the obvious follow-up questions that one search can fan out into[6] | Fix next |
| No page per variant | No set of near-identical pages written for each variant of a question; Google treats that as scaled content abuse when it is done to manipulate results[6] | Fix first |
| Structure helps readers | Headings, lists and tables where the content has steps or comparisons; Google says no special chunking is needed[6] | Keep an eye on |
Area 5: evidence worth quoting
A quotable page says something specific that a reader can check. In the research that named GEO, the edits that raised a source’s visibility most were adding citations to credible sources, quotations and statistics[12].
Specific claims
- How to test it
- Vague words are replaced by measurable facts: sizes, prices, limits, dates[8]
- If it fails
- Fix next
Sources shown
- How to test it
- Facts that come from elsewhere carry their source, linked
- If it fails
- Fix next
First-hand content
- How to test it
- The page adds experience or data of your own, not a summary of what is already online[6]
- If it fails
- Fix next
Kept current
- How to test it
- Prices, plans and policies match today’s offer; Bing points to IndexNow to tell engines when a page changes[9]
- If it fails
- Fix next
Honest dates
- How to test it
- A new date on a page reflects a real change to it
- If it fails
- Look closer
| Check | How to test it | If it fails |
|---|---|---|
| Specific claims | Vague words are replaced by measurable facts: sizes, prices, limits, dates[8] | Fix next |
| Sources shown | Facts that come from elsewhere carry their source, linked | Fix next |
| First-hand content | The page adds experience or data of your own, not a summary of what is already online[6] | Fix next |
| Kept current | Prices, plans and policies match today’s offer; Bing points to IndexNow to tell engines when a page changes[9] | Fix next |
| Honest dates | A new date on a page reflects a real change to it | Look closer |
Area 6: mentions on other sites
AI answers draw on what others say about you as well as on your own pages. Google says its AI features can show what is said about products and services across the web, in blogs, videos and forum discussions[6].
Cited sources mapped
- How to test it
- For each prompt you lose, open the sources the answer cites and list those that cover your category
- If it fails
- Fix next
You are in them
- How to test it
- Check whether those pages mention you, and whether what they say is right
- If it fails
- Fix next
Mentions are real
- How to test it
- No paid or placed mentions; Google says inauthentic mentions are not as helpful as they might seem[6]
- If it fails
- Fix first
| Check | How to test it | If it fails |
|---|---|---|
| Cited sources mapped | For each prompt you lose, open the sources the answer cites and list those that cover your category | Fix next |
| You are in them | Check whether those pages mention you, and whether what they say is right | Fix next |
| Mentions are real | No paid or placed mentions; Google says inauthentic mentions are not as helpful as they might seem[6] | Fix first |
Area 7: the prompt baseline
The last area records where you stand before anything changes. The baseline is your AI search visibility on the first day: how many answers name you, and how many cite your pages.
Prompt set written
- How to test it
- A fixed list of buyer questions, split by stage, with brand prompts kept apart
- If it fails
- Fix first
Baseline run
- How to test it
- Each prompt asked in each tool you track, in a new chat, two or three times, every answer saved with its sources
- If it fails
- Fix first
Search Console AI report
- How to test it
- Export impressions in AI Overviews and AI Mode by page[13]
- If it fails
- Fix next
Bing AI Performance
- How to test it
- Export citations, cited pages and grounding queries[9]
- If it fails
- Fix next
GA4 AI Assistant
- How to test it
- Read sessions from AI assistants by landing page; the channel leaves out AI Overviews and AI Mode[14]
- If it fails
- Fix next
| Check | How to test it | If it fails |
|---|---|---|
| Prompt set written | A fixed list of buyer questions, split by stage, with brand prompts kept apart | Fix first |
| Baseline run | Each prompt asked in each tool you track, in a new chat, two or three times, every answer saved with its sources | Fix first |
| Search Console AI report | Export impressions in AI Overviews and AI Mode by page[13] | Fix next |
| Bing AI Performance | Export citations, cited pages and grounding queries[9] | Fix next |
| GA4 AI Assistant | Read sessions from AI assistants by landing page; the channel leaves out AI Overviews and AI Mode[14] | Fix next |
How to score the answers and report the results over time is set out in How to measure GEO results.
Turn the findings into a plan
An audit ends with an order of work, not a list of problems. Sort the failed checks by label, then, within each label, by how close they sit to a sale: a wrong price in an answer comes before a vague sentence in a blog post.
- Give each Fix first finding an owner and put it into the next two weeks.
- Schedule the Fix next findings in the weeks after, grouped by page, so that each page is edited once.
- Turn each Look closer finding into a question with a date to answer it.
- Add the Keep an eye on findings to the monthly check.
- Run the baseline again after the first fixes, and again at the end of the quarter, the same way.
The checks above are the GEO part of a site audit, not a replacement for one. Speed, internal links and content quality still decide whether a page is worth retrieving in the first place; the practical guide to generative engine optimization sets this audit inside the rest of the work.
Sources
Primary sources only, each read on the day given.
- 1.Overview of OpenAI Crawlers. OpenAI. Read .
- 2.Perplexity Crawlers. Perplexity. Read .
- 3.RFC 9309: Robots Exclusion Protocol. IETF, RFC Editor. Read .
- 4.AI features and your website. Google Search Central. Read .
- 5.State of AI crawler access, October 2026 edition. QueryRev Research. Read .
- 6.Optimizing your website for generative AI features on Google Search. Google Search Central. Read .
- 7.Search generative AI control. Search Console Help, Google. Read .
- 8.Optimizing Your Content for Inclusion in AI Search Answers. Microsoft Advertising Blog. Read .
- 9.Introducing AI Performance in Bing Webmaster Tools Public Preview. Bing Webmaster Blog, Microsoft. Read .
- 10.Organization structured data. Google Search Central. Read .
- 11.sameAs. Schema.org. Read .
- 12.GEO: Generative Engine Optimization. arXiv (accepted at KDD 2024). Read .
- 13.Generative AI performance report (Search). Search Console Help, Google. Read .
- 14.Default channel group. Analytics Help, Google. Read .
Cite this guide
A suggested citation. Its link goes to this page, whose address does not change.
QueryRev Research. “GEO audit checklist: what to check, and what to fix first.” QueryRev AI, 18 Oct 2026. https://queryrev.com/guides/geo-audit-checklist