The GEO Checklist: 30 Checks in the Order They Break
In our study of 149,912 AI citations, only 2.9% pointed to a brand's own site, yet most GEO checklists audit your own pages and add FAQ schema and llms.txt, which Google says its AI features don't need. AI visibility breaks in an order: access, the page, format, off-site sources, measurement. Here are 30 checks in that order, each with how to verify it.
A GEO checklist works best in the order AI visibility breaks. First, can each AI engine's fetcher reach your pages? Then, does the page state its answer, do you have the formats engines cite, do the third-party pages they already cite name you, and do you measure often enough to tell a real change from noise? Below are 30 checks in that order, each with why it matters and how to verify it.
The order matters more than the count. Google's own guide says optimizing for generative AI search is “still SEO”, so most of the work is not new. What is new sits at the two ends of the list: whether a bot gets your page at all, and whether the pages AI reads name you. In our study of 149,912 AI citations, only 2.9% pointed to the brand's own site.
What Most GEO Checklists Get Wrong
Most GEO checklists are an SEO audit with three additions: FAQ schema, an llms.txt file, and pages cut into short answer blocks. Google's guide to its AI features says none of the three is needed. It says Google Search ignores files like llms.txt, that structured data isn't required for generative AI search, and that there's no requirement to break content into tiny pieces. The FAQ rich result itself is gone: Google removed its documentation on 15 June 2026.
Meanwhile three layers get a line each, or nothing. Access stops at robots.txt, although a firewall or CDN can refuse a bot that robots.txt allows, and Cloudflare changed its AI crawler controls on 15 September 2026. Off-site sources get “earn mentions”, although 75.2% of the citations in our study went to corporate and competitor pages. And measurement gets “test your prompts monthly”, although answers change between runs, Google and Bing both added free AI reports in 2026, and GA4 now has an AI Assistant channel. This checklist gives each of those layers the checks it needs.
Five layers, in the order they break
A failure in an earlier layer makes the later ones moot: a page no engine can fetch has no answer to improve.
1. Access: Can Engines Reach the Page?
Everything later depends on this layer. OpenAI and Anthropic each run separate bots for search, for training and for fetches a person asks for, and only some of them decide whether you appear in answers. Some user fetchers, such as ChatGPT-User, may not follow robots.txt at all, so the firewall is where they get stopped. Cloudflare says less than 1% of its sites block search crawlers, so when a search bot is refused, ask whether anyone meant it.
Layer 1 of 5 · Checks 1-8
Access
Can each engine's fetcher reach the page?
Check 1: Write robots.txt rules by purpose, not by company
OpenAI and Anthropic each run separate bots for search, for training and for fetches a person asks for. Blocking a search bot takes you out of the answers: OpenAI's OAI-SearchBot decides whether a site can appear in ChatGPT search. Blocking a training-only bot such as GPTBot or ClaudeBot is a licensing decision. Google says Google-Extended does not affect inclusion in Google Search, but it also decides whether Gemini Apps and Vertex AI can ground their answers in your pages.
How to check
- Read robots.txt for OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot and Bingbot.
- Confirm each one can reach the pages you want cited.
- Repeat on every subdomain, since each serves its own robots.txt.
Check 2: Don't block user-initiated fetchers at the edge
ChatGPT-User, Claude-User and Perplexity-User fetch a page when someone asks about it. OpenAI says robots.txt rules may not apply to ChatGPT-User, and Perplexity says its user fetcher generally ignores them, so for those two a firewall or CDN rule is what stops the visit, whether or not anyone meant it to. Anthropic says Claude-User follows robots.txt.
How to check
- In your CDN or server logs, find the requests from these three agents.
- Confirm each one gets a 200, not a 403 or a challenge page.
Check 3: Fetch your key pages as each bot and as a browser, and compare
An Allow line in robots.txt means nothing if the server answers the bot with a 403 or a challenge page. Fetching the same URL as a bot and as a browser tells a block aimed at AI apart from a page that fails for everyone.
How to check
- Fetch the homepage and three pages you want cited with each bot's User-Agent, then with a browser's.
- Compare the status codes and word counts.
- Confirm with your logs, since a borrowed User-Agent is not the real bot (check 8).
Check 4: Review your CDN's AI crawler settings since 15 September 2026
Cloudflare now offers new domains a preset. On a site that shows ads, training is set to Disallow AI Training and agents are blocked on pages with ads, while search crawlers stay allowed. On existing domains, a training setting of Block was moved to Disallow AI Training, which keeps search crawlers allowed. A training Block chosen now also stops Googlebot, Bingbot and Applebot, search included, so check what each setting does before you change it.
How to check
- Open your CDN's AI crawler controls.
- Confirm search crawlers are allowed and agents are set the way you intend.
- Check that the training setting reflects a decision someone made.
Check 5: Put the core content in the initial HTML
Most AI companies don't say whether their fetchers run JavaScript. A page whose answer appears only after scripts run may reach a bot as an empty shell.
How to check
- Compare the word count of the raw HTML with the rendered page.
- If the answer is missing from the raw HTML, render it on the server.
Check 6: Stay indexed and snippet-eligible in Google
Google says a page must be indexed and eligible to show with a snippet to appear as a supporting link in AI Overviews or AI Mode, and nosnippet stops a page's content being used as a direct input for them. Since 31 August 2026 every site also has a Search Console control that opts it out of AI Overviews, AI Mode and Discover's AI features.
How to check
- Run URL Inspection on your key pages.
- Search your templates for nosnippet, data-nosnippet and max-snippet.
- Confirm the Search generative AI control in Search Console is set the way you intend.
Check 7: Get indexed in Bing, and send changes with IndexNow
Bing's AI Performance report counts your citations in Copilot and AI summaries in Bing. Bing says accurate, up-to-date content matters for inclusion and citation in AI-generated answers, and that IndexNow helps keep pages fresh.
How to check
- Inspect your key URLs in Bing Webmaster Tools.
- Confirm your CMS or CDN sends an IndexNow ping when a page changes.
Source: Bing Webmaster Blog
Check 8: See AI crawlers in your logs, grouped by purpose
Logs are the only proof that the real bots reach you, rather than something using their names. Grouping visits into search, training and user fetches shows which kind of access you actually have.
How to check
- Pull 30 days of server or CDN logs.
- Keep the visits whose IP addresses fall in each company's published ranges.
- Count them by bot and by page.
2. The Page: Can the Engine Lift the Answer?
Once a bot can fetch the page, it needs an answer it can quote and trust. Nothing here is a trick: Google's guide asks for the same indexable, clear, accurate pages it always has. The checks are the places where pages most often fall short of that.
Layer 2 of 5 · Checks 9-13
The page
Can the engine read the answer and lift it?
Check 9: Answer the question in the first lines, in plain prose
An engine can only quote an answer it can find. Google says there is no requirement to break content into tiny pieces for AI, so this is about clear writing, not formatting tricks.
How to check
- Read the first 100 words of each target page.
- If they don't answer the page's question, rewrite the opening.
Check 10: Keep one page per question, with one set of facts
Two pages giving different prices, limits or dates for the same thing leave an engine to pick one, and it may pick the old one.
How to check
- Search your own site for the claims that matter most, and fix the pages that disagree.
- Check the canonical tags on near-duplicates.
Check 11: Back specific claims with named sources
A number with its source and date can be checked and quoted. A vague claim gives an engine nothing to lift.
How to check
- Count the unsourced numbers on each target page.
- Source each one, or cut it.
Check 12: Date pages honestly
A visible date and dateModified that change only when the content changes tell an engine how current the page is. A date bumped without an edit is a claim the page can't back up.
How to check
- For each target page, compare the visible date, the dateModified in the markup and the CMS revision history.
- Fix any date that changed without a real edit.
Check 13: Make structured data match the visible page
Google says structured data isn't required for generative AI search and there is no special schema.org markup to add. Markup still serves ordinary rich results, and it must describe only what the page shows. The FAQ rich result is gone: Google removed its documentation on 15 June 2026.
How to check
- Run each page template through the Rich Results Test.
- Check that every marked-up fact is visible on the page.
3. Format and Entity: Are You the Kind of Page AI Cites?
A clear page can still be the wrong kind of page. In our study, ranked lists were the most-cited content format, with 35.7% of the citations that went to content pages. The engine also has to know who you are, which depends on facts it finds about you elsewhere.
Layer 3 of 5 · Checks 14-17
Format and entity
Are you the kind of page AI cites?
Check 14: Have the page formats AI cites for your questions
In our study of 149,912 AI citations, ranked lists took 35.7% of the citations that went to content pages, how-to guides 9.7% and comparisons 4.5%. The format that answers a question shapes whether a page gets cited for it.
How to check
- For your core questions, note the formats of the pages the engines cite.
- Check you have a page in each of those formats.
Check 15: Publish an honest comparison or alternatives page
Comparison questions get answered from pages that compare. A page that names the real alternatives, states its criteria and carries a date gives an engine something fair to quote.
How to check
- Check the page exists and names the options a buyer would actually weigh.
- Confirm it states its criteria and its date.
Check 16: Keep your entity facts the same everywhere
Engines piece a company together from its site, LinkedIn, directories and knowledge bases. A different name, description or founding date in each place blurs who you are.
How to check
- Put your name, one-line description, location and key facts from each profile side by side.
- Fix the ones that differ.
Check 17: Own one asset that others cite
Original data, a benchmark or a free tool gives other sites a reason to name you, and other sites are where most citations come from (layer 4).
How to check
Name the URL and the number or result someone would quote from it. If you can't, you don't have one yet.
4. Off-Site: Do the Pages AI Cites Name You?
This layer is where most citations live. With 2.9% of citations going to the brand's own site, the rest of your visibility is decided on pages you don't own. The checks start from the pages the engines already cite for your questions, because a generic list of popular sites tells you little about your category.
Layer 4 of 5 · Checks 18-23
Off-site sources
Do the pages AI already cites name you?
Check 18: Build your citation set, engine by engine
Only 2.9% of the 149,912 citations in our study pointed to the brand's own site, and 75.2% went to corporate and competitor pages. Most of the pages that decide whether you are named belong to someone else.
How to check
- Ask your core questions on each engine several times, and log every cited URL.
- Use that list, not a generic list of popular sites, as your off-site plan.
Check 19: List the cited pages whose answers name rivals and not you
A page that feeds answers naming your competitors, and not you, is the most direct gap to close.
How to check
- For each cited URL, note whether the answers citing it named you, a rival or neither.
- Judge by the answer, not by whether the page itself mentions you.
Check 20: Claim and update the review and directory profiles in your citation set
Review and directory sites show up in answers for some categories and some engines, not all. Your own citation set tells you which ones count for you.
How to check
- Match your citation set against review and directory sites.
- Claim, complete and date each profile that appears.
Check 21: Take part honestly in the threads AI cites
Where forum threads appear in your citation set, a useful answer from someone who knows the product, and says who they work for, can be quoted. Undisclosed promotion gets removed, and can be quoted against you.
How to check
- List the threads and communities in your citation set.
- Check that every reply from your team says who they work for.
Check 22: Cover your buyer questions on video where video gets cited
Video is one of the source types AI engines cite: YouTube and other video took 4.2% of the citations in our study. Where videos appear in your citation set, a video that answers the question can be cited beside the text pages.
How to check
- Check your citation set for video links.
- Search the video platform for your top questions, and see whose videos answer them.
Source: Ranqo's GEO study
Check 23: Earn mentions, never buy or fake them
Google says seeking inauthentic mentions across the web isn't as helpful as it might seem. Paid placements dressed as editorial and planted reviews also fail the moment a reader or moderator spots them.
How to check
- Ask your PR and link vendors what they place and where.
- Drop anything you couldn't disclose.
5. Measure, Fix, Re-Measure
AI answers change from run to run, so one check is an anecdote. In our study, 77.5% of brand, question and engine combinations tracked over at least three runs named the brand every time or never; every real change happens in the remaining 22.5%. Google, Bing and GA4 now each report part of the picture for free. None of them shows which brands AI names in your place, and only Bing lists the queries where other sites are cited and yours is not.
Layer 5 of 5 · Checks 24-30
Measure, fix, re-measure
Can you tell a real change from noise?
Check 24: Freeze a set of unbranded buyer questions
In our study, on a brand's first tracking run, each engine named the brand in 94% to 100% of answers to questions that named it, so branded questions tell you little. A question set that keeps changing makes every trend meaningless.
How to check
- Keep a versioned list of the questions you track, mostly unbranded.
- Change it only with a dated note.
Check 25: Ask each question more than once before you read a change
In our study, 77.5% of brand, question and engine combinations tracked over at least three runs named the brand every time or never. Change happens in the rest, which is exactly where a single check misleads.
How to check
Don't call a win or a loss until at least 2 runs agree.
Check 26: Export a baseline from Search Console's generative AI report
Since 31 August 2026 every site has a Search Console report of impressions in AI Overviews, AI Mode and Discover's AI features, by page, country, device and date. It shows impressions of your own links, not the questions behind them or who else was cited.
How to check
- Export the report for your key pages before you change anything.
- Date the export.
Source: Google Search Central Blog
Check 27: Read Bing's AI Performance report and its opportunity queries
Bing reports your citations, cited pages and the grounding queries its AI used, across Copilot and AI summaries in Bing. Since 13 October 2026 its Topics Portfolio also lists queries where other sites are cited and yours is not.
How to check
- Compare the grounding and opportunity queries with the questions you track.
- Add the ones your buyers would ask.
Check 28: Use GA4's AI Assistant channel, and know what it leaves out
GA4's default channel group now files visits from assistants such as ChatGPT, Gemini, Copilot and Grok under AI Assistant. Clicks from Google's AI Overviews and AI Mode are not in it: they count as Organic Search.
How to check
- Open Traffic acquisition by default channel group.
- Check the AI Assistant row against your referral logs.
Check 29: Re-measure each fix in runs, not days
A week with no new answers tells you nothing, however long it feels. Counting the measurement runs since a fix makes the before and after comparable.
How to check
- Log the date of each fix.
- Compare the questions it targeted across the following runs with the baseline.
- Claim a lift only when it holds.
Check 30: Give the work an owner and a cadence
Crawler settings, cited pages and the engines themselves all change between audits. Without an owner, this list gets done once and goes stale.
How to check
- Name the owner.
- Measure weekly or every two weeks.
- Book a full pass through this checklist every quarter.
What We Left Off, and Why
These items appear on most GEO checklists. Each is left off here for a stated reason:
- FAQ schema as a citation lever. Google's FAQ rich result is gone, and Google says there is no special schema.org markup to add for generative AI search. Mark up FAQs only where a real FAQ is on the page. (Search Central changelog; Google's AI optimization guide)
- llms.txt as a requirement. About 2.1% of sites had one in HTTP Archive's 2025 crawl, and Google says Google Search ignores these files. It does no harm, so publish one if another tool you use asks for it. (HTTP Archive Web Almanac 2025; Google's AI optimization guide)
- Rewriting pages into 40-word answer blocks. Google says there is no requirement to break content into tiny pieces for AI. Answer first (check 9) and write for the reader. (Google's AI optimization guide)
- Allowing every AI bot so AI can see you. Training-only bots such as GPTBot and ClaudeBot don't decide whether you appear in answers; search bots and user fetchers do (checks 1 and 2). Allowing one is a licensing decision. Google-Extended is the exception: it also decides whether Gemini Apps can ground answers in your pages. (Google's crawler docs)
- Generic technical SEO. HTTPS, page speed and mobile layout matter for search in general, and an SEO audit already covers them. They are left out here to keep the list about what decides AI answers.
Copy the Checklist
The 30 checks as plain text, in order, to paste into a doc or a ticket:
GEO CHECKLIST: 30 CHECKS IN THE ORDER THEY BREAK
1. ACCESS: Can each engine's fetcher reach the page?
[ ] 1. Write robots.txt rules by purpose, not by company
[ ] 2. Don't block user-initiated fetchers at the edge
[ ] 3. Fetch your key pages as each bot and as a browser, and compare
[ ] 4. Review your CDN's AI crawler settings since 15 September 2026
[ ] 5. Put the core content in the initial HTML
[ ] 6. Stay indexed and snippet-eligible in Google
[ ] 7. Get indexed in Bing, and send changes with IndexNow
[ ] 8. See AI crawlers in your logs, grouped by purpose
2. THE PAGE: Can the engine read the answer and lift it?
[ ] 9. Answer the question in the first lines, in plain prose
[ ] 10. Keep one page per question, with one set of facts
[ ] 11. Back specific claims with named sources
[ ] 12. Date pages honestly
[ ] 13. Make structured data match the visible page
3. FORMAT AND ENTITY: Are you the kind of page AI cites?
[ ] 14. Have the page formats AI cites for your questions
[ ] 15. Publish an honest comparison or alternatives page
[ ] 16. Keep your entity facts the same everywhere
[ ] 17. Own one asset that others cite
4. OFF-SITE SOURCES: Do the pages AI already cites name you?
[ ] 18. Build your citation set, engine by engine
[ ] 19. List the cited pages whose answers name rivals and not you
[ ] 20. Claim and update the review and directory profiles in your citation set
[ ] 21. Take part honestly in the threads AI cites
[ ] 22. Cover your buyer questions on video where video gets cited
[ ] 23. Earn mentions, never buy or fake them
5. MEASURE, FIX, RE-MEASURE: Can you tell a real change from noise?
[ ] 24. Freeze a set of unbranded buyer questions
[ ] 25. Ask each question more than once before you read a change
[ ] 26. Export a baseline from Search Console's generative AI report
[ ] 27. Read Bing's AI Performance report and its opportunity queries
[ ] 28. Use GA4's AI Assistant channel, and know what it leaves out
[ ] 29. Re-measure each fix in runs, not days
[ ] 30. Give the work an owner and a cadenceHow Ranqo Fits
Ranqo runs this checklist as a loop, layer by layer:
- Access. Site Access fetches your site as 16 crawlers, from GPTBot to Googlebot, and as a browser, then reads your robots.txt, llms.txt and sitemap, every week and whenever you ask (checks 1 to 4).
- The page. Page Audit scores a page on six dimensions, crawlability, content quality, page speed, AI readiness, citation potential and authority, and lists what to fix first (checks 5 and 9 to 12).
- Format and off-site. The Sources page shows the sites and pages each engine cites for your questions, labelled by source type and page format, and the Citation Pool shows who AI names in the answers that cite each site (checks 14 and 18 to 20). Outreach ranks the Reddit communities, threads and publications to approach and drafts the comment, post or pitch; it never posts for you (checks 21 and 23).
- Measurement. Your buyer questions are asked every week on each engine you track, and the Action Center counts each fix's lift in runs, showing a lift only when it clears the noise band (checks 24, 25 and 29). The Traffic page shows AI referral sessions from GA4 beside Site Tracking's server-side record of AI bot visits (checks 8 and 28).
Start with layer one, free
Check whether AI crawlers can reach a page with the free AI Crawler Inspector, then run the rest of the list in Ranqo: try it free for 7 days.
Start your free trialWritten by
Nisha Kumari
Nisha Kumari is Co-Founder at Ranqo, where she leads growth strategy and client acquisition. With a background in digital marketing and financial management, she specializes in SEO, Generative Engine Optimization, and helping brands build visibility across AI platforms.
Share this article
Related articles
How to Tell If a GEO Change Worked: Count the Answers, Not the Score
Winning two of 20 tracked questions on every engine moves an overall AI visibility score 10 points. Between one run and the next, it moves 14.5 points or more about one time in twenty. So the score cannot tell you whether one GEO change worked; the answers it targeted can. Here is a six-step protocol, the two-run rule behind it, and a log to copy.
Is Your Site Blocking AI Crawlers? A Ten-Point Check After Cloudflare's September Change
Most sites are not blocking AI crawlers, and most that are did not decide to. Refusing the crawler that trains a model costs nothing in AI answers; refusing the one that fetches pages at answer time is what removes you. What Cloudflare changed in September, which bot is which, and a ten-point check to run on your own site.
Off-Site GEO: A Playbook for the 97% of AI Citations You Don't Control
Only 2.9% of AI citations point to your own domain, so 97% of your AI visibility lives on pages you don't control. The usual advice, get mentioned everywhere, is a checklist, not a plan: the six off-site surfaces differ 20x in payoff and 10x in effort. Off-site GEO sequences them by leverage, share times winnability times durability. Here's the playbook, with data from 149,912 citations.