CHECKLIST · 7 MIN READ

AEO and GEO checklist: 25 checks for AI search (2026)

By the Seamwire team · Updated 20 October 2026

THE SHORT ANSWER

An AEO and GEO checklist covers five things: AI search crawlers can reach your pages, the pages work without JavaScript and allow snippets, each service and question has its own answer-first page, other trusted sites describe you consistently, and you measure how often engines name you. Most sites that are missing from AI answers fail one of the first two, often without knowing. Work through the 25 checks below in order; the early ones are quick and the later ones take months.

Crawler access: which bots should you allow?

Each AI company runs separate bots for search, for fetching a page when a user asks, and for collecting training data. Blocking a training bot does not remove you from that company's answers; blocking a search bot does. Decide each one on purpose.

1. Allow the search bots in robots.txt. OAI-SearchBot: OpenAI says sites that opt out are not shown in ChatGPT search answers. PerplexityBot: surfaces and links sites in Perplexity search, and Perplexity says it is not used to train foundation models. Claude-SearchBot: Anthropic says blocking it may reduce your visibility in Claude's search results. Bingbot: Microsoft says Copilot's web search sends its queries to Bing. Googlebot: AI Overviews and AI Mode use Google's normal index.
2. Do not block the user-initiated fetchers: ChatGPT-User, Perplexity-User and Claude-User visit a page when someone asks about it. OpenAI and Perplexity say robots.txt may not apply to these, so a firewall rule is what usually stops them. Anthropic says disabling Claude-User stops Claude retrieving your content for a user's question.
3. Decide on the training controls deliberately: GPTBot (OpenAI), ClaudeBot (Anthropic), Google-Extended and Applebot-Extended. Google says Google-Extended does not affect inclusion or ranking in Google Search; it governs Gemini training and grounding outside Search. Apple says Applebot-Extended does not crawl and that pages which disallow it can still appear in Spotlight, Siri and Safari results. Blocking these is a choice about training, not about being found.
4. Remove leftovers: a site-wide “Disallow: /” copied from a staging site, or old rules naming bots that have since changed purpose.
5. Give every subdomain its own correct robots.txt. Rules on www do not cover docs, blog or app subdomains.

CDN, firewall and hosting: is something blocking bots before robots.txt is read?

A robots.txt that allows everything means little if the CDN or firewall answers the bot with a block or a challenge page.

6. Check Cloudflare's AI settings. Cloudflare has blocked AI training crawlers by default on new domains since July 2025. Its documentation says that from 15 September 2026 the old “Block AI bots” toggle gives way to AI bot policies in three groups (Search, Agent and Training), that crawlers mixing search and training are blocked under any setting that blocks training, and that new domains block Training and Agent bots on pages showing ads, with Search allowed. Agent covers the user-initiated fetchers in check 2.
7. Check other layers too: bot-management rules on other CDNs, hosting-level firewalls, WordPress security plugins and rate limits. A challenge page is a blank page to a crawler; Anthropic states its bots do not bypass CAPTCHAs.
8. Read your server or CDN logs for these user agents. You want 200 responses on the pages that matter, not 403, 429 or challenge responses.

Can the engines read and quote the page?

9. Serve the important copy in the HTML. Vercel's 2024 analysis found that none of the major AI crawlers it studied rendered JavaScript; Gemini (through Googlebot) and Applebot were the exceptions. If your text appears only after scripts run, most AI bots see an empty page.
10. Keep key facts in text: what you do, for whom, where, and how. Not only in images, sliders or PDFs. Google lists this among its best practices for AI features.
11. Check snippet controls. Google says nosnippet, data-nosnippet, max-snippet and noindex limit what its AI features can show, and Bing says data-nosnippet also keeps text out of its AI answers and Copilot. Use them on purpose, never by template default on pages you want quoted.
12. Use noindex only where you mean it. OpenAI's FAQ notes that a page it learns about elsewhere can still appear as a bare link unless it carries noindex.
13. Keep an accurate XML sitemap, and notify Bing of changes with IndexNow, which Bing recommends so its AI experiences reference the current version of a page.

Pages and passages: does each question have one clear answer?

14. One page per service, per industry and per common buyer question. A single “Services” page listing everything gives an engine nothing specific to quote.
15. Open each page with the answer: two to four sentences that make sense on their own, naming who you are and what you do.
16. Use questions as headings, in the buyer's words: “How long does an ERP integration take?” rather than “Our timelines”.
17. Add specifics an engine can lift: figures with named sources, comparisons, steps and tables. Bing advises clear headings, tables, FAQ sections and supporting evidence, and the 2024 GEO research paper found that adding statistics, quotations and citations raised visibility in generative engines.
18. Show a visible date and update it when the page changes. Keep any structured data consistent with the visible text, as Google requires.

Entity and reputation: does the rest of the web agree about you?

Engines lean heavily on what others write. Muck Rack's May 2026 analysis of more than 25 million links cited by ChatGPT, Claude and Gemini found about 84% came from earned media such as journalism, research and reference sites, against about 0.3% from paid or advertorial content.

19. Use the same company name, one-line description, address and phone number on your site, LinkedIn, Google Business Profile, Bing Places and every directory listing. Bing advises keeping entities consistent and local details current in Bing Places.
20. Add Organization structured data with sameAs links to those profiles, for clarity about who you are rather than as a ranking lever.
21. Collect real reviews from real clients on the directories your buyers use, such as Google, Clutch or G2. Never write or buy them.
22. Earn mentions where your buyers' questions are answered: independent “best of” lists, trade press, expert comment and the communities your buyers read, on their rules.

Measurement: how will you know it worked?

23. Track a fixed set of 20 to 50 buyer questions across the engines your buyers use, many runs each, and report your share of answers with a margin of error. Single runs vary too much to mean anything.
24. Track AI referral traffic: GA4's AI Assistant channel (added May 2026), a custom channel group for sources it misses, and utm_source=chatgpt.com, which ChatGPT adds to its links. Google AI Overviews and AI Mode clicks stay in Organic Search.
25. Tie it to leads: pass source and landing page into your CRM with each enquiry, and check Bing Webmaster Tools' AI Performance report for Copilot citations and grounding queries.

How Seamwire works through this list

We start with a free AI visibility check: ten real buyer questions put to four engines, plus a look at whether AI bots can read your site at all. If checks 1 to 13 are fine and there is no gap worth closing, we say so. If not, an audit covers the whole list with repeated runs, and a programme makes the technical and content fixes and reports the results monthly.