AI SEO Checklist: Access, Accuracy and Measurement

99Visibility Research · Updated October 21, 2026 · 9 min read

Short answer

An AI SEO checklist has three parts. Access: make sure each AI search crawler can fetch your pages through robots.txt, your CDN and firewall, server-rendered HTML and snippet controls. Accuracy: state your key facts once, consistently, on pages crawlers can read. Measurement: check each engine separately and note whether the answer used live search. Access fixes remove blocks. Nothing on the list guarantees a citation.

Key takeaways

Access items have the strongest evidence because they remove blocks that stop a crawler from reading your pages at all.
Robots.txt rules are per bot: GPTBot controls training, while OAI-SearchBot controls ChatGPT search eligibility.
Accuracy work means stating each key fact once, the same way everywhere, in HTML that loads before JavaScript runs.
Measure each AI engine separately and record whether the answer used live retrieval, because one blended score hides what changed.
Schema, llms.txt, E-E-A-T scores and chunking do not belong on an evidence-based checklist as citation levers.

Why the checklist has three groups

AI answers pass through four stages, and each stage has a different owner.

Parametric: what the model knows with no web access, frozen at a training cutoff. Nobody changes this quarter's answers here.
Access: whether a crawler may fetch and store your page. You and your CDN control this.
Retrieval: whether your page enters the model's context for a given query. The engine decides, per query.
Generation: which retrieved pages the model uses and cites. The model decides.

Most published GEO tactics were tested at the generation stage, on pages already inside the model's context. The July 2026 critical survey of 45 GEO studies (arXiv 2607.14035) concluded that no reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability.

That shapes the checklist. Access gets the most items because the mechanisms are documented by the engines themselves. Accuracy covers the problem brands actually report: AI platforms getting specific facts wrong. Measurement exists so you can see whether anything changed, engine by engine. Each item below says what it can achieve and what it cannot.

Access checklist: robots.txt, CDN, rendering, snippets

1. Robots.txt, bot by bot

Check the rules for each retrieval bot: OAI-SearchBot (ChatGPT search), PerplexityBot, Claude-SearchBot, Meta-WebIndexer, Googlebot and bingbot. Treat training bots (GPTBot, ClaudeBot, Google-Extended) as a separate decision. OpenAI's crawler documentation states that each setting is independent, so you can block training and still allow search:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Can: make you eligible for that engine's search index.
Cannot: get you retrieved or cited. User-triggered fetchers such as ChatGPT-User and Perplexity-User may not follow robots.txt. There is no robots.txt token for Google AI Overviews or AI Mode; Googlebot rules are the control there.

2. CDN and firewall reachability

OpenAI's help centre says eligibility needs both things: allow OAI-SearchBot, and confirm your host or CDN allows traffic from OpenAI's published searchbot IP addresses. A firewall can return 403 to a bot your robots.txt allows. As of September 15, 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers, while Search stays allowed. Check your zone settings even if you changed nothing.

Can: remove a silent, total block.
Cannot: tell you how common this failure is. Nobody has measured that.

The AI crawler access audit walks through this test step by step, and the AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

3. Server-side rendering

Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files (11.50% and 23.84% of requests) but do not execute them, so they cannot read client-side rendered content. Googlebot does render. Fetch your key pages with JavaScript off and confirm prices, product names and headings are in the raw HTML.

Can: put your content into the candidate set for non-Google engines.
Cannot: make the engine more likely to select it.

4. Index and snippet eligibility

Google states that a page must be indexed and eligible to show with a snippet to appear as a supporting link in AI Overviews or AI Mode, with no additional technical requirements. Search the templates for noindex, nosnippet, max-snippet:0 and data-nosnippet. Bing documents that data-nosnippet content is excluded from snippets and AI summaries. In Search Console, check whether the generative AI inclusion control has been changed; inclusion is the default, and child properties can inherit an opt-out from the parent. Check Bing index status in Bing Webmaster Tools separately.

Can: find accidental suppression, often inherited from a template.
Cannot: add anything. No markup makes a section more likely to be used.

Accuracy checklist: consistent facts

Access decides whether an engine can read you. Accuracy decides whether what it reads is right.

List the facts that matter. Prices and plan names, product names, founding year, headquarters, integrations, certifications, ownership.
Write one canonical version of each. Then find every place it appears: pricing page, about page, footer, help docs, structured data, press pages. Fix every contradiction. If your own site says two prices, an engine has two to choose from.
Put each fact in a plain sentence on a fetchable page. A specific stated fact (a price, a number, a one-line definition) gives the engine a sentence it can quote. The 252,000-trial factorial study (arXiv 2605.25517) found explicit price information helped consistently, though on content already in context.
Check third-party pages. When engines cite other sites instead of yours, being named accurately on those sites is associated with appearing in the answer. Update directory listings, partner pages and review profiles that state old facts.
Record which engine says what. Note the wrong fact, the engine, and whether the answer showed citations.

Can: give retrieval-on answers a correct sentence to work from.
Cannot: change retrieval-off answers. Those come from training data with a fixed cutoff. OpenAI's model documentation, checked in August 2026, listed a knowledge cutoff of August 31, 2025 for ChatGPT's Instant path. A fact you changed after that date is unknown to that model until it is retrained.

If you are shipping a new product or price, the product launch checklist for AI accuracy covers timing. For the wider question of keeping your brand facts consistent across the web, see managing your brand as an entity.

Measurement checklist: per engine, per retrieval state

Score each engine on its own. Google says AI Overviews and AI Mode may use different models and techniques. In Otterly's 2026 experiment, Microsoft Copilot kept citing two sites after Google deindexed them, while Google's AI surfaces almost never did. A blended number would have hidden that.
Record retrieval state. If an answer shows no citations, it may come from training data. Retrieval is per query and discretionary on every platform.
Track both kinds of prompt. Branded prompts test accuracy; unbranded prompts test whether you get named at all. See branded vs unbranded prompts.
Use first-party reports for what they show. Bing Webmaster Tools' AI Performance report shows grounding queries, the phrases the AI actually used to retrieve content. Search Console's generative AI report shows impressions only, with no clicks or position.
Never record an "AI rank". No engine publishes position data to validate it against.

Can: show what each engine says, and whether a fix changed it.
Cannot: observe the real queries engines run. They rewrite prompts, so any prompt set is a sample.

For a quick snapshot, the free checker on this site shows whether AI engines name you for up to three buyer questions. It is a one-time check, not tracking. For a recurring routine, use the quarterly AI visibility review.

What is and is not in your control

Stage	In your control	Not in your control
Parametric	Nothing this quarter	Training data, cutoff dates, retraining schedule
Access	Robots.txt, CDN and firewall rules, rendering, snippet and index directives	Whether a bot honours your rules
Retrieval	Index eligibility in Google and Bing, topical fit of your pages	The queries the engine writes, which pages it pulls
Generation	Clear, consistent facts on readable pages	Which retrieved pages get cited, and how

The honest summary: you can remove every block and correct every fact, and the model still decides what to say. Work the left column fully, then measure.

A practical order of work

Robots.txt. Confirm each retrieval bot is allowed. Fix any blanket AI block that also removes search bots.
Network reachability. Request a page as each AI crawler and confirm a 200, not a 403 or challenge page. Check CDN bot settings.
Rendering. Compare raw HTML with the rendered page on your pricing, product and about pages.
Snippet and index controls. Remove accidental noindex, nosnippet and data-nosnippet. Check the Search Console AI control and Bing index status.
Facts. Write the canonical fact list and fix contradictions on your own site first, then third-party pages.
Baseline measurement. Record per-engine answers and retrieval state.
Re-check after changes. Repeat steps 1 to 4 after any redesign, CDN change or migration. The site migration checklist covers that case. For engine-specific detail, see the ChatGPT visibility checklist.

Steps 1 to 4 are mostly a day of work for one technical person. Step 5 needs whoever owns pricing and product copy.

Common mistakes

Allowing GPTBot to get into ChatGPT search. GPTBot governs training only. OAI-SearchBot governs search.
Treating schema as a quick win. Google says structured data isn't required for generative AI search. Ahrefs' matched-control study found AI Overview citations fell 4.6% after pages added schema. Existing markup does no harm; it just isn't a lever for AI answers.
Publishing llms.txt as a fix. Google says Search ignores it, and no other engine documents reading it.
Chasing an E-E-A-T score or domain authority. Google says E-E-A-T isn't a specific ranking factor. Domain Authority is a Moz metric no engine computes.
Rewriting or chunking for AI. Google's AI optimization guide says you don't need to write in a specific way for generative AI search, and there is no ideal page length.
Assuming strong SEO covers it. Ahrefs found 37.9% of URLs cited in AI Overviews also appeared in the top 10 (March 2026). Rank helps but does not settle it. See whether strong SEO means AI visibility.
Hidden text or prompt injection. Content written to manipulate AI crawlers exists, and research shows it can sway agents once retrieved. It is illegitimate, it breaches search spam policies, and it puts your brand at risk. Leave it off the list.

Frequently asked questions

Does schema markup belong on an AI SEO checklist?

Not as a citation lever. Google says structured data isn't required for generative AI search, and Ahrefs' matched-control study found AI Overview citations fell 4.6% after pages added schema. That study only covered pages already cited heavily, so whether schema helps a new page enter the answer is untested. Keep existing markup if it serves other purposes, since Google says it does no harm, but don't expect it to get you cited.

Should I block GPTBot?

That is a training decision, not a visibility decision. Blocking GPTBot tells OpenAI not to use your content for training foundation models. It does not remove you from ChatGPT search, which OAI-SearchBot controls. OpenAI documents the settings as independent. Nobody has measured whether blocking training changes how models recall your brand later, so treat it as a policy choice about your content.

How often should I rerun the checklist?

Rerun the access checks after any redesign, framework change, CDN or firewall change, or domain migration, because each can silently block crawlers. Otherwise a quarterly pass is a reasonable rhythm for access, facts and measurement together. Also re-check whenever a vendor changes defaults, as Cloudflare did for AI crawlers on September 15, 2026.

Can I fix what ChatGPT says when it doesn't search the web?

Not directly. Answers without live retrieval come from training data frozen at a cutoff date. Changes you make now reach those answers only when a newer model is trained and released. What you can do is make sure your correct facts are on readable pages, so answers that do use live search have the right sentence available.

Is llms.txt worth adding?

It isn't a visibility fix. Google says Search ignores llms.txt and that adding it will neither help nor harm visibility. No other major engine documents reading it. Google also says it's fine to keep one for other services that use it. If you have spare time, spend it on the access checks instead, where the mechanisms are documented.

Sources

This page is AI-assisted. It was written from a verified evidence base and checked against our editorial policy. Engine behaviour changes often; facts are as of October 21, 2026.