Insights · GEO

Make the work discoverable.
Keep the claims defensible.

Romaan Sheikh, Cofounder · Published 18 August 2026 · Updated 15 October 2026

AI-search readiness starts with clear information that people and search systems can access. This checklist separates technical eligibility, useful content and observed visibility. Passing it cannot guarantee a recommendation or citation.

Start with the response actually served

Our August 2026 site review found a managed CDN policy rewriting robots.txt. The lesson is practical: hosting and security settings can change crawl instructions independently of the source file. Check the live response before diagnosing a discovery problem.

That policy mentioned agents with different purposes. GPTBot, ClaudeBot and Google-Extended must not be treated as interchangeable controls for AI Search. Separate search crawling, training and user-triggered visits using the provider’s current documentation.

1. Check access and indexability

Inspect status codes, redirects, robots instructions, canonical URLs and noindex directives on the important pages. Confirm that your CDN serves the intended policy. Preserve deliberate privacy and access controls.

2. Answer a real buyer question

Explain the service, intended customer, scope and price basis in ordinary language. A relevant case with dated results is more useful than a page of unsupported superlatives. Check that essential information is available without playing a film or completing an interaction.

3. Match structured data to visible facts

Use appropriate organisation, person and service markup to describe what the page actually says. Validate syntax and meaning. Do not invent ratings or add misleading types to obtain a richer search result.

4. Maintain reliable discovery files

Keep the sitemap aligned with public canonical URLs and genuine content updates. An optional llms.txt can summarise key pages and facts. It is not a prerequisite for Google’s AI Search features or a guaranteed route to citations.

5. Corroborate the organisation

Use consistent, verifiable names and relevant official profile links. Distinguish legal incorporation from a brand launch date, and real client work from fictional demonstrations. Consistency reduces ambiguity; it does not guarantee a citation.

6. Test delivery and accessibility

Measure loading, layout stability and interaction on representative devices. Keep important content readable and keyboard accessible. A laboratory performance score is not a field Core Web Vitals pass, and TBT is not real-user INP.

7. Measure visibility, not just crawler visits

Crawler requests show retrieval activity, not recommendations. Use the search reports your account actually provides and retain the dates, filters and limitations. Keep impressions, clicks, citations and enquiries separate. Results from one platform do not prove visibility on another.

Google now documents a Generative AI performance report and a Search generative AI inclusion setting in Search Console. Check the actual property and inherited setting. An unavailable report can mean insufficient impressions; it does not establish zero visibility. Retain provider, date, country, device and report limitations when comparing results.

What we can control

We can improve access, page quality, attribution, structure and the evidence behind a claim. Search providers decide what to index and show. There is no universal delay or guaranteed outcome after a technical change.

Sources and review

Reviewed 15 October 2026 against Google’s AI features guidance, OpenAI’s crawler documentation, Anthropic’s crawler guidance and Google’s sitemap guidance. Provider behaviour can change, so check the current source before changing policy.