AI SEO and GEO prepare website content for search and answers powered by AI. We start with three practical questions: can the system open the page, does it find useful information, and can it identify the publisher? Technical access is something we can test. Each system decides which sources to reference.

Sources last checked: . Originally published on 28 July 2025. This article has been revised to distinguish search, page retrieval and model training.

ChatGPT, Gemini and Claude: which bot does what?

Settings serve different purposes across providers. Before changing robots.txt or the firewall, record which uses the business wants to allow.

  • OpenAI: OAI-SearchBot supports ChatGPT search. GPTBot collects content that may be used for training. ChatGPT-User visits pages following a user action; robots.txt rules may not apply to these requests. Search and training controls are independent. OpenAI documentation.
  • Google and Gemini: Googlebot serves Google Search. Google-Extended is a robots.txt control for training and grounding in certain Gemini products. It has no separate HTTP user-agent and does not control Search inclusion or ranking. Google-Extended documentation.
  • Anthropic: Claude-SearchBot supports search, Claude-User retrieves pages at a user’s request, and ClaudeBot supports training. Anthropic crawler policy.

How do we check that a page can be read?

We follow the public URL through redirects, server responses, CDN and firewall rules, robots.txt and indexing instructions to the final content. For Google Search, the basic technical requirements include Googlebot access, an HTTP 200 response and indexable content.

  • The content is present in the HTML. The title, main text, services and links should be available without signing in. We prefer the substance of a page to be accessible without depending entirely on JavaScript execution.
  • The firewall allows the actual visit. An HTTP 200 response containing a challenge or an error message is insufficient. We compare the response body with the normal page.
  • The test has clear limits. Sending a bot name in the user-agent tests the server’s response to that name. Confirming a real crawler requires logs and origin verification using the provider’s published information.
  • Both languages are checked separately. We maintain correct canonical URLs, translations, internal links and an updated sitemap. Cache changes must not leave an old version in one language.

If your site operates in Greek and English, continue with our multilingual SEO and page mapping checks to keep content and internal links consistent within each version.

What about Google AI Overviews, AI Mode and llms.txt?

Google requires an indexed page, snippet eligibility and participation in its generative AI features. Include is the default; owners do not all need to activate a new option. Check the effective setting in Search Console under Settings → Search generative AI, following the participation control documentation. No special “AI schema” is required and Google Search ignores llms.txt. Eligibility does not guarantee appearance. Google’s official AI optimization guide.

What should the content and schema communicate?

Write for the visitor’s decision: what you offer, who it serves, how you work and what information is needed for the next step. Show specific projects, limitations and original examples instead of broad promises. This approach relates to Google’s guidance on helpful, reliable content.

Schema should describe what the reader can see: the actual author or organisation, accurate dates, the service and business details. Do not add invented offices, reviews or offers. Valid markup alone does not guarantee a special search appearance. General structured data guidelines.

For a local service, distinguish the business address from the area served. A team based in Patras can work with businesses in Crete or Cyprus. Project examples should explain their relevance and actual location.

In eCommerce, this starts with accurate descriptions, specifications, availability and purchase terms. See how to organise them on a useful eCommerce product page, without promises of guaranteed inclusion in AI answers.

What do we measure after the changes?

Keep a baseline and a dated change log. Monitor successful crawler visits, errors, indexing, visibility and enquiries reaching the business. Assess speed on real devices and connect it to essential visitor actions, such as reading and completing a form.

The Search Console Generative AI performance report records impressions in Google’s AI features, with page, country and device breakdowns. It may be unavailable when data is limited. Impressions are not visits or sales. In analytics, separately track identifiable referrals from AI services and the actions that follow.

Frequently asked questions

Is allowing GPTBot enough to appear in ChatGPT?

No. GPTBot relates to training. For search, examine OAI-SearchBot and its actual access. A reference to the page is still not guaranteed.

Do I need a special writing format for each AI?

Start with content your visitors can understand. Use headings, examples and questions where they help reading. Formatting alone does not demonstrate that a source will be selected.

Start with a specific review

Send us your URL, key services and target market. We will define the scope and priorities: access, content, schema and measurement. Explore our SEO service or discuss your project with Firstidea.