AI search optimisation. Be understood by answer engines.

In its guide to generative AI features, Google explains that AI Overviews and AI Mode are rooted in its usual ranking and quality systems: they retrieve pages from Google’s index to write their answer and show links to those pages. For Google, optimising for generative AI search is optimising for search: it is still SEO.

Google adds that no special file, no particular markup and no writing “for AI” is needed. At Microsoft, Copilot relies on the same crawling and index as Bing search. So I don’t sell a secret recipe: I strengthen the foundations that matter to these engines.

Google recommends helpful, reliable content with a point of view of its own: first-hand experience, expertise and concrete cases, rather than a summary of what already exists. It also advises pages organised into paragraphs and sections, with clear headings.

With you, I spell out what you do, for whom, how and on what terms, and I back every important claim with a source or a verifiable fact. No artificial chunking and no set length: Google says there is no ideal page length. No pages multiplied for every search variation either: when the aim is to manipulate rankings, Google treats it as scaled content abuse.

AI companies use separate crawlers that can be set in the robots.txt file, the file that tells crawlers what they may read. At OpenAI, GPTBot collects content for model training, while OAI-SearchBot is used to show sites in ChatGPT search; the two settings are independent. Anthropic likewise separates ClaudeBot, for training, from Claude-SearchBot, for the quality of its search results. Perplexity states that PerplexityBot is not used to train models. At Google, the Google-Extended token (a crawler name set in robots.txt) controls whether your content is used to train Gemini models, with no effect on your presence in Google Search.

I check that the crawlers you accept actually get through: robots.txt, firewall and anti-bot protection. Anthropic states, for example, that its crawlers do not bypass CAPTCHAs, so overly strict protection can stop them. I also point out the limits: user-triggered fetches, such as ChatGPT-User or Perplexity-User, may not follow robots.txt.

Structured data (schema.org) describes your business, your services and how your pages are organised. Google states that it is not required for generative AI search and that there is no special markup for AI, but that it remains useful for SEO. I add it when it faithfully describes what the page shows.

The llms.txt file is a proposal, not a standard: Google says its search ignores it, with no positive or negative effect. So I don’t present it as a visibility lever. Instead, I make sure essential content is in the page’s code, since Google points out that not all crawlers can run JavaScript. For Bing and Copilot, the IndexNow protocol lets you report every page you add, change or delete.

Google Search Console measures how your site appears in Search, including its generative AI features (AI Overviews, AI Mode): impressions, clicks, pages, countries, devices and trends over time. I connect these figures to the enquiries you actually receive.

Google warns against tools that promise rankings or claim to use its internal data: no third-party tool has access to its ranking or AI systems. The official reports are therefore the reference, and every action is judged by its measured effects.