On-Page GEO starts where you decide what is written about your offering: on your own website. Whether AI systems such as ChatGPT, Gemini, Claude, and Perplexity describe your offering accurately and cite your website as a source depends in part on whether they can fetch your pages and how unambiguous the information on them is. On-Page GEO is one of the three areas of Generative Engine Optimization (GEO), alongside monitoring and Off-Page GEO, which deals with a brand’s presence beyond its own website, for example in media, directories, and reviews.

What On-Page GEO covers

Work on your own website falls into three layers that build on each other:

  • Access: The crawlers that AI systems rely on are allowed and able to fetch the pages. This includes suitable rules in robots.txt, bot protection that does not turn away wanted crawlers by accident, main content that is already in the HTML the server delivers, for example through server-side rendering, rather than appearing only after JavaScript runs, and an XML sitemap that makes pages easy to discover. How well this works is described by a website’s crawlability.
  • Readability: Software can tell how a page is structured and which facts it contains. Semantic HTML with a clear heading hierarchy, which also serves web accessibility, and structured data contribute to this. An llms.txt file gives an overview of the most important pages, optionally complemented by plain-text Markdown versions of those pages. The shared goal is machine readability.
  • Content: Copy names companies and products explicitly, states facts such as services, prices, and locations clearly, and remains understandable as a single passage. This citability makes it easier for an AI system to use a passage in an answer and cite the page as its source. Beyond citability, the content layer includes current information (content freshness), names and company facts that read the same everywhere (brand fact consistency), and, where readers expect it, details on which company and which people stand behind the content (authorship).

Crawlability and machine readability thus describe properties of a website; On-Page GEO is the work on them and on the content itself. Several of these measures take little effort, such as a rule in robots.txt or an llms.txt file.

What AI providers say

In Bing’s Webmaster Guidelines, which also cover Copilot, Microsoft describes GEO as work on making content eligible as a basis (grounding) and source for AI answers. Many of the practices listed there concern a brand’s own website: a logical heading hierarchy, structured data that accurately reflects the visible content, explicitly stated facts, and content that is updated when facts change.

Google states that its AI features, AI Overviews and AI Mode, require no new machine-readable files, no special markup, no content broken into small pieces, and no writing style just for AI. On-Page GEO, however, does not target a single system: an accessible, clearly structured website can benefit many AI systems, including AI agents that read and use websites on people’s behalf.

What On-Page GEO can achieve

No single measure guarantees that an AI system cites a page or mentions an offering; Microsoft, too, states that GEO does not guarantee AI citations. How much individual changes to copy achieve is disputed in research. In the paper that coined the term GEO in November 2023, adding references to sources within a website’s text, quotations from credible sources, and statistics clearly increased visibility in a test setup. A benchmark study presented at the NeurIPS 2025 conference, by contrast, found that most such methods for rewriting text were largely ineffective and frequently even had a negative effect. Both studies tested targeted rewrites of existing text, such as added statistics or more fluent wording, not whether facts are stated clearly on a page at all.

On-Page GEO therefore mainly creates the preconditions: an AI system cannot use content as a source if it cannot fetch it, and it cannot reproduce facts accurately if they are not stated clearly anywhere.