What is GEO? Generative Engine Optimization, explained for experts

September 25, 2026 · 5 min

What GEO actually means

The term comes from a 2023 paper by researchers at Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi. Their question was simple: when a language model composes an answer from web pages, which pages does it choose to cite — and can a page make itself more citable? Their answer was yes, and measurably so.

In practice, GEO covers three things: access — can the assistant’s crawler read your page at all? — comprehension — can it tell, in a few lines, who you are and what you claim? — and trust — does the page give it reasons, such as sources, figures and credentials, to quote you rather than someone else?

GEO and SEO: what changes, what does not

Everything that makes a page good for Google still matters: fast loading, clean HTML, one address per page, honest titles. Google itself states that its AI features draw on the same index and the same signals as classic search. GEO is not a second website; it is a second reader.

• SEO rewards the page that best matches a query; GEO rewards the passage that best answers a question.
• SEO counts the links pointing to you; a generative engine also weighs the sources you cite and the facts it can check.
• SEO lives in ten results; a generative answer usually names two to five sources. The gap between first and nowhere is wider.

The consequence for experts is uncomfortable but useful: a beautiful site that says little in its first paragraph is nearly invisible to machines, while a plain page that states its claim clearly, with a source, can be quoted to millions of people.

What measurably increases your chances of being quoted

The Princeton study tested nine ways of rewriting a page. The methods that helped most were not tricks but habits of scholarship:

• Cite your sources. Pages that referenced authoritative sources were chosen more often.
• Add statistics. A precise figure is easier to quote than an adjective.
• Include quotations from recognised authorities in your field.
• Write fluently and plainly. Keyword stuffing, the oldest SEO reflex, performed poorly.

Together, these changes raised visibility in generative answers by up to 40%. For a researcher, none of this is new: it is the discipline of the abstract, applied to the web.

The technical layer: let the machines in

None of the above matters if the crawler never reads your page. Three checks come first:

• Your robots.txt must not block GPTBot, OAI-SearchBot, ClaudeBot or PerplexityBot. Many templates block them by default.
• Your host may block them too. Some providers reject AI crawlers at the firewall, before your site is even reached — test with each crawler’s real user agent.
• Your content must be in the HTML. Most AI crawlers do not run JavaScript. Text injected by scripts, and pages that mix several languages on one address, reach them garbled or empty.

Then give the machines structure: schema.org data describing who you are (Person or Organization), what you offer (Service), what you wrote (Article) and the questions you answer (FAQPage). An llms.txt file — a short, plain summary of your site, proposed in 2024 as a convention for language models — takes ten minutes to write and removes ambiguity.

A one-month GEO plan for an expert’s website

• Week 1 — Access. Open robots.txt to AI crawlers, test them against your host, make sure your text is in the HTML.
• Week 2 — Clarity. Rewrite the first paragraph of your five key pages as an answer: who you are, for whom, with what result.
• Week 3 — Proof. Add sources, figures, credentials and dates; add structured data and an llms.txt.
• Week 4 — Depth. Publish one substantial article answering the question your clients ask most, with an FAQ.

Then measure. The only honest GEO metric is observed behaviour: which AI crawlers visit, how often, and how many visitors arrive from ChatGPT, Perplexity or Gemini. Every site The Abstract hosts reports exactly that, in a dashboard its owner can read.

Updated · September 25, 2026