Generative Engine Optimization (GEO): How AI Answers Choose Their Sources

When someone asks ChatGPT, Perplexity or Google's AI Overviews a question, the answer names only a handful of pages. GEO is the work of making your page one of them. Here is what it means, what the research found, and what to change on your own site.

By Aji Paul · Published 3 January 2026 · Updated 29 September 2026

What is generative engine optimization?

The term comes from a 2023 research paper, "GEO: Generative Engine Optimization" by Pranjal Aggarwal and five co-authors, later accepted at the KDD 2024 conference. The authors treat answer engines as a black box: you cannot see the ranking code, so you test which changes to the text move the result.

The difference from classic SEO is the prize. A results page gives ten links and lets the reader choose. An AI answer reads several sources, writes one reply and credits a few of them inline. Ranking eighth for a query can still bring clicks; being the eighth source an engine read and did not cite brings nothing. GEO is about crossing the line from "read" to "cited".

How does an AI answer engine pick the pages it cites?

Most answer engines work in two steps. First they retrieve: they run searches against a web index and pull back candidate passages. Then they generate: a language model writes the answer from those passages and attaches citations to the parts it used. Your page has to survive both steps.

Google AI Overviews and AI Mode

Google's documentation, AI features and your website, is blunt about the first step: there are no extra requirements and no special optimizations for AI Overviews or AI Mode. To be shown as a supporting link, a page must be indexed and eligible to appear in Google Search with a snippet. You do not need new machine-readable files or AI text files, and the usual snippet controls (nosnippet, data-nosnippet, max-snippet, noindex) also govern how your content appears there.

ChatGPT search

OpenAI runs separate crawlers for separate jobs, listed in its crawler documentation. OAI-SearchBot surfaces websites in ChatGPT's search features; sites that block it are not shown in ChatGPT search answers. GPTBot collects content that may be used to train OpenAI's models. The two are independent, so a site can refuse training while still allowing search. Robots.txt templates copied from elsewhere often block both without the owner meaning to.

Gemini training versus Google Search

Google offers a similar split. The Google-Extended token in robots.txt controls whether content Google crawls may be used to train future Gemini models, and Google's crawler overview states that it has no effect on a site's inclusion or ranking in Google Search.

What does the research say gets a page cited?

The GEO paper built a benchmark of real user queries across many subjects, called GEO-bench, and rewrote source pages in several ways to see which rewrites raised a page's visibility in the generated answer. The best methods boosted visibility by up to 40%. The detail is more useful than the headline:

• Adding quotations and adding statistics worked best. The paper reports relative improvements of roughly 30 to 40% on its main visibility measure for these methods.
• Clearer writing helped too. Rewrites that only improved fluency and readability still raised visibility by about 15 to 30%.
• Keyword stuffing did not help. The authors found it offered little to no improvement in generative answers.
• Smaller sites gained the most. Pages that ranked lower in normal search benefited far more. In one test, citing sources raised visibility by 115.1% for pages ranked fifth, while the top-ranked page lost 30.3% on average.

A GEO checklist for your own pages

1. Confirm the page can be crawled and indexed. Check it with the URL Inspection tool in Google Search Console, and read your robots.txt line by line for Googlebot, OAI-SearchBot and any other search crawler you want to reach you.
2. Put the answer in the first sentence. Use a heading shaped like the question people ask ("How long does an ERP rollout take?") and answer it in one or two plain sentences directly under it. Detail can follow.
3. Give every figure a named source and a link. An unsourced number reads as opinion; a number with the survey, the year and a link is evidence an engine can lift and credit.
4. Quote real documents and people accurately. Short, exact quotations from standards, official documentation or a named expert are the kind of passage the research found engines reuse.
5. Show who wrote it and why they know. A visible author with a profile page, a published date and an updated date help readers and machines judge the page. Change the updated date only when the content changes.
6. Keep structured data truthful. Article, Organization and FAQ markup should repeat what the page visibly says, as Google recommends. Markup for questions that are not on the page does more harm than good.
7. Give each question one home. Several near-identical pages split the signal and invite the engine to pick none. Merge them, or make each one about a genuinely different question.
8. Do not choke the snippet. A strict max-snippet or a nosnippet tag limits what Google can show from the page, and that includes its AI features.

How do you measure GEO?

There is no single dashboard for AI citations yet, so measurement combines three sources.

• Search Console. Google says traffic from AI Overviews and AI Mode is counted in the Performance report under the "Web" search type. It is not broken out separately, so watch the pages you reworked rather than the site total.
• Referral traffic. Visits from chatgpt.com, perplexity.ai and similar domains appear as referrals in most analytics tools. For most sites the numbers are small; the trend is what matters.
• A fixed question panel. Write down 20 to 30 questions your buyers actually ask. Once a month, ask them in the engines you care about and record which pages are cited, including competitors' pages.

GEO, AEO and SEO: how do they differ?

In practice the three overlap heavily. A fast, crawlable page with a clear answer, real evidence and a named author does well in all of them. What GEO adds is discipline about sources and structure, because an engine assembling an answer needs passages it can lift cleanly and attribute with confidence.