GEO, or Generative Engine Optimization, describes work intended to help a source appear in answers from AI search systems such as Google AI Overviews, AI Mode, or ChatGPT. GEO does not replace SEO. For search-backed answers, a page has to be reachable and eligible as a source for that system. Google, for example, requires a page to be indexed and snippet-eligible for AI Overviews and AI Mode; that still does not guarantee a mention. Beyond this, retrieval and selection differ by provider and are only partly documented.
The topic attracts a lot of confident claims and very little evidence. This guide separates the two. By the end you will know where the term comes from, which parts of the job really change with AI search, which parts do not, and what to work on first when you have limited time for both.
What GEO is and where the term comes from
The term comes from a research team led by Pranjal Aggarwal. Their paper, "GEO: Generative Engine Optimization", was posted to arXiv in November 2023 and accepted at KDD 2024. The authors built a test set called GEO-bench with 10,000 queries across many subject areas. They then measured how changes to a source affected how much of it showed up in a generated answer.
The findings are useful once you read them carefully. Three plain edits performed best: citing sources, adding relevant quotations and adding statistics. In the paper's measurements they improved visibility by roughly 30 to 40 percent in relative terms. Keyword stuffing barely moved anything and in some tests did worse than the untouched text. Another notable result: sources that ranked lower in traditional results gained much more than the sources at the top.
Two caveats belong next to those numbers. The experiments ran on a research benchmark, with additional tests on Perplexity, not on Google or today's ChatGPT. And "up to 40 percent more visibility" describes a metric from the paper, not a promise for your site. The direction still holds up: specific, sourced content is more likely to be used in a generated answer than padded content.
For a short definition and related terms, see generative engine optimization in the glossary.
GEO vs SEO: what actually changes
Retrieval differs by provider and is only partly documented. Google describes query fan-out for AI Overviews and AI Mode: several related searches across subtopics and sources may feed into an answer. OpenAI documents OAI-SearchBot for ChatGPT Search but does not publish the full retrieval and source-selection process. It is therefore not safe to say every system always retrieves passages rather than whole pages. A useful editorial check is still simple: can a section answer one concrete question without the preceding paragraphs while keeping its conditions and limits? That is a quality rule for readers, not a documented ranking lever. Do not use retrieval-augmented generation as a blanket description of every provider.
Google may fan one question out into several searches. Google Search Central says AI Overviews and AI Mode may use several related searches across subtopics and sources. Google calls this query fan-out. Other providers do not disclose all comparable steps. Google’s documentation therefore does not establish a general rule for every AI answer, nor does it predict which subquery will surface a particular page as a source.
For a practical check of which search crawlers can reach your site, see the guide to AI crawlers and robots.txt.
Citations and visits are different signals. In a browser sample of 900 U.S. adults, Pew Research Center observed fewer clicks on classic results during visits with an AI summary than during visits without one. That is an observation for this sample and time period, not a forecast for your site. A source mention and a visit are therefore not interchangeable; measure them separately.
The output is not a fixed rank. A list of blue links can be measured as a position. A generated answer names brands in a changing order and links one source in one run and another in the next. In our assessment, a single prompt tells you very little. A fixed set of questions that you check repeatedly tells you a lot more.
What stays the same in GEO and SEO
This is the good news for anyone who already does SEO properly: the fundamentals carry over. In the same documentation, Google says a page needs to be indexed and eligible to appear in Search with a snippet to show up as a supporting link in AI Overviews or AI Mode. According to Google, there are no additional technical requirements, no special optimizations, no special schema markup and no special AI text files.
That leaves a sober list of things that have not changed:
- Crawlability. Anything a bot is not allowed to fetch, or cannot reach because of a firewall, is missing from every index, classic or generative.
- Indexing. Only indexed pages can be sources for AI Overviews. If you exclude a page with
noindexornosnippet, you exclude it there too. The glossary covers the basics under indexing. - Search intent. A page that misses the intent behind a question is as weak a source for an answer as it is a candidate for ranking.
- Quality and trust. Google recommends the same principles for AI features as for the rest of Search: unique, useful content made for people, a good page experience and structured data that matches the visible text.
- Internal links. Content that is well linked internally is easier to find. Google lists this explicitly in its recommendations for AI features.
For planning purposes, this means a solid SEO foundation is not a stage you leave behind for GEO. It is the precondition without which none of the steps below will work.
A GEO example: rewriting one paragraph
Take a made-up section from the services page of an accounting firm:
> We have supported clients with passion and expertise for many years. For everything that matters to freelancers, we are your trusted partner. Get in touch today.
For a human reader the paragraph is interchangeable. For an AI system it is useless. It names no subject, no condition and no answer. A revised version could read:
> We work with freelancers and sole proprietors from their first tax year. We handle bookkeeping, quarterly estimated tax payments and the annual return. We do not prepare financial statements for corporations.
This paragraph answers a concrete sub-question, stands on its own and states a limit. That makes it easier to retrieve and easier to quote correctly. The example is deliberately simple. It points in the same direction as the GEO paper: fewer claims, more statements someone can check.
How to measure GEO: AI visibility instead of rank
Classic SEO measures rankings, impressions and clicks. GEO adds another question: do AI systems mention you for the questions that matter to your business, and how? That measure is called LLM visibility.
A workable setup has four parts:
- A fixed question set. A few dozen questions real customers would ask, phrased the way people talk rather than the way they type keywords.
- Several systems. ChatGPT, Gemini and Google AI Overviews answer differently. A result from one tells you little about the others.
- Separate signals. Record whether your brand is mentioned, in which position and whether your site is linked as a source. A mention without a link is not the same as a citation.
- Repetition. Check the same questions at fixed intervals and compare trends rather than single results.
Add the data you already have. According to Google, traffic from AI features is counted in the Search Console Performance report under the "Web" search type. Based on that description it is not broken out separately, so you only see its effect in the total.
In Crawl Foundry, the AI Visibility feature handles this measurement. You define the questions, and Crawl Foundry checks them regularly in ChatGPT, Gemini and Google AI Overviews, plus AI Mode where available if you choose it. You see which brands get mentioned and cited, a 0 to 100 score per engine and your share of voice against the competitors you track. You see the maximum price of a check before it starts.
To confirm that search and AI crawlers can fetch and read a specific page at all, run it through the AI Crawler Checker first. It is free and needs no account.
SEO vs GEO: an honest priority list
When time is short, order matters. The logic is simple: a page first has to be reachable, then it has to say something useful, and only then does fine-tuning pay off.
- Check technical access. robots.txt, firewall and CDN rules must not block Googlebot or the AI providers' search crawlers if you want to appear in their answers. For OpenAI, that crawler is OAI-SearchBot, not the training crawler GPTBot.
- Answer questions directly. Every important section opens with the answer and names its subject and conditions, so it can stand alone.
- Show your evidence. Cite sources, give figures with their origin and, where it fits, quote named experts. These were the edits that helped most in the GEO paper.
- Add something of your own. If ten pages say the same thing, an answer needs only one of them. Original data, tests or documented experience give a page a reason to be chosen.
- Make your brand and offer unambiguous. An AI system can only mention you if it is clear who you are and what you sell. That means a consistent name, clear product names and an about page with facts someone can verify.
- Measure before you rebuild. Set up the question set before you change content. Otherwise you will not know later whether a change did anything.
You can skip keyword stuffing, mass-produced pages for every conceivable question and files that promise a shortcut into AI answers. An llms.txt file can make sense for specific purposes. For AI Overviews and AI Mode, Google says no special AI text files are needed.
For the evidence behind the file, see how to evaluate llms.txt and what needs measurement.
Common GEO misconceptions
"GEO replaces SEO." For Google's AI features, GEO draws on the same index. Without SEO fundamentals there is nothing to cite.
"I need to write differently for AI." Clearly structured text with direct answers helps people and machines alike. There is no separate writing style for language models. Whatever tires readers, such as empty intros and filler sentences, does not help a retrieval system either.
"Citations can be guaranteed." Each system decides which sources its answer uses. Anyone who guarantees mentions in AI answers is promising more than they control.
"One check is enough." A single answer is a snapshot. Only repeated checks over several weeks show whether anything is moving.
Bottom line: GEO is SEO with one more checkpoint
GEO changes how content gets found and how you measure success. It does not change what a page fundamentally needs: access for crawlers, a clean index, a clear answer to a real question and evidence. If you have those fundamentals in place, most of the GEO work is already done. What you add is mainly two things: passages that stand on their own and regular measurement of AI visibility.
Start with the question set. Write down the twenty questions where you want to appear in an AI answer, and check who gets mentioned there today.