What is GEO? In marketing, GEO stands for Generative Engine Optimisation: the work you do to get mentioned and cited in the answers of AI assistants such as ChatGPT, Gemini and Perplexity, and in Google’s AI Overviews. It has nothing to do with geography or geolocation. The aim is to end up in the answer itself, which is a different goal from ranking high in a list of links.
Where does the term GEO come from?
From a 2023 paper by six researchers, later presented at the KDD 2024 conference. They use “generative engine” for a search system that pulls information from several sources and has a language model summarise it into a single answer. In their words, Generative Engine Optimisation is a way to help content creators become more visible in those answers.
GEO builds on SEO and does not replace it. How the two relate in practice is covered in our article on the difference between GEO and SEO.
How does an AI assistant arrive at its answer?
There are two routes: what the model learned during training, and what it looks up on the web at the moment you ask. As a business, you have the most influence over the second.
The GEO paper describes that second route in four steps. The assistant first rewrites your question into one or more search queries, and a search engine returns a ranked list of sources for them. The assistant summarises those sources. Finally it writes one answer, with references to the pages it relied on.
Google describes the same principle for its own AI Overviews and AI Mode. The system runs several related searches across subtopics and assembles the answer from the results, something Google calls “query fan-out”. A page can only appear as a source in such an answer if it is indexed and eligible to be shown with a snippet in the normal search results.
Assistants fetch pages with their own crawlers, programs that visit your site in the same way Googlebot does. You decide which ones to let in through your robots.txt.
| Crawler | From | Used for | If you block it |
|---|---|---|---|
| OAI-SearchBot | OpenAI | Including sites in ChatGPT’s search feature | Your site will not appear in ChatGPT’s search answers |
| GPTBot | OpenAI | Collecting content that may be used to train models | You signal that your content may not be used for training |
| ChatGPT-User | OpenAI | Opening a page when a user asks for it in ChatGPT | Robots.txt rules may not apply here, because the user starts the visit |
| PerplexityBot | Perplexity | Showing and linking sites in Perplexity’s search results | Your site will not appear in Perplexity’s search results |
Plenty of people miss the difference between the first two rows. Turn away every AI crawler because you don’t want your writing in a training set, and you also drop out of the search answers. You can make those two choices separately.
What does GEO optimisation involve in practice?
It comes down to four kinds of work. The first is technical; the other three concern content and reputation.
1. Make sure assistants can read your site
Check your robots.txt against the crawlers in the table above. Put your most important information on the page as plain text and make it reachable through internal links. Google asks the same of any site it crawls. A change to your robots.txt does not take effect straight away at OpenAI: their documentation says to allow about a day.
In a first technical check we most often find AI crawlers that have been blocked without anyone noticing, and structured data that is missing or broken. Pages that sell but never answer a concrete question come up nearly as often.
2. Answer the questions people really ask
People type whole questions into an assistant, not a few keywords. The assistant breaks that question into sub-questions and searches for sources for each one. So write pages that answer one question clearly, with the answer at the top and the explanation below it. Put the question in the subheading, so people and machines see at once what the paragraph covers.
3. Back up what you claim
This is the part that has been researched most. The researchers who coined the term GEO tested nine ways of adjusting a text. Adding quotations, adding statistics and citing sources worked best. Old-fashioned keyword stuffing achieved little or nothing. A claim with a source the assistant can check seems to be more useful to it than the same claim without one.
4. Be findable beyond your own website
An assistant builds its answer from several sources, and your own site is at most one of them. The more often your name turns up in the places an assistant searches, the better your chance of being in the answer. Think of trade media, reviews and comparisons in your sector.
What about llms.txt?
An llms.txt is a text file in the root of your site that summarises your most important pages in Markdown, written for language models. Jeremy Howard proposed it in September 2024, and it is not an official standard. Google says explicitly that you don’t need new machine-readable files or AI text files to appear in AI Overviews or AI Mode.
We still add the file. It takes little effort and does no harm, and writing it makes you decide which pages really matter. Just don’t expect miracles from it. On WordPress we handle llms.txt and AI crawler access with our own plugin, Odyssee SEO AI. For NLO it was part of a broader technical overhaul, together with the sitemap.
What has been proven, and what hasn’t?
Less than many vendors would have you believe. GEO is a young field, and the assistants change faster than the research can follow. The table sets out what public sources support.
| Claim | Status |
|---|---|
| A site that blocks OAI-SearchBot will not appear in ChatGPT’s search answers | Stated in OpenAI’s documentation |
| Google’s AI Overviews have no extra requirements on top of normal SEO | Stated in Google’s documentation |
| Adding sources, quotations and statistics increases your visibility in AI answers | Shown in an experiment, with a maximum gain of around 40 per cent. The effect varies by subject area |
| An llms.txt gets you mentioned more often | Not shown. It is a proposal, and Google says it doesn’t need it |
The researchers measured on their own test set of ten thousand queries and, separately, on Perplexity. That is no guarantee for your site in every assistant.
We often hear that SEO is dead and that an llms.txt makes all the difference. We disagree with both. The first doesn’t fit how the assistants work, because they search first and summarise afterwards. Nobody has demonstrated the second.
How do you measure GEO?
You measure two things: whether you get mentioned, and whether that brings in visitors.
Mentions. Draw up a fixed list of questions your customers would put to an assistant. Put them to several assistants at regular intervals and record whether you are mentioned, whether your site is used as a source and who else appears in the answer. A single answer tells you little, so repeat the measurement.
Traffic. Visits that arrive through an AI assistant show up in GA4, and we see them in our clients’ data. Google is harder. In Search Console, clicks from AI Overviews and AI Mode are counted in the ordinary total for the Web search type, so you can’t read them out separately there.
Keep it in proportion, though. SE Ranking analysed Analytics data from more than 100,000 websites and found that in 2026 AI platforms accounted for 0.32 per cent of all website traffic, and 0.16 per cent in the UK. Google still sends 134 times as many visitors, according to the same study. AI traffic is growing fast, mind: sixteen times as much as in 2024.
On the websites we manage, the share is higher. In the third quarter of 2026, 1.6 per cent of measured sessions across ten sites demonstrably came from an AI assistant, 95 per cent of it from ChatGPT. Google organic search accounted for 45 per cent over the same period, nearly thirty times as much. These are mostly Dutch businesses, with a fair number of hospitality and local firms among them, so this is our own small sample and no market benchmark. On the seven sites where we also have figures for the previous year, visits from AI assistants more than doubled.
So don’t move your whole budget to AI. Most traffic still comes from Google, and what you do well there carries through into the answers assistants give.
Then there is what people say themselves. Leads increasingly tell us they found a business through an AI assistant. Ask about it in your contact form or on the first call; you won’t find a cheaper way to measure.
For clients we track mentions with a tool that puts the same questions to several assistants at regular intervals. That shows how things develop over time, which a snapshot can’t.
To see where your business stands now, our free growth check looks at how findable you are in Google and in AI answers. How we take it from there is described on our SEO and GEO service page.
Frequently asked questions
What does GEO stand for?
In this context GEO stands for Generative Engine Optimisation. It means optimising your site and your brand so that AI assistants such as ChatGPT, Gemini and Perplexity mention you and use your pages as a source. The term comes from a research paper published in 2023.
Do I need an llms.txt for GEO?
No. It is a proposal and not yet an official standard, and Google says you don’t need separate AI files for its AI features. It is quick to make and harmless, so we add it anyway. It won’t make up for weak content or a site that crawlers can’t access.
Can I see in Search Console how many clicks come from AI Overviews?
Not separately. Google says impressions and clicks from AI Overviews and AI Mode are included in the Performance report under the Web search type. Traffic from other assistants you check in your analytics, for example GA4.
Is GEO worth doing if you don’t rank at the top of Google?
Yes, as long as you can be found at all. In the GEO study, pages ranked lower in the search results gained the most from adding sources, which suggests an assistant doesn’t simply pick the top result. If you don’t rank anywhere, there is little to cite, and the place to start is the SEO groundwork.
Sources
- Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024
- Google Search Central, AI features and your website
- OpenAI, Overview of OpenAI crawlers
- Perplexity, Perplexity crawlers
- Jeremy Howard, The /llms.txt file
- SE Ranking, AI Traffic Research Study, June 2026
- Odyssee, our own analysis of GA4 data from ten websites we manage, Q3 2026