What is GEO? Generative Engine Optimization explained

What does GEO mean?

Generative Engine Optimization (GEO) is the practice of optimising content for search systems that write their own answers. These include ChatGPT Search, Perplexity, Gemini, and AI Overviews and AI Mode in Google Search. The goal is for the system to mention your brand in its answer and to link your page as a source.

The term comes from research. Pranjal Aggarwal and five co-authors introduced it in a paper published in November 2023 and accepted at the KDD 2024 conference. You will find a short definition in our glossary entry on GEO.

GEO does not replace SEO. Classic SEO aims for a position in the results list. GEO aims for a mention inside a generated answer, often without a click (the click is more of a bonus there). Our article on GEO vs. SEO explains where the two overlap.

How do AI search tools pick sources?

AI search tools search first and write second. The system retrieves matching documents from a search index and hands them to a language model. The model writes the answer and links some of the pages it used. This pattern is called retrieval-augmented generation, or RAG.

That creates a hard requirement. A page that the responsible crawler may not fetch cannot appear as a source. Each provider names its crawler in its own documentation.

Google AI Overviews and AI Mode

According to Google Search Central, AI Overviews and AI Mode may run several related searches across subtopics. Google calls this query fan-out. Any page that is indexed and eligible to show with a snippet can be used as a source. Google says there are no additional technical requirements or special optimisations. Read more in our article on Google AI Overviews.

ChatGPT Search

OpenAI uses the crawler OAI-SearchBot for its search feature. If you block it in robots.txt, OpenAI says your site will not appear in ChatGPT search answers. GPTBot is a separate crawler that collects content for model training. You can therefore allow search and opt out of training.

Perplexity and Gemini

Perplexity uses PerplexityBot to surface and link websites in its results. Perplexity states that this crawler is not used to train foundation models. For Gemini and other Google systems outside Search, the Google-Extended token controls whether content may be used for training and grounding.

What did the GEO study find?

Evidence in the text makes content more visible. That is the central result of Aggarwal et al. The researchers built a dataset called GEO-bench with 10,000 queries from many domains. They then rewrote source texts with nine methods and measured how prominently each text appeared in the generated answers.

• Citing sources, adding quotations and adding statistics delivered a relative improvement of 30 to 40 percent.

• Keyword stuffing brought little or no gain and even scored below the baseline on one metric.

• Lower-ranked pages benefited most: citing sources raised the visibility of a page in fifth place by 115.1 percent, while the page in first place lost 30.3 percent.

What can brands do?

The following steps are derived from the providers' documentation and the GEO study. The order reflects our recommendation.

1. Check crawlers: allow Googlebot, OAI-SearchBot and PerplexityBot in robots.txt. Decide separately on GPTBot and Google-Extended.

2. Answer first: answer the page's main question in the first paragraph, then add detail.

3. Add evidence: cite sources, figures and quotations with their origin. These were the methods that worked in the study.

4. Clarify entities: use the same company name, offer, people and location everywhere. Our glossary entry on entities explains why.

5. Earn mentions: in our assessment it helps to be named on sites that AI search tools often link as sources, such as trade media and comparison sites.

6. Measure: define a fixed set of prompts and check it regularly.

New file formats are optional. Google states explicitly that AI Overviews require neither AI text files nor special schema markup. Whether an llms.txt file helps with other providers is still an open question.

How do you measure success in AI search?

Single queries are not a measurement. A SparkToro analysis from January 2026 makes this clear. When the same prompt was repeated, ChatGPT and Google's AI features in Search returned the same list of brands in less than 1 percent of runs. The same order appeared in less than 0.1 percent of runs.

A better metric is the share of answers that mention your brand, measured across many runs. Specialised tools automate this. Our comparison of AI visibility tools gives an overview.

Google leaves a gap here. According to Google, clicks from AI Overviews and AI Mode count as normal search traffic in the Search Console performance report under the search type “Web”. There is no separate filter for AI features.

Where should you start?

Start by checking robots.txt and the indexing status of your most important pages. Then pick ten to twenty questions your customers would ask an AI search tool and record today's answers. If you are looking for outside help, read our guide on choosing a GEO agency.