What Is GEO? The Complete Guide to AI Search Optimization
What is GEO?
GEO, short for Generative Engine Optimization, covers all the techniques that increase the probability of an AI answer engine citing your site, your brand or your products in the answer it generates. It is also called AI search optimization, generative search optimization or optimization for answer engines.
The term was popularized by a research paper, “GEO: Generative Engine Optimization”, by Pranjal Aggarwal and his co-authors, presented at the KDD 2024 conference. [1][1] Pranjal Aggarwal et al., « GEO: Generative Engine Optimization », KDD 2024 The authors show that it is possible to increase a piece of content's visibility in generative engine answers by up to 40%, and that some classic techniques, such as keyword repetition, do not work.
What is the difference between GEO, SEO, AEO and LLMO?
These acronyms refer to related approaches. What sets them apart is the goal.
| Acronym | Meaning | Goal |
|---|---|---|
| SEO | Search Engine Optimization | Rank pages in search engine results |
| GEO | Generative Engine Optimization | Be cited in AI-generated answers |
| AEO | Answer Engine Optimization | Provide the direct answer to a question: snippets, voice assistants, answer engines |
| LLMO | Large Language Model Optimization | Influence what a language model “knows” about a brand |
In practice, GEO encompasses AEO and LLMO, and it builds on SEO. Google says so itself for its AI features: there is no additional requirement or special optimization to appear in AI Overviews or AI Mode; SEO best practices remain the foundation. [2][2] Google Search Central, « AI features and your website »
How does an AI choose the sources it cites?
An AI assistant answers in two ways, each with its own levers.
From its internal knowledge
The model answers with what it learned during training, from an immense amount of public text. A brand that is often mentioned, described consistently and associated with a topic is more likely to be “known” by the model. This knowledge updates at the pace of new training runs, so slowly.
By searching the web in real time
For recent or specific questions, the assistant runs searches, fetches pages, extracts the useful passages, writes a synthesis and cites its sources. This is how Perplexity, ChatGPT with search, Copilot, AI Overviews and Google's AI Mode work. Everything happens in a few seconds: the page must be accessible to bots, readable without JavaScript and contain a passage that clearly answers the question.
One technical point is decisive. According to a study by Vercel and MERJ published in December 2024, none of the major AI bots, including those of OpenAI, Anthropic and Perplexity, executes JavaScript. [3][3] Vercel et MERJ, « The rise of the AI crawler », 17 décembre 2024 A site whose content only appears after JavaScript runs is therefore largely invisible to them. Google, which also powers Gemini, and Apple are the exceptions.
Which levers increase your chances of being cited?
Available studies and engine recommendations converge on eight levers.
- Be accessible: allow search engine and AI assistant bots in your robots.txt, and check that no firewall blocks them.
- Serve complete HTML: server-side rendering or pre-rendering, so content is readable without JavaScript.
- Answer first: a clear answer in the first two sentences of each section, details afterwards.
- Provide evidence: dated figures, cited sources, expert quotes. In the GEO study, adding citations, statistics and sources is among the most effective methods, while keyword stuffing performs worse than the original text. [1][1] Pranjal Aggarwal et al., « GEO: Generative Engine Optimization », KDD 2024
- Structure: headings phrased as questions, lists, tables, short definitions, FAQs.
- Clarify the entity: same name, same activity description and same contact details everywhere (site, profiles, directories), with structured data matching the visible text. [2][2] Google Search Central, « AI features and your website »
- Be confirmed elsewhere: mentions in media, directories and reference sites, customer reviews on third-party platforms. In practice, an AI is more likely to reuse information that several sources confirm.
- Stay up to date: visible update dates, revised content, accurate practical information, a well-kept Google Business Profile. [2][2] Google Search Central, « AI features and your website »
Which AI bots should you allow?
Each AI publisher uses several bots, with different roles.
| Publisher | Bot | Role |
|---|---|---|
| OpenAI | OAI-SearchBot | Surfaces sites in ChatGPT search results [4][4] OpenAI, « Overview of OpenAI Crawlers » |
| OpenAI | GPTBot | Collects content that may be used to train models [4][4] OpenAI, « Overview of OpenAI Crawlers » |
| OpenAI | ChatGPT-User | Visits a page at a ChatGPT user's request [4][4] OpenAI, « Overview of OpenAI Crawlers » |
| Anthropic | Claude-SearchBot | Crawls the web to improve the quality of Claude's search results [5][5] Anthropic, article d'aide sur les robots d'Anthropic |
| Anthropic | ClaudeBot | Collects content that may be used to train models [5][5] Anthropic, article d'aide sur les robots d'Anthropic |
| Anthropic | Claude-User | Visits a page at a Claude user's request [5][5] Anthropic, article d'aide sur les robots d'Anthropic |
| Perplexity | PerplexityBot | Surfaces and cites sites in Perplexity results [6][6] Perplexity, « Perplexity Crawlers » |
| Perplexity | Perplexity-User | Visits a page at a user's request [6][6] Perplexity, « Perplexity Crawlers » |
| Googlebot and the Google-Extended token | Googlebot crawls for search and its AI features; Google-Extended is not a bot but a robots.txt token that lets you limit the use of your content for training and grounding some of Google's other systems [2][2] Google Search Central, « AI features and your website » |
Blocking training bots (GPTBot, ClaudeBot) or the Google-Extended token is a legitimate choice to protect your content. But if your goal is visibility, let at least the search bots access your site: Googlebot, Bingbot, OAI-SearchBot, Claude-SearchBot and PerplexityBot. Also check your host or CDN settings: Cloudflare announced in July 2025 that it blocks AI bots by default. [7][7] MIT Technology Review, 1er juillet 2025
Why is server-side rendering essential?
Because most AI bots only read the HTML sent by the server. [3][3] Vercel et MERJ, « The rise of the AI crawler », 17 décembre 2024 To check, open a page of your site, right-click and choose “View page source”: if the page's text is not there, most AI bots cannot see it. My free crawler Arachnide SEO Spider automatically compares raw HTML and post-JavaScript HTML, page by page.
Do structured data help GEO?
Google states that no special markup is required for its AI features, but asks that structured data match the visible text. [2][2] Google Search Central, « AI features and your website » schema.org markup remains useful to describe unambiguously who you are (Organization, Person, LocalBusiness), what you sell (Product, Service, Offer) and who wrote your content (Article and its author). It strengthens the consistency of your entity, without ever replacing visible content.
Do you need an llms.txt file?
llms.txt is a proposed file that summarizes a site for language models. It costs almost nothing to create, but Google states that no special file is required for its AI features. [2][2] Google Search Central, « AI features and your website » Treat it as a bonus, never as a strategy.
How do you write a page an AI wants to cite?
A citable page looks like a good reference sheet: it answers quickly, proves what it claims and breaks down easily into self-contained passages.
- A title that mirrors the question as your customers ask it.
- A 40-to-60-word answer right under the title.
- Short sections, each focused on one sub-question.
- Dated figures, with their source and a link.
- Lists and tables for comparisons, steps and prices.
- An identified author, with their credentials, and an update date.
- Information found nowhere else: your data, your cases, your experience.
- An FAQ that picks up your customers' real questions.
How do you measure your visibility in AIs?
Visibility in AIs can be measured, provided you accept that answers vary from one time to the next. So we reason in frequency and trend.
- A question panel: 30 to 100 real questions from your customers, tested regularly in each engine.
- Citation rate: the share of answers that mention your brand or cite your site.
- Share of voice: your presence compared with your competitors' on the same panel.
- Assistant traffic: visits from chatgpt.com, perplexity.ai, copilot.microsoft.com or gemini.google.com in your analytics.
- Search Console: Google includes appearances in its AI features in the Performance report, search type “Web”. [2][2] Google Search Central, « AI features and your website »
To automate this tracking, I created Vigie, a free, open-source AI visibility tracker.
Which mistakes make a site invisible to AIs?
- Content that only exists after JavaScript runs.
- AI bots blocked without knowing it, by robots.txt or by the host.
- Contradictory information about the company from one site to another.
- Claims without evidence and statistics without sources.
- Fake reviews or testimonials: the European “Omnibus” directive bans publishing fake consumer reviews, and they destroy trust. [8][8] Directive (UE) 2019/2161 du 27 novembre 2019, dite « Omnibus »
- Generic pages, rewritten by AI, with no new information.
- No identified author, no date.
Where to start?
- Test 10 key questions from your market in ChatGPT, Perplexity and Google, and note who is cited.
- Check that your pages display without JavaScript and that AI bots are not blocked.
- Rewrite your 5 most important pages in answer format, with evidence.
- Harmonize your company description across your site, your Google Business Profile, LinkedIn and your directories.
- Measure again in a month.
Frequently asked questions about GEO
Will GEO replace SEO?
Can you pay to appear in ChatGPT's answers?
Does GEO work for a small local business?
Do you need to rewrite your whole site?
How much does a GEO approach cost?
Sources
- [1] Pranjal Aggarwal et al., « GEO: Generative Engine Optimization », KDD 2024 (opens in new tab) ↩
- [2] Google Search Central, « AI features and your website » (opens in new tab) ↩
- [3] Vercel et MERJ, « The rise of the AI crawler », 17 décembre 2024 (opens in new tab) ↩
- [4] OpenAI, « Overview of OpenAI Crawlers » (opens in new tab) ↩
- [5] Anthropic, article d'aide sur les robots d'Anthropic (opens in new tab) ↩
- [6] Perplexity, « Perplexity Crawlers » (opens in new tab) ↩
- [7] MIT Technology Review, 1er juillet 2025 (opens in new tab) ↩
- [8] Directive (UE) 2019/2161 du 27 novembre 2019, dite « Omnibus » (opens in new tab) ↩