GEO SEO: getting cited by AI search engines

What is GEO (Generative Engine Optimization)?

GEO, short for Generative Engine Optimization, makes content understandable, reliable and reusable by the engines that write an answer instead of showing a list of links. It is also called “GEO SEO”, AEO (Answer Engine Optimization) or LLMO (Large Language Model Optimization). These terms refer to the same work, with no standard definition.

The term comes from a research paper published on November 16, 2023 by six researchers. It was accepted at the KDD 2024 conference (Aggarwal et al., “GEO: Generative Engine Optimization”).

This figure comes from an experimental protocol. It varies with the method and the domain, and does not predict the gain for a real website.

GEO and SEO: what are the differences?

GEO does not replace SEO. When an answer engine searches the web, it relies on a search index. What changes is the form of the result, and therefore what you measure.

Dimension	SEO	GEO
Target result	A position and a click	A citation or a mention in the answer
Unit of measurement	Impressions, clicks, position	Mention rate, citation rate, share of voice
Sources used	The ranked page	Several sources: website, comparison sites, reviews, forums, press
Stability	Ranking measurable day after day	Answers vary from one session to the next
Main levers	Technical, content, links	SEO foundation, citable content, brand awareness

The foundation stays the same: crawling, indexing, useful content, authority. Everything covered in our SEO guide also serves GEO. The point-by-point comparison, with what Google says about it, is in GEO vs SEO in detail.

Which AI answer engines should you track?

Google rolled out AI Overviews and AI Mode in France on July 22, 2026. The launch came two years late, because of neighbouring rights (Blog du Modérateur, July 22, 2026, in French). Any analysis of French organic traffic in 2026 must therefore separate the period before and after this date.

Engine	Type	What you can measure
Google AI Overviews	AI summary at the top of the results	Impressions and pages in Search Console
Google AI Mode	Conversational tab	Same, aggregated with AI Overviews
ChatGPT	Assistant with web search	No publisher report: prompt panel
Perplexity	Hybrid engine, sources displayed	Visible sources, manual tracking
Microsoft Copilot	Assistant backed by Bing	AI performance report in Bing Webmaster Tools
Gemini, Claude	Assistants with web search	No publisher report

Each engine cites its sources in its own way. How they work is compared on our page about AI search engines. Google’s case is detailed in AI Overviews and SEO.

How does an AI engine choose its sources?

A language model answers from two reservoirs. The first is what it learned during training, frozen at a given date. The second is what it retrieves on the fly from a search index. A little-known brand depends almost entirely on the second, and therefore on SEO.

1. Interpretation: the engine rephrases the intent behind the question.
2. Searches: it runs several related queries (Google calls this query fan-out for AI Mode).
3. Retrieval: it selects pages from its index.
4. Sorting: it isolates the useful passages and ranks them.
5. Generation: it writes the answer and attributes sources, or not.

For AI Overviews and AI Mode, Google says it relies on its usual ranking and quality systems (Google Search Central, AI optimization guide, updated on July 10, 2026). A page that does not rank therefore has little chance of being picked up.

How do you optimise your content to get cited?

The same Google guide states clearly what not to do. No chopping content into tiny fragments. No special writing “for AI”. No dedicated file such as llms.txt, which Google Search ignores. What helps is what already helps a reader in a hurry.

The signals that make a page citable

• a direct two-sentence answer at the top of the page, then the detail;
• sections that make sense on their own, with a heading that announces the question covered;
• figures with their source, date and scope;
• clear tables, lists and definitions, easy to extract;
• an identified author and a visible update date;
• a point of view of your own: a test, a case, a method that other pages do not have.

Example of rewriting a passage

A passage gets picked up when it answers a specific question on its own. Compare these two versions of the same idea:

Vague version	Citable version
“There are several ways to manage AI crawlers, each with its pros and cons depending on your strategy.”	“To stop appearing in ChatGPT search, block OAI-SearchBot in robots.txt. To refuse only the training of OpenAI’s models, block GPTBot and allow OAI-SearchBot.”

The citable version is not longer. It replaces a generality with names, an action and a consequence. This work, section by section, carries more weight than markup or an extra file.

Structured data remains useful for classic search. Its direct effect on AI citation has not been demonstrated (see our structured data guide). The full approach, content and brand, is described in the GEO optimisation method.

Why does the brand carry so much weight in GEO?

Being cited as a brand and being cited as a source are two different things. The first depends mainly on what the web says about you. The second depends on your pages and how accessible they are.

A correlation is not a cause, and the study comes from a tool vendor. It nonetheless guides priorities: specialist press, comparison sites, customer reviews, communities. Brand information must also stay consistent from one site to another.

Should you allow AI crawlers to crawl your site?

Each vendor uses several crawlers, with different roles. At OpenAI, OAI-SearchBot feeds ChatGPT search: a site that blocks it does not appear in those results. GPTBot collects content for model training. ChatGPT-User acts at a user’s request. For the latter, OpenAI states that robots.txt may not apply (OpenAI documentation on its crawlers, accessed on September 24, 2026).

Crawler family	Examples	Effect of blocking
Training	GPTBot, ClaudeBot, CCBot	The content is no longer used for future models
Answer engine index	OAI-SearchBot, PerplexityBot, Googlebot, Bingbot	No more citations as a source
User action	ChatGPT-User, Claude-User	The user can no longer have the page read

robots.txt is not enough. A firewall or a CDN can block these crawlers without anyone having decided to. The checking procedure is described on our page about GPTBot and AI crawlers.

How do you measure your visibility in AI engines?

Since August 31, 2026, Search Console has offered all sites a report dedicated to generative AI features. It shows impressions, pages, countries, devices and dates, but neither queries nor clicks (Google Search Central Blog, June 3, 2026).

Seer points out that the 0.94% is dragged down by a single account; without it, the figure is close to 1.61%. Being cited therefore limits the loss of clicks, without making up for it. Outside Google, measurement relies on a fixed prompt panel, checked every month on several engines. The metrics to track every month:

• mention rate: share of the panel’s prompts in which the brand is named;
• citation rate: share of prompts in which a page of the site is cited with a link;
• AI share of voice: the brand’s place among the competitors cited;
• AI impressions: the generative AI report in Search Console;
• AI traffic: visits from chatgpt.com, perplexity.ai, gemini.google.com or copilot.microsoft.com, and their conversions.

Metrics, tools and limits: see measuring your AI visibility.

Where do you start with GEO?

1. Access: robots.txt, firewall and CDN let the answer engines’ index crawlers through.
2. Baseline: 50 to 100 prompts representative of your customers, checked on ChatGPT, Gemini, Perplexity and Copilot.
3. High-performing SEO pages: these are the most likely candidates for citation.
4. Rewriting: answer at the top, self-contained sections, sourced figures.
5. Off-site brand: mentions, reviews, comparison sites, press.
6. Monthly measurement: same panel, same settings. A change of a few points proves nothing.

One last principle: nobody can guarantee a citation in ChatGPT or in an AI Overview. GEO increases a probability. It is managed alongside SEO, on a single roadmap. That is what the expression “GEO SEO” means: one team and one technical and editorial foundation. Only the metrics are split: clicks on one side, citations on the other.