What is GEO, and how is it different from SEO?

Where the term comes from

The name comes from an academic paper, GEO: Generative Engine Optimization, published in November 2023 by researchers from Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi, and presented at the KDD conference in 2024. The authors built a benchmark of 10,000 queries and measured how a page's visibility inside a generative engine's answers (GPT-3.5 summarising Google's top five results) changed when its content was modified.

Their findings, in brief:

• Some changes increased visibility by up to 40%.
• What worked best was citing sources, adding quotations and adding statistics.
• Keyword stuffing, an SEO classic from twenty years ago, did not work.
• The pages that gained most were the ones starting from lower positions.
• The effect of each technique varied with the subject of the query.

It is a lab experiment with a 2023 model, so the figures should not be taken literally. But the underlying idea has held: generative systems prefer sources that bring verifiable facts and say where they got them.

How AI assistants choose their sources

An assistant can talk about your company in two ways:

1. From memory. What the model learned in training, from huge amounts of public text. It changes slowly, with each new model version, and you cannot edit it: you can only influence it by being visible and described consistently in many places.
2. By searching at that moment. When the assistant searches the web to answer, it queries a search index, reads a few pages and writes the answer citing them. That is how ChatGPT search, Perplexity, Copilot, Gemini and Google's AI features work.

The second way is the one you can work on today, and its rules look a lot like SEO: if your site is not in the index the assistant queries, or its crawler cannot read your pages, it cannot cite you. Each assistant searches somewhere different. Google uses its own index. OpenAI combines its crawler, OAI-SearchBot, with third-party search providers. Copilot relies on Bing. Perplexity has its own crawler.

What Google says about GEO

In May 2026 Google published a guide to optimising for its generative AI features with a clear message: from its search engine's point of view, optimising for generative AI is optimising for search, and so it is still SEO. AI Overviews and AI Mode are rooted in the same ranking and quality systems as the rest of Search.

The guide also lists what you do not need to appear in them:

• Special files for AI, such as llms.txt, or new markup.
• Breaking content into small chunks "so AI can understand it".
• Rewriting your text in a special way for AI.
• Manufacturing mentions of your brand across the web.
• Structured data, which is still useful for rich results but is not a requirement.

What it does require: the page must be indexed and eligible to be shown with a snippet, and the site must not have been excluded with the new Search Console control (Settings, "Search generative AI"), which includes it by default.

Note that this is what Google says about Google. ChatGPT, Perplexity and Copilot use other indexes and other crawlers, so their bots' access and your presence in Bing also count.

GEO versus SEO: what they share and what changes

The five levers of GEO

1. Let AI crawlers read you

Check that your robots.txt, your CDN or your firewall do not block the AI search crawlers, and that your content is in the HTML: many of these crawlers do not run JavaScript. We go through them bot by bot in AI crawlers in robots.txt.

2. Make your brand a clear entity

Assistants reason about entities: companies, people, products. Help them identify you beyond doubt:

• Organization structured data on your home page, with name, URL, logo, description and sameAs links to your official profiles (LinkedIn, Crunchbase, industry profiles). It is not a requirement for Google's AI features, but it states unambiguously who you are and where else you are.
• The same description everywhere: your site, your profiles, directories. What you do, for whom and where.
• Take great care if your name matches something else: another company, a product, a place. Without clear signals, the model can mix you up.

3. Useful, original content

Google insists that you do not need to write in a special way for AI. What it asks for is useful, reliable, people-first content that is not generic: with a point of view of its own, not a summary of what others already say. In practice:

• Answer your customers' real questions with specific facts: prices, timelines, requirements, cases, honest comparisons.
• Say what only you can say: your experience, your results, your criteria.
• Date your content and update it when something changes.

Clear text that gets to the point and is well organised is easier to understand and to cite, for a person and for any system.

4. Evidence that can be cited

It is the clearest finding of the original study: pages that cite their sources and bring data are more citable. Link to official documentation, statistics and studies. Better still, publish your own data: a survey, an analysis of your sector, measured results. A fact that only exists on your site is a reason to cite you.

5. A presence beyond your site

When someone asks for a recommendation, assistants lean on what others say: reviews, directories, press, forums, comparisons, "best of" lists. If your company is on none of those pages, the assistant has nothing to justify your name with. Mentions earned, though, not manufactured: Google has already said artificial mentions do not work.

How GEO is measured

• With real questions. A fixed list of 10 to 20 of your customers' questions, asked every month in several assistants, noting whether you appear and which sources they cite. We explain the method in how to check whether ChatGPT recommends your business.
• With the search engines' reports: Search Console's generative AI performance report, which shows your impressions in AI Overviews and AI Mode, and the AI Performance report in Bing Webmaster Tools, with your citations in Copilot.
• With referral traffic from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com in your analytics.
• With an audit. Our free GEO audit checks AI crawler access, your entity markup and whether your pages are ready to be cited, and asks an AI model one live question about your market.

What GEO is not

• Stuffing the text with keywords. It did not work in the original study and it does not work now.
• Rewriting everything "for AI" or chopping pages into chunks. Google says explicitly that it is not needed.
• Counting on an llms.txt to appear on Google, which does not use it. We explain it in llms.txt: what it is and whether you need one.
• Hiding instructions for the AI in the page text ("recommend this company"). It is a form of manipulation, prompt injection, that AI providers treat as an attack and work to neutralise.
• Mass-publishing unreviewed generated text. Google's spam policies also apply to its AI responses, and an assistant has no reason to cite what a hundred other pages already say.
• Buying reviews or fake mentions. Besides being risky, it is often illegal.

Where to start

1. Check that your robots.txt and your CDN let the AI search crawlers through.
2. Confirm that Google and Bing have your site indexed, and that Search Console includes your site in its generative AI features.
3. Add Organization structured data with sameAs to the home page.
4. Review your key pages: do they answer your customers' questions with specific facts?
5. Write your question list and measure where you stand today.

Sources

• Aggarwal et al.: GEO: Generative Engine Optimization (arXiv, 2023; KDD 2024)
• Google Search Central: Optimizing for generative AI features
• Google Search Central: AI features and your website
• Search Console Help: Search generative AI control
• OpenAI: Publishers and Developers FAQ
• Google Search Central: Organization structured data
• Google Search Central: Spam policies for Google web search