Generative engine optimisation (GEO) is the practice of improving how a brand shows up in answers written by AI engines: whether it is named, how it is described, and which sources the engine relies on when it talks about it. It covers ChatGPT, Gemini, Perplexity, Claude, Microsoft Copilot, and Google's AI Overviews and AI Mode.
The term comes from a 2023 research paper, GEO: Generative Engine Optimization, which showed that the way content is written and sourced changes how often generative engines cite it. Since then GEO has moved from research topic to marketing budget line. You will also see it called answer engine optimisation (AEO), AI search optimisation or LLM optimisation. They describe the same job.
This guide explains what GEO is, how it differs from SEO, and what the GEO Compass benchmark of more than 900,000 AI answers shows about how it works in practice.
Why GEO matters now
Two years ago many AI answers came from a model's training data alone. That has changed. On a fixed panel of long-running studies in the GEO Compass data, the share of ChatGPT answers that ran a live web search rose from 35% in September 2025 to more than 95% a year later. Gemini went from 15% to over 80%. Perplexity and Google's AI surfaces were already searching almost every time.
GEO Compass fixed panel of long-running studies, September 2026. See the Trends page for the monthly series.
So nearly every AI answer is now built from pages the engine fetches at the moment of the question. What sits on the open web, who wrote it and how clearly it is structured shapes what the engine says about you. That is what GEO works on. See the Trends page for the full series.
GEO vs SEO: what is different
SEO and GEO share foundations. Engines still need to crawl, index and trust your pages. But the goal and the levers differ.
| SEO | GEO | |
|---|---|---|
| Goal | Rank in a list of links | Be named, described accurately and cited inside the answer |
| Unit of success | Position and clicks | Share of answers that name you (share of model), position in the answer, sentiment, accuracy |
| Main lever | Your own pages and backlinks | Your own pages plus the third-party sources engines cite: earned media, reviews, reference sites, video, communities |
| Query style | Short keywords | Long, conversational questions, often with follow-ups |
| Volatility | Algorithm updates | Model updates, which can change cited sources overnight |
The biggest practical difference is where the work happens. In the GEO Compass data, company and brand websites supply 31% of the domains AI engines cite. The other 69% is earned media (20%), academic, institutional and government sources, social and community platforms, reference sites such as Wikipedia, and marketplaces. You can't optimise most of the answer by editing your own site.
Brands are named, but rarely cited, in category questions
GEO is hardest where it matters most: questions that don't name your brand.
When someone asks about a brand by name, AI engines cite that brand's own website in 73% of answers with sources. When they ask a category question, such as "best travel insurance for families" or "which EV has the longest range", the brand the question set was built around is named in 27% of answers, and its own website is cited in only 12%.
Share of answers with sources that cite the focus brand's own site. GEO Compass benchmark, all markets.
In other words, most category discovery happens on other people's pages. That is why GEO is as much a PR, content and reputation discipline as a technical one. Read more in Branded vs unbranded.
Every engine works differently
There is no single "AI search". Engines differ in how many sources they cite and which platforms they favour:
- Google AI Mode cites about 10 distinct sites per answer; Gemini and Microsoft Copilot fewer than 5. With fewer sources, each citation carries more weight.
- YouTube appears in 38% of Google AI Overviews answers with sources, and 37% of AI Mode answers, but only 4% on ChatGPT and under 1% on Claude.
- LinkedIn shows up in 16% of Perplexity answers. The OpenAI API cites Wikipedia in 24% of answers.
The engine comparison covers this in detail. The practical point: track and optimise engine by engine, and by model version, because a model update can change which sources are cited from one week to the next.
And every market works differently
Where a question is asked changes the answer. In the GEO Compass benchmark, earned media supplies about twice its global share of citations in Australia. Wikipedia is cited about four times as often as average in Malaysia. Marketplaces and aggregators carry three times their usual weight in India. Questions asked in local languages pull in local publishers. A GEO plan for Asia-Pacific needs a market-by-market view of the sources that matter.
What a GEO programme involves
A GEO programme usually runs in three phases:
- Audit. Track the real questions customers ask across the AI engines that matter in your markets. Measure how often you are named, how you are described, where you rank in the answer and which sources are cited, for you and your competitors.
- Update core assets. Fix the facts AI engines take from you: your website's product, comparison and FAQ pages, structured data, messaging, press materials and executive bios. Make them easy to extract and consistent across every place your brand appears.
- Activate and optimise. Work on the third-party sources engines rely on for your category: the publishers, reviewers, creators, communities and reference sites. Then measure monthly and adjust, because engines and models change.
Our GEO checklist breaks this down into practical steps.
How to start
Begin with measurement. Pick 50 to 200 questions your customers actually ask, across branded and unbranded intents, in the markets and languages you serve. Run them across the engines your audience uses, repeatedly, and look at who is named and which sources are cited. Then compare with your category in the GEO Compass explorer to see what is typical.
Frequently asked questions
- What does GEO stand for?
- GEO stands for generative engine optimisation: improving how a brand appears in answers written by generative AI engines such as ChatGPT, Gemini, Perplexity, Claude, Microsoft Copilot and Google's AI Overviews and AI Mode.
- Is GEO the same as AEO?
- Mostly. Answer engine optimisation (AEO) and AI search optimisation describe the same goal: being named, described accurately and cited inside an AI answer. GEO is the term most used in research and by agencies.
- Does GEO replace SEO?
- No. AI engines now run a live web search for almost every answer, so pages that rank and can be crawled are still the raw material. GEO adds work on the sources AI engines rely on beyond your own site, such as earned media, reviews, reference sites, video and community platforms.
- How do you measure GEO?
- By tracking a fixed set of real customer questions across AI engines over time, and measuring how often your brand is named, how it is described, where it ranks in the answer and which sources are cited. Share of model, answer position, answer sentiment and answer accuracy are the usual KPIs.
Sources
- GEO: Generative Engine Optimization (Aggarwal et al.), arXiv / KDD 2024
- AI features and your website, Google Search Central
Figures marked as GEO Compass data come from the anonymised GEO Compass benchmark of AI answers collected by We AI & Data Advisors. Related: AI search in Australia · AI search in Singapore · AI search in Malaysia · AI search in Indonesia · AI search in India · AI search in Europe