What is generative engine optimization (GEO)?
Generative engine optimization (GEO) is the work that gets a business named, and ideally recommended, in the answers written by generative AI engines: ChatGPT, Gemini, Claude, Perplexity and Google AI Overviews. It is measured by asking the engines what buyers ask and reading what they answer, because no console reports it.
- GEO is measured in answers: named, position, recommended, cited. Not in rankings.
- The term is from a 2023 paper (KDD 2024) that reports visibility gains of up to 40% and says what works varies by domain.
- In our data a business is named in 35.2% of answers, recommended in 20.6%, and its site is cited in 0.6%: the engines read reddit.com, en.wikipedia.org, youtube.com, linkedin.com, github.com.
- Google: AI Overviews and AI Mode need no special optimization, files or markup. The fundamentals apply.
What is GEO?
When a buyer asks ChatGPT which accountant to call, the answer is a short piece of text that names a few firms and often picks one. Generative engine optimization is everything that makes your firm one of the names, and the one picked. It is not about ranking a page; the answer has no page two. It is about what the engine knows about you, where it learned it, and whether it trusts it enough to say your name.
Four things can happen to a business in a generated answer, and GEO counts all four: it is named; it sits at a position among the names; it is presented as the recommendation; its own site is cited as a source. The first and the third are the ones buyers act on.
Where the term comes from
The phrase was introduced in November 2023 by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande in a paper titled "GEO: Generative Engine Optimization", later presented at the KDD 2024 conference. The paper proposes a way to measure visibility in generated answers and tests changes to pages against a benchmark of queries.
GEO can boost visibility by up to 40% in generative engine responses. [...] The efficacy of these strategies varies across domains, underscoring the need for domain-specific optimization methods.
Aggarwal et al., GEO: Generative Engine Optimization, abstract
Two things from the paper survive every later study: visibility in generated answers can be moved, and no single trick moves it everywhere. What works for a restaurant is not what works for a software vendor.
How generative engines build an answer
The chat engines and Google's AI features do two things when you ask a buying question: they search the web, in their own way, and they write. The search decides which pages the answer can draw on; the writing decides who gets named. GEO works on both. Google's documentation says the search part has no special requirements: a page needs to be indexed and eligible for a snippet, and no new files or markup are needed. The chat engines publish no requirements at all.
What the engines actually draw on is measurable, and it is not the business's own site. Across our answers, the sources cited most are reddit.com, en.wikipedia.org, youtube.com, linkedin.com, github.com. The 2026 citation studies find the same from much larger samples, and add that the engines barely overlap with each other: about 11% of domains are cited by both ChatGPT and Perplexity.
What the answers look like, measured
| Engine | Answers | Business named | Recommended | Own site cited |
|---|---|---|---|---|
| Perplexity | 1,777 | 32.2% | 20.1% | 0.6% |
| ChatGPT | 1,776 | 31.8% | 21.6% | 0.3% |
| Claude | 1,687 | 33.8% | 18.8% | 0.7% |
| Gemini | 1,543 | 40.7% | 26.1% | 0.5% |
| Google AI Overviews | 769 | 42.4% | 12.7% | 1.2% |
7,552 answers, 35 businesses, 24 August to 7 October 2026. Full tables, by country and by cited platform, on the data page.
Three facts to build on. Being named is common and being chosen is not: the gap between 35.2% and 20.6% is where GEO earns its keep. The engines name on average 7 brands per answer, and when a business is named it is first in 45.1% of those answers. And the engines disagree: asked the same question on the same day, two of them name the same business first 26 to 35.7% of the time.
GEO vs SEO
| Row | SEO | GEO |
|---|---|---|
| Goal | Rank a page for a query | Be named and recommended in a generated answer |
| Unit | Position, impressions, clicks | Named, position among names, recommended, cited |
| Where it shows | Result lists | ChatGPT, Gemini, Claude, Perplexity, AI Overviews, AI Mode |
| Decided by | Crawlability, relevance, links, content quality | What the engine finds about you across the web, and trusts |
| Measured with | Search Console, rank trackers | Asking the engines daily and reading the answers |
| Changes | Weeks to months | Day to day; use a rolling average |
Same raw material, different scoreboard. The full comparison, with AEO and AIO, is in its own guide.
SEO is the floor: a page that cannot be crawled cannot be cited. GEO is what happens above the floor, and it is mostly about pages that are not yours. Read GEO vs SEO vs AEO for the side-by-side with answer engine optimization.
How to do GEO
- Measure first. Write the ten to fifty questions your buyers ask, put them to every engine, read the answers. Without this there is nothing to optimise against.
- Find the sources. Every answer lists what it used. Those pages, reviews, directories, forums, articles, are where your name has to be. In our data they are reddit.com, en.wikipedia.org, youtube.com, linkedin.com, github.com.
- Fix your own site for the questions you lose. One page per question, the answer first, evidence after; structured data that describes you accurately; a crawlable site. Google needs nothing more, and the chat engines read the same pages.
- Earn the mentions. Reviews on the platforms of your trade, a complete presence on the directories the engines cite, comparison articles that name you with the facts right. Named is 35.2%; cited is 0.6%. The mentions come from elsewhere.
- Re-measure, daily. Answers move from day to day. A rolling average over four weeks shows the trend; one check shows noise.
GEO tools and services
A GEO tool asks the engines and scores the answers. The measures that matter are per question, per engine, per day: named, position, recommended, cited, sources used, competitors named. Citeon AI records all of them across five engines, keeps every answer, and turns each losing question into a fix card with the file or copy ready. GEO services do the applying and the content for you: what the Optimize plan includes.
What to measure
| Measure | Question it answers | Baseline in our data |
|---|---|---|
| Named rate | In what share of answers does the engine mention you? | 35.2% |
| Position | When named, where among the brands? | First in 45.1% of named answers |
| Recommended rate | In what share of answers are you the pick? | 20.6% |
| Own site cited | In what share of answers is your site a source? | 0.6% |
| Cited sources | Which pages did the engine use? | reddit.com, en.wikipedia.org, youtube.com most |
| Competitors named | Who does the engine name next to you, or instead of you? | 7 brands per answer on average |
Start with your own numbers: the free audit puts your buyers' questions to three engines and emails the answers, usually within 5 to 10 minutes.
Questions people ask
What does GEO mean?
Generative engine optimization: improving how visible a business or a page is in the answers that generative AI engines write. The term was introduced in a 2023 paper by Pranjal Aggarwal and colleagues, presented at KDD 2024. It is unrelated to geographic targeting, which marketers also shorten to geo.
Is GEO different from SEO?
Yes in what it measures, no in what it starts from. SEO is measured in rankings and clicks; GEO in whether a generated answer names and recommends you, and which sources it used. Both start from a crawlable site and what the web says about you. Google says its AI features need no special optimization beyond the standard fundamentals.
Does GEO work?
The paper that coined the term reports visibility gains of up to 40% in generative engine responses from its methods, and notes that what works varies by domain. In our own measurement of one restaurant in The Hague, the share of answers naming it went from 4.7% in June 2026 to 57% in August 2026, with fixes measured one at a time. One business, not a benchmark.
What are generative engine optimization tools?
Software that asks the engines the questions buyers ask, scores every answer (named, position, recommended, cited), lists the sources used and the competitors named, and, in Citeon AI's case, turns each gap into a fix on the site. Rank trackers cannot do this because there is no rank to track.
What are GEO services?
The same work done for you: daily measurement, fixes applied to your site with your approval, content that answers buyer questions. Citeon AI's Optimize plan is one, at 199 euro a month; details on the services page.
Does llms.txt or special markup help GEO?
Google says no new files or markup are needed for AI Overviews and AI Mode. For the chat engines there is no published requirement either. What the engines read, in our data, is third-party pages far more than the business's own site (cited in 0.6% of answers), so presence on those pages matters more than any file.
Sources
- Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande: GEO: Generative Engine Optimization, KDD 2024 (arXiv 2311.09735, November 2023). "GEO can boost visibility by up to 40% in generative engine responses"; "the efficacy of these strategies varies across domains"
- Wikipedia, Generative engine optimization
- Google Search Central, AI features and your website. no additional requirements; no new files or markup
- Google Search Central, Creating helpful, reliable, people-first content
- 5WPR, The state of AI citations 2026. about 11% of domains cited by both ChatGPT and Perplexity; vendor sites largely absent from category top tens
- Citeon AI measurements, 24 August to 7 October 2026. 7,552 answers, 35 businesses, five engines