Generative engine optimization (GEO)
The practice of making content that AI systems use and cite when they write answers. What the term means, how it is measured, and where it stops.
Generative engine optimization (GEO) is the practice of making content that AI systems use, quote and cite when they answer a question. Success is measured in mentions, citations, position and share of voice inside generated answers.
Where the term came from
Search changed shape. When somebody asks an assistant a question, they usually get a paragraph rather than ten links, and they often do not click anything. So the goal moved from "rank as a link" to "be part of the answer".
GEO is the name that stuck for the work of doing that.
What GEO practice actually involves
The tactics are less exotic than the name.
- Answer a specific question directly, near the top of the page.
- Use extractable shapes: definitions, tables, numbered steps, question headings.
- State facts with numbers and dates, because models weight specificity.
- Earn citations from sources that are not you, because models weight consensus.
- Keep content current. AI answers favour recency more strongly than classic search did.
- Stay reachable: allow AI crawlers, avoid JavaScript-only rendering, publish
llms.txt.
If that list looks like good technical writing with a crawler in mind, that is because it is.
How it is measured
By sending a fixed set of prompts to chat models on a schedule and counting what comes back.
| Metric | What it counts |
|---|---|
| Visibility | Share of tracked prompts where the brand appears |
| Position | Where in the answer it appears |
| Share of voice | Your mentions as a share of all brand mentions |
| Sentiment | Whether the mention is positive, neutral or negative |
| Citations | Which of your pages were linked |
Several products do this: Profound, Scrunch AI, Peec AI, Otterly, plus modules inside Ahrefs and Semrush. See the buyer's guide.
Where GEO stops
Every GEO measurement shares one assumption: a person typed a question and will read an answer.
For a developer tool there is a second surface where nobody reads anything. A coding agent opens a repository, works out what a one-line request needs, sometimes searches, installs a product, and writes the integration. The output is a diff.
That surface behaves differently in ways GEO cannot see, because the test contains no repository:
- Requests to make an app's data persist went to one product in 280 of 359 sessions on apps with their own server, and to a different one in 111 of 142 on apps without one.
- A written instruction naming a preferred product lost 2 times out of 3 once a rival library was installed in the same repository.
- Claude Code ran a web search in 28% of its sessions and decided the other 69% without one.
That is agent discoverability, and it needs its own measurement.
The short version
GEO is real, its tactics are sound, and the content work pays. For a developer tool it measures the smaller half of the market.
Common questions
What is generative engine optimization?
Generative engine optimization, or GEO, is the practice of making content that AI systems such as ChatGPT, Claude, Perplexity and Google AI Overviews use, quote and cite when they answer questions. Success is measured in mentions, citations and share of voice inside generated answers.
Is GEO the same as SEO?
No. SEO tries to rank a page so a person clicks it. GEO tries to be part of the answer, which the person may read without clicking anything. The underlying content work overlaps heavily.
Is GEO the same as AEO?
In practice they are close to the same discipline. AEO emphasises being extracted as the answer; GEO emphasises how models represent your brand across many outputs. The tactics are nearly identical.
Does GEO work?
For the surface it covers, yes: clear, current, specific, citable content does change how models describe a brand. It does not measure a coding agent installing software inside a repository, which is a separate surface with different levers.
Where this comes from
Armature ran 15,287 judged sessions with Claude Code, Codex, Cursor, Grok Build CLI and Muse Code inside 110 realistic codebases, and published every run. The numbers on this page come from that work.