Generative engine optimization: what you can measure

Generative engine optimization, or GEO, is the work of making your content easy for AI answer tools to find, understand and cite. Answer engine optimization, or AEO, is the same idea with a narrower aim: being the source an answer engine quotes when someone asks a question.

The terms are new and the field is unsettled. Nobody outside these products knows exactly how they pick sources. So this guide sticks to two things: what you can measure, and what good page hygiene looks like.

What GEO is not

It is not a switch. No tag, file or phrase makes ChatGPT, Perplexity or Gemini name you. Anyone who promises a ranking inside those answers is guessing. Answers change between runs, between users and between model versions.

That makes measurement more important, not less. You cannot control the answer, but you can watch it.

What you can measure

• Whether an answer to a buyer question names your brand at all.
• Which other names appear in the same answer.
• Which pages the answer cites, and whether yours is among them.
• Which source types show up: reviews, forums, news, docs, comparison sites.
• How these change when you repeat the same question over several weeks.

Run each question more than once. A single answer is an anecdote. A set of 25 questions asked on a schedule you choose starts to show a pattern.

A checklist for your pages

These are habits that help any reader or crawler, human or machine. None of them guarantees a citation.

• Write a clear definition near the top of a page. One or two sentences that say what the thing is.
• Use plain headings that match the questions people ask.
• Add structured data (JSON-LD) that matches the visible content: Article, FAQPage, BreadcrumbList.
• Add a short FAQ with real questions and direct answers.
• Publish an llms.txt file that maps your important pages in one place, and offer a plain markdown version of key pages.
• State facts with a source and a date, and update them when they change.
• Make sure crawlers you want are allowed in robots.txt, and that pages load without needing a login.
• Get mentioned where your buyers look, such as reviews, directories and communities, with accurate descriptions.

Let an agent run the checks

Checking 25 questions by hand across three tools is slow. An agent can do it. The AI answer visibility check recipe asks ChatGPT, Perplexity and Gemini your buyer questions, then reads the pages they cite. It reports which answers name your brand and which name others.

The recipe lists the providers it can use for each step. Your agent searches the looot catalog, reads what each endpoint takes and returns, and shows the price before it spends. Compare results by date so you can see change.

The last line of that prompt is the useful part. The cited pages tell you what the engine treated as a source, which is a better brief for your next page than any guess.

What to do with the results

Look for questions where rivals appear and you do not. Open the cited pages and ask what they have that yours lacks: a clear definition, a comparison table, a recent date, a data point. Write or fix one page, then run the check again later. Keep the changes small so you can tell what moved.

Treat the checks as a log, not a score. A row that says you were named once in four runs is information. It is not a ranking.

For the data layer behind these checks, see what is an MCP server and MCP vs API for agents. To keep the results in a database you own, read the build-your-own page.

Open the AI answer visibility recipe