What is AEO? A plain guide to answer engine optimisation

What does AEO mean?

AEO stands for answer engine optimisation. An answer engine gives people a written answer instead of a list of links. ChatGPT, Gemini, Perplexity, Copilot and Google’s AI Overviews all work this way.

Ask one of them “What’s the best help desk software for a small SaaS team?” and the answer usually names a few products and stops. AEO is the work of being one of those names, and of being described correctly when you are.

You’ll also see GEO (generative engine optimisation) and AI SEO. People draw small distinctions between these terms, but there’s no settled line between them. This guide uses AEO for all of it.

How is AEO different from SEO?

SEO aims to get your pages ranked in a list of results. AEO aims to get your brand named, correctly, inside the answer.

AEO builds on SEO and doesn’t replace it. Google says a page must be indexed and eligible to show with a snippet to appear as a link in AI Overviews or AI Mode, and that there are no additional technical requirements.

How do AI assistants decide what to cite?

An AI assistant passes three gates before it cites your page. It decides whether to search at all, since some answers come from what its model learned in training. It turns the question into its own searches, and your page has to come back in the results. Then it picks what to cite from those results.

Getting into the results seems to matter most. A University of Toronto preprint from October 2026 found that every page ChatGPT cited, across 22,315 URLs, was among those its own search had returned. Comparing pages on the same website, retrieval explained 32% of the variation in citations and page properties explained 2.8%. This is one correlational study, not yet peer reviewed, of API models in narrow markets, so treat it as the best evidence so far and not as settled.

Two page properties did hold up in that study: a title that closely matches the question, and one paragraph that closely matches it. For a writer, that points to something plain. Give each specific question a page or section that answers it directly, in words people would use to ask it.

Searches can also go beyond the question typed. Google says its AI features may run “multiple related searches across subtopics and data sources”, so a page can be used because it answers a narrower question behind the original one.

Crawler access matters too. OpenAI says sites that block OAI-SearchBot won’t be shown in ChatGPT’s search answers, while GPTBot is a separate crawler for training. You can block one and allow the other.

Why isn’t being cited the same as being named?

A citation links to your page. A mention puts your brand’s name in the answer. They often don’t come together. In a Semrush study of 3,981 appearances across 115 prompts, 61.7% cited a domain without naming the brand, 25.1% named it without citing it and 13.2% did both. Semrush sells a tool that measures this, and the sample is modest.

So a definition page, including this one, can earn citations without anyone hearing a brand’s name. Being named seems to depend more on other sites. In the Toronto study’s small B2B dataset of 55 queries, when a cited third-party page named a vendor, the odds that ChatGPT’s answer named that vendor were 39.7 times higher. In its larger consumer dataset, they were 4.2 times higher. An Ahrefs study of about 75,000 brands, published in May 2025, found branded web mentions correlated 0.664 with mentions in AI Overviews, and backlinks 0.218. Ahrefs sells SEO tools. Both findings are correlations, and larger brands naturally get more of everything.

The practical point is to look beyond your own site, and to earn mentions honestly. Google says that seeking inauthentic mentions “isn’t as helpful as it might seem”.

What have we seen in our own scans?

Measuring AI answers is easy to get wrong, and we’ve got it wrong ourselves. These are from our own early scans in October 2026.

• Our extraction invented mentions. When ChatGPT asked a clarifying question and named nobody, our software sometimes still recorded businesses from its own instructions. One pilot’s report put a brand at 9% visibility and first position after two answers that named nobody were recorded as naming it. Across our database at the time, 114 of 705 recorded mentions named a business the answer didn’t contain. We now keep a name only if it appears in the answer’s text. If you’re using any tool, including ours, spot-check its results against the answers themselves.
• Location can change the answer. A developer tool’s scans ran as if from Dubai, and its answers filled up with UAE providers, which said little about its global market. We now ask without a location unless a business has a specific market.
• One product can have many names. A single API product appeared under four spellings in one pilot. Unless those are merged, its share gets split and looks smaller than it is.
• Small wording changes can decide whether an answer appears. When we tested Google, “What is the best aesthetic clinic in Dubai Marina?” returned an AI Overview, and “Best aesthetic clinic in Dubai Marina?” didn’t.

How we know this: these come from small pilot scans and one test search, so they show what can go wrong and not how often. Quotable sells a tool that measures AI answers.

Which AEO advice doesn’t hold up?

Much popular AEO advice has weak evidence behind it, or has failed when re-tested. No one has published a randomised test on live pages in live engines, so none of this is final.

Clear headings, tables and facts are still worth having, because people scan and need the answer. Just don’t expect formatting alone to earn citations.

Where should you start?

Start by finding out where you stand, before changing anything. You can do this without paying for a tool:

• Run the AI readiness checker and the AI crawler checker to catch access problems.
• Use the AI question generator to list questions customers might ask in your category, then correct them from what you hear in sales calls.
• Ask each question several times in ChatGPT, Gemini and Google. Note who’s named, how you’re described, what’s wrong and which sites are cited.
• Fix the clearest gap. That might be a blocked crawler, a question with no page on your site, or a review site where rivals appear and you don’t.
• Ask the same questions again, and compare.

Doing this by hand gets slow once you have many questions. Quotable, which we make, runs your questions on each engine on a schedule and tracks the results. Other tools do this too, so compare them on how they count and whether you can read the answers behind each number.

How should you report AEO to your team?

Report AEO as shares across many runs, not as ranks or single answers. Answers vary from run to run, so a single check tells you little: an Ahrefs study published in October 2026 found 50% to 83% of cited sites had gone by the next day, depending on the assistant (about 50% to 60% once answers citing nothing are left out), though about 75% of those that dropped out came back within 20 days.

A report someone can defend has four features:

What can’t AEO do?

Nobody can guarantee that an assistant will name you, and that includes us. You don’t control the answer. You can improve what assistants are able to find and read, earn mentions elsewhere, and measure what happens.

AEO also won’t fix a weak position. If people rarely talk about your product, or a rival really does suit the question better, no page will change that for long.