AnswersAnswer engine optimisation

What answer engine optimisation is, and how to tell whether it is working

Answer engine optimisation (answer engine optimization, in US spelling), or AEO, aims at the answer an AI assistant gives rather than at a ranked page. Much of it is search work a marketing team already knows, and the rest happens on pages the brand does not own. What is new is how you find out whether it worked: by reading the answers themselves, more than once, before the work and after it.

What is answer engine optimisation (AEO)?

Answer engine optimisation is the work of getting a brand named, and described accurately, when somebody asks an AI assistant a question the brand should be the answer to. An answer engine is anything that replies with a written answer, often citing a few sources, rather than a ranked list of results: a chat assistant, an AI summary on a search results page, or a voice assistant.

The target is the answer, not the page. Somebody asks which accounting software suits a small firm, or which running shoe suits a flat foot, and the assistant often replies with a few names. AEO is the work on whatever influences whether your brand is one of those names, and whether what is said about it is true.

A brand left out of the answer is left off the shortlist, and a buyer who never saw the name has no reason to go looking for it afterwards.

Where a brand stands in those answers is its AI visibility, and Delphi measures it. It puts the questions a brand's buyers ask, drafted for the brand and edited by its team, to ChatGPT, Gemini, Claude, Perplexity and Grok, stores every answer word for word, and reads each one for whether the brand was named, how it was described, and whether it was the one recommended.

Is AEO different from generative engine optimisation (GEO)?

AEO and GEO are not different in practice: both mean getting a brand into the answers AI assistants give, and where a vendor splits them, it is splitting one job into parts.

Where each name came from, and the other acronyms in circulation, are set out on our page about generative engine optimisation.

Is AEO just SEO under another name?

AEO is partly SEO under another name: the groundwork is shared, from pages a crawler can fetch to facts stated plainly on them, but the target is often a page the brand does not own, and success is read from the answer rather than from a rank. Google says there are no additional requirements or special optimisation for its own AI features beyond the SEO best practices that already apply, as long as the site has not been excluded from Search's generative AI features in Search Console; no site is excluded by default.

Google says the best practices for SEO remain relevant for AI Overviews and AI Mode, that there are no additional requirements to appear in them, and that a supporting link has to be indexed and eligible to appear in Google Search with a snippet (Google Search Central), as long as the site has not been excluded from Search's generative AI features in Search Console; no site is excluded by default. Google's statement covers its own features only; other assistants run crawlers of their own, and our page on GEO versus SEO sets out what carries over from SEO and what changes.

What does AEO work actually involve?

AEO work comes down to four jobs: letting the crawlers that feed answers read your pages, writing passages that answer one question completely on their own, making the brand one clear entity with the same facts everywhere, and earning accurate mentions on the third-party pages assistants read and cite.

Be readable. An assistant can only find, quote and describe your pages from its search index if the crawler that builds the index was allowed to fetch them, both by your robots.txt and by any CDN or hosting layer in front of the site, which is the first item on Google's list of SEO fundamentals for its AI features. A blocked page can at most turn up there as a bare link, as both Google and OpenAI say of their own crawlers. Several operators run that crawler separately from the one that collects training data, so refusing the wrong token can keep your own pages out of an assistant's answers; which crawlers to allow has its own page. Content that appears only after a script runs is a second gap, because not every crawler runs JavaScript, so Delphi reads your home page as a crawler that does not run JavaScript would.
Answer in passages. A question is not always searched for whole. Google says AI Overviews and AI Mode may issue multiple related searches across subtopics to build one response, so a page can be retrieved for one part of a buyer's question. Each section should answer its own question in its first sentence, with the specifics a buyer weighs, such as price, compatibility and who it suits. A paragraph that only makes sense after the one above it is hard to lift out and use.
Be one clear entity. The name in a buyer's question, the company on a review site and the brand on your home page need to read as one organisation. State the brand's names, including any legal or trading name, in the structured data on your own site, link it to your profiles elsewhere, and keep the facts identical in every place. Google lists making structured data match the visible text on the page among those same fundamentals. Delphi checks whether your home page declares the organisation and links it to profiles elsewhere.
Be described well elsewhere. Assistants also read and cite pages that belong to somebody else: comparison articles, reviews, community threads, retailer listings. Delphi records which publishers are cited in answers to buying questions where you were not the recommendation. Sources are only observable where an answer discloses them, so that list is a partial view and says so. A publisher can be written to and a thread answered under your own name, but neither owes you a mention, and Delphi does not draft a post for anybody to pass off as their own.

What can AEO not do?

Answer engine optimisation cannot control what other sites say about a brand, and it cannot make an assistant give the same answer twice. In ChatGPT, OpenAI states that its ads do not influence the answers, so a placement there is bought below the answer rather than inside it.

Other sites are influenced, not controlled. Your own pages are yours to change. A publisher's review, a retailer's listing and a community thread belong to somebody else: you can correct, contribute and persuade, and the owner decides.
The same question does not get the same answer. In Delphi's study of 3,116 answers, the most recent measured on 24 August 2026, an assistant asked the same question more than once inside one measurement returned the same set of names 21.5% of the time. One check before the work and another after it are two draws from something that moves on its own, and they cannot show whether the work did anything.
An ad is not a recommendation. OpenAI says its ads in ChatGPT appear below the answer and do not influence it, and our page on being recommended by ChatGPT sets out what an ad in ChatGPT cannot buy. Google works differently: it says Text and Shopping ads can show within AI Overviews in English in a set of countries that includes the US, as well as above and below them. Either way paid placement is a separate budget, and adding it to what the organic answers say counts different things as one.

How do you know whether AEO is working?

You know AEO is working by reading the answers, not the traffic reports: put the same buying questions to the same assistants, more than once each, before the work and again after it, and count a change only when it is larger than the answers vary on their own.

Search Console's generative AI report counts how often links to your pages were shown in Google's AI features, and Bing Webmaster Tools' AI Performance preview counts how often your pages were cited across Microsoft Copilot and Bing's AI summaries, but neither says whether the answer recommended you, a rival or nobody. So the direct measure is the answer: the same buying questions put to the same assistants, more than once each, before the work ships and after it. Our tracking page sets out how to tell whether a change is real. Delphi runs that measurement, asking ChatGPT, Gemini, Claude, Perplexity and Grok each question more than once, and its dashboard and history page report a movement in the Index smaller than the spread between the assistants as no confirmed change. Some of those assistants search the web as they answer and the others answer most questions from what their model has already learned, so a change to your pages or crawler rules shows first where an assistant searches as it answers, and elsewhere only once a newer model has learned it.