An answer engine does not rank you. It describes you, in a sentence, to somebody who may never see your website. Everything else follows from that one difference.
Including, unfortunately, most of the advice being sold about it.
What is answer engine optimisation?
Answer engine optimisation (AEO, spelled answer engine optimization in the US) is the practice of getting your business named, described accurately and recommended inside the answers that AI systems generate – Google’s AI Overviews and AI Mode, ChatGPT, Perplexity, Copilot. Success is not a position in a list. Success is being retrieved, selected and cited, and having what the answer says about you be true. It is a discipline applied to assets you already own, not a channel you buy separately.
You will also see GEO – generative engine optimisation – used for the same thing. Treat them as synonyms. You will see LLMO, AIO and a few others besides; those are vendors trying to own a category name, and nobody is typing them into a search box.
How is AEO different from SEO?
Four ways that actually matter. The unit of success is a description rather than a position. There is no fixed results page to audit, because two people asking the same question get different answers. The competitive set is whoever the model retrieves, not whoever ranks – you can lose to a source that has never appeared above you in Google. And a factual error about you propagates as a statement about your business, repeated confidently, which ranking never did.
That last one is the part operators underestimate. If three pages of your own site give three different answers about what is included, an answer engine will pick one and assert it. I have watched a model tell a prospective guest something about a trip that had not been true for two years.
How do answer engines pick what to cite?
Mostly by retrieving from a search index and then grounding the generated answer in what it pulled. That means classical crawlability and indexability are still the gate: if a crawler cannot reach the page, nothing downstream matters.
Which crawler is a question worth getting right, because the names are not interchangeable. OpenAI alone runs four, and they do different jobs: OAI-SearchBot surfaces you in ChatGPT search, GPTBot is for training, ChatGPT-User fetches a page because a user asked for it, and there is a separate ads crawler. Blocking GPTBot does not remove you from ChatGPT’s search results, and blocking the wrong one removes you from the surface you wanted to be on. This is the most common self-inflicted wound in the whole category.
Ranking and citation are correlated, and the correlation is weakening. Ahrefs found 76% of cited pages were also ranking in the top 10 in mid-2025; by early 2026 the same analysis put it at 38%. Rank well and you are more likely to be cited. Rank well and you are not guaranteed to be, and you can be cited without ranking at all.
What actually moves citation, and what is folklore?
The best evidence available is a peer-reviewed study across roughly 10,000 queries that tested what changes to a page did to its visibility in generated answers. Adding cited sources, direct quotations and specific statistics raised source visibility by up to 40%. Keyword stuffing did nothing. Read that carefully: 40% is the ceiling across nine tested methods, measured on the researchers’ own visibility metrics against 2023–24 engines, not a lift you should expect. It is the best evidence available and it is still thin. Anyone quoting it as a reliable number is doing the thing this article is about.
The rest of what works is unglamorous: one canonical set of facts about your business stated identically everywhere it appears, genuine depth on the pages that describe what you sell, and third-party corroboration – models trust agreement, so the same facts repeated by sources that are not you carry weight.
Now the folklore. There is no AEO schema: Google’s own guidance says no special markup or machine-readable files are required and that Search ignores them. The llms.txt file is a proposal nobody adopted – Google’s own search advocate compared it to the meta keywords tag, which should tell you enough. And any product selling you a guaranteed citation, or an “AI visibility score” presented as though the engine publishes one, is selling you a number it invented.
How do you measure it with no rank and often no click?
Imperfectly, and you should be suspicious of anyone who says otherwise. Google added a generative AI performance report to Search Console in June 2026, and it is genuinely useful, but it reports impressions only – no clicks, no click-through rate, no queries. That is the ceiling on what the platform will tell you.
So you triangulate. Segment assistant referrals in analytics, knowing they will be a small share of sessions. Run a fixed set of prompts across the main assistants every month and log two things: whether you were named, and whether what was said about you was accurate. And watch branded search and direct sessions, because influence from an answer usually surfaces later as somebody looking you up by name. Pew’s work is the clearest picture of the click side: with an AI summary present, people clicked a result on 8% of visits versus 15% without, and only 1% clicked a link inside the summary. Plan for influence you cannot fully attribute.
Where does AEO sit in a growth programme?
Inside the SEO and content work, not beside it with its own invoice. Every fix above – clearer facts, better structure, real depth, corroboration from elsewhere – is a fix classical SEO has asked for since long before anyone said “answer engine”. That is the honest version of AEO: it is not a separate channel to buy, it is the discipline of being describable, and it pays twice. You get cited, and the same work lifts conventional organic performance.
Which gives you something free to do this week. If you are already paying for SEO and someone has quoted you extra for AEO, ask them what is in that line that was not already their job. Citation measurement and entity clean-up are genuinely extra work and a good answer will name them. A bad answer will describe schema markup and better content, which you were already buying. That single question is worth more than most of what is written about this, including, probably, this.
“If a product promises you a citation, it is selling you a number it invented.”
If you want the applied version of this rather than the definition, we wrote up what it looked like on one account – the four things we changed, in order, and what each one moved.