Answer Engine Optimization: A Useful Label for Work That Is Still SEO

Answer engine optimization, usually shortened to AEO, is the practice of getting your content used inside a generated answer rather than shown as one of ten blue links. The label is roughly three years old, it is being sold hard, and the most important thing to know about it comes from Google itself: in its guide to optimizing for generative AI features, Google states that from Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.

I sell services under this label, so read the next paragraph with that in mind. AEO is a useful word and a dishonest invoice. It usefully names a shift in what winning looks like. It does not name a separate discipline with separate deliverables, and anyone quoting you for AEO as a workstream running alongside your SEO is charging you twice for one set of work.

What the label usefully names

Something real did change, and pretending otherwise is the opposite error. For most of search’s history the unit of success was a ranking position and the reward was a click. Increasingly the unit of success is being the source a generated answer draws on, and the reward is often a citation rather than a visit. That changes what you measure, what you write, and what you promise a client. It does not change the machinery underneath.

The machinery is worth stating precisely because it explains why there is no back door. Google’s AI features are built on the same index and the same core ranking systems as the blue links. Google describes using retrieval-augmented generation to improve the quality, accuracy and freshness of AI responses by relying on its core Search ranking systems to retrieve relevant, up-to-date web pages, and describes query fan-out as concurrent related queries the model issues alongside the original. For eligibility it is unambiguous: a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements.

Retrieval first, generation second. You cannot be cited if you cannot be retrieved, which means every classic ranking problem is also an AEO problem.

The answer surfaces, and how much we actually know about each

Treating every assistant as one undifferentiated thing is how bad AEO advice gets written. They differ, and so does the quality of public information about them. I am separating the two in this table because most vendor material does not.

Surface	What decides inclusion	How well documented	What to do
Google AI Overviews and AI Mode	Core ranking retrieval, then generation. Must be indexed and snippet-eligible.	Documented by Google in detail	Classic SEO, plus non-commodity content. Measure in Search Console.
Featured snippets	Ranking plus a clear, extractable answer on the page	Long-standing and well understood	Answer the question in the opening of the page
Assistants with live web retrieval	Their own retrieval layer, frequently built on a search index	Thin. Most public detail is inference, not documentation.	Be retrievable and be worth quoting. Distrust anyone claiming more precision.
Assistants answering from training data	What was in the training corpus, frozen at a cutoff	Effectively opaque	Nothing reliable. Treat claims to influence this as unfounded.

The bottom two rows are where the industry’s overclaiming lives. When a vendor tells you precisely how to get cited by a given assistant, ask where the documentation is. For Google there is a published guide and a Search Console report. For most of the rest there is neither, and confident advice in the absence of either is someone’s guess with a price attached.

What actually earns a citation

Google’s own list is short and unglamorous, and it is the same list that earns rankings.

Be retrievable. Indexed, snippet-eligible, no nosnippet, no crawl blocking, no demoted section. Everything else is downstream of this.
Publish something non-commodity. Google names this directly: create valuable content and bring a unique point of view. A page that restates the consensus is competing with a model that already has the consensus.
Answer in the opening. Not because generative systems demand a special format, but because grounding works from what the page says, and a buried answer is a worse document to ground against.
Cover the intent on one URL. Fan-out rewards a page that genuinely handles the neighbouring questions. It does not reward seven thin pages each handling one.
Stay current. Freshness is named explicitly in Google’s description of grounding.
Measure in Search Console. Google warns against third-party tools claiming to use internal Google metrics, because none have that access.

What to ignore, in Google’s own words

Google published a list of things that do not help. Every one of them is currently being sold as AEO.

Sold as AEO	Google’s published position
An llms.txt file	Google Search itself does not use them. They will neither harm nor help.
Chunking content for machine parsing	No requirement to break content into tiny pieces.
Rewriting pages in an AI-friendly voice	No need to write in a specific way just for generative AI search.
AI-specific structured data	Structured data is not required for generative AI search, and no AI-specific type exists.
Brand mention campaigns	Pursuing inauthentic mentions is not as helpful as it might seem.
A page per fan-out query variant	Violates the spam policies and is ineffective long term.

That last row is not a minor point. Publishing a page for every predicted sub-query is scaled content abuse by Google’s own definition, and it puts at risk the index eligibility that citation depends on. The tactic sold as the way to win AI visibility is among the more reliable ways to lose it.

How to price it, if you are buying

Three questions will tell you whether an AEO proposal is real work or a relabelled invoice.

What are the deliverables? If they are technical health, content with original substance, internal linking and measurement, that is SEO and should be priced as SEO. If they are an llms.txt file and a schema audit, you are buying the mythbusting list.
How will you measure it? Search Console’s generative AI reporting is the first-party source. Anything claiming internal Google data is not what it says it is.
What happens to my rankings? If the answer treats rankings as a separate concern, the proposal does not understand the retrieval step, and everything built on it will be guesswork.

The measurement trap worth naming

A page can hold its position, gain impressions and lose clicks, because a growing share of searches end without one. That is a different diagnosis from a ranking loss and it needs a different response, but it gets reported as a collapse and triggers panic changes that destroy attribution. Report impressions and average position alongside clicks, every time, or the number you are reacting to is not the number you think it is.

If you want the tactical version of this rather than the definitional one, the mechanics of grounding and fan-out and what to do about them are in how to rank in AI Overviews. If your pages are not being retrieved at all, that is a recovery problem before it is an AEO problem, and the diagnostic order is in Google penalty recovery. The retrievability checks themselves are Pass A of my SEO audit checklist, and if what is holding you back is a large volume of interchangeable pages, start with thin content.

Royking Niba is an SEO and GEO consultant specialising in penalty and spam-update recovery, with more than 8 million organic visits recovered for clients. Every Google position quoted here was checked against its published generative AI guidance on 22 September 2026.