Answer Engine Optimization: What AEO Is and How to Get Cited

Most people searching for answer engine optimization are not really asking for a definition. They are asking whether the last ten years of work on their website just quietly stopped counting.

I am Armen, an SEO consultant in Barcelona, and I run the agency ACERO Digital with a team of specialists. That worry lands on almost every first call now, usually dressed up as a joke about firing whoever does their SEO. The honest answer is narrower than the panic and much narrower than the sales pitch.

Your SEO still counts. What changed is the thing that gets selected, and that is really all AEO is about.

What is answer engine optimization?

Answer engine optimization, or AEO, is the practice of shaping your content and your technical setup so an AI driven answer system can retrieve your text and quote it inside a direct answer. You are competing for a citation in the answer, not for a position in a list.

That sounds like a small distinction until you look at what it changes on a Tuesday morning. Classic SEO asks whether your page deserves to be in the running. An answer engine has already decided the user does not want a running order. It wants two sentences and a source.

So the work shifts down a level. You stop optimising the page and start optimising the paragraph that has to survive being torn out of it.

What counts as an answer engine?

More surfaces than most businesses count. The AI Overview at the top of Google and its AI Mode tab are answer engines. So are ChatGPT with browsing, Perplexity, Gemini and Copilot. So are the older extractive features that have been quietly doing this for years: featured snippets, People Also Ask, and the voice assistants that read one result aloud and stop.

Here is where I disagree with how this usually gets framed. Businesses treat AEO as the ChatGPT problem, because ChatGPT is the thing they read about, and then they ignore the surface that is already costing them money.

For a plumber in Sant Martí, the AI Overview sitting above the results for “how much does it cost to fix a boiler barcelona” moves far more revenue than anything happening inside a chat window. It is the same optimisation work. Only one of the two shows up in your Search Console.

How does an answer engine choose what to quote?

In two steps, and the first one is where most content dies. The system retrieves a small set of candidate passages from an index, then a language model writes an answer using only what came back. Retrieval decides who is in the room. Generation decides who gets quoted.

That retrieval step does not work on whole pages. Content gets broken into chunks, and chunks get matched against the question. The model is not reading your article. It is reading a handful of fragments that survived a filter, and it never sees the rest of what you wrote.

Google adds another layer that Ahrefs calls query fan out: the original search gets broken into related sub queries, and sources are selected against those instead of the phrase the person typed. Your page might be cited for a question you have never had in a keyword tool.

Does ranking on Google still get you cited?

Ranking helps, but it stopped being a gate, and the two best datasets on this openly contradict each other.

Ahrefs analysed 863,000 keywords and 4 million AI Overview URLs and found that 38 percent of cited pages appeared in the top 10 for the same query, down from 76 percent in their July 2025 study. The rest split almost evenly: 31.2 percent came from positions 11 to 100, and 31.0 percent from beyond position 100. BrightEdge, tracking the same overlap for sixteen months, reports the opposite trajectory, climbing from 32.3 percent in May 2024 to 54.5 percent in September 2025.

Anyone quoting you one of those numbers without the other is selling something. What survives the disagreement is the useful part: a very large share of AI citations go to pages outside the top 10, so “we rank first” no longer buys you the answer, and “we rank eleventh” no longer disqualifies you from it.

The payoff is worth the effort. Seer Interactive tracked 3,119 informational queries across 42 organisations and found that brands cited in the AI Overview earned 35 percent more organic clicks than uncited competitors on the same query. In the same dataset, organic click through rate on those queries fell from 1.76 percent to 0.61 percent. The pie shrank. The citation is what is left worth winning.

Is AEO different from SEO and GEO?

Mostly they are three names for overlapping work, sold separately. The distinction that holds up is what the engine does with your text: an extractive surface lifts your sentence, a generative one rewrites it and credits you.

You do not need three budgets or three vendors for this. If the GEO column is the one you care about, how a language model decides which business to name is the longer version of that story. The crawlability and the clear entity underneath all three columns are the same asset, which is why most of what we do in an SEO consulting engagement now covers all of it without renaming the invoice.

What changes in the way you write?

The unit. You are writing passages that have to stand alone, so the first two sentences under every heading have to answer that heading outright.

A few concrete edits do most of the work. Put the number and its date in the same sentence as the claim, because a model lifting “38 percent” without “Ahrefs, 2026” attached will not cite you for it. Keep any definition under about fifty words so it fits a snippet. Phrase your H2s as the question a person types, not as a clever label.

And write something that is actually worth lifting. A sentence that needs the paragraph above it to make sense will not survive retrieval, and a paragraph that says what forty other pages already say gives no engine a reason to pick yours.

What does the technical side of AEO involve?

Crawler access first, then structured data, then a clear statement of what your business is. In that order, because the other two do nothing if the bot never arrives.

Check your robots.txt for GPTBot and OAI-SearchBot (OpenAI), PerplexityBot, and ClaudeBot. Watch out for Google-Extended: it governs whether your content helps Gemini apps and grounded responses, not whether you appear in AI Overviews inside Search, which runs on ordinary Googlebot. I have seen a site block Google-Extended believing it was opting out of AI Overviews and get neither outcome it wanted.

On structured data, be honest about what it does. No engine has confirmed schema as a citation factor. What it reliably does is remove ambiguity about which entity a price or a review belongs to, and a retrieval system with less to guess at is a retrieval system more likely to use you. Organization and FAQPage markup take an afternoon. A thorough technical SEO audit finds the crawl blocks that quietly cap everything else.

How do you measure AEO without inventing numbers?

With three sources, none of which is a vendor dashboard. Search Console shows impressions and clicks from Google’s AI experiences. Your analytics shows referrals from chatgpt.com, perplexity.ai and gemini.google.com, usually a small number with a conversion rate that makes you look twice.

The third source is manual and it is the one that actually tells you something. Write down twenty questions a customer asks before buying from you, run them in ChatGPT and Google on the first of every month, and log who gets named. Vary the phrasing, because the answer moves with the wording more than anyone expects.

Google has said plainly that no external tool has access to its internal AI metrics. So when a platform sells you an AI visibility score to two decimal places, you are buying an estimate with a confident font.

When is AEO not worth your time?

When the fundamentals underneath it are missing. If your site is slow and thinly indexed, AEO is a distraction dressed as a strategy, and the fix is ordinary technical and content work that will improve your AI visibility as a side effect.

It is also a poor fit if your customers do not research before buying. A takeaway near Sagrada Família is not losing orders to ChatGPT. Its problem is the map pack, and pretending otherwise is how agencies sell expensive answers to cheap questions. If someone has already quoted you a monthly figure for this, look at what an AI SEO retainer actually buys before you sign it.

The blunt version: you cannot format your way into being worth quoting. If the underlying page has nothing specific in it, no schema, no heading rewrite and no llms.txt file will make an engine choose it.

What I would fix first

If you gave me one afternoon on your site, I would not touch the code. I would open ChatGPT and Google side by side, run the ten questions your customer asks before they hire anyone, and write down who gets named each time. That list is your real baseline, and it is usually uncomfortable reading. Then I would take the one page that should have been named and rewrite the top of it so the first two sentences answer the question completely, without the run up.

After that, robots.txt, and a plain sentence on your homepage saying what you do and where you do it. It sounds too simple to matter. It is the single change that most often moves what a model says about a business, because an engine cannot recommend a company it cannot summarise.

If you want a second pair of eyes on where your business currently stands in AI answers, tell me what you sell and I will look at the queries that matter for it.