What is Answer Engine Optimization (AEO)?
Answer Engine Optimization is the practice of getting your brand named and cited inside the answers that ChatGPT, Claude, Gemini, Perplexity, Grok and Google AI Overviews generate. It is measured by how often you appear in those answers, not by where you rank on a page of links.
Search used to hand you a page of ten links and let you choose. Increasingly it hands you a paragraph. That paragraph names two or three companies, cites a handful of sources, and the user acts on it. If your brand is not in the paragraph, the click never existed to be lost.
AEO is the discipline of getting into that paragraph. The mechanics overlap with SEO — crawlable pages, clear structure, credible sources — but what you optimise for, and how you measure it, are different.
How AEO differs from SEO
The single biggest difference is the unit of measurement. SEO measures a URL's position for a keyword. AEO measures a brand's presence in an answer to a question. One page can be ranked #1 for a keyword and never be mentioned when a user asks the corresponding question in ChatGPT, because the model synthesised its answer from three other sources it found more quotable.
- SEO tracks keywords; AEO tracks prompts — the actual sentences people type into a chat box.
- SEO tracks position; AEO tracks presence, position within the answer, and share of the mentions against competitors.
- SEO results are relatively stable day to day; answers are regenerated per query and drift constantly, which is why sampling frequency matters far more.
- SEO has one dominant surface; AEO has at least six, and they disagree with each other regularly.
That last point is the one teams underestimate. It is entirely normal to be cited in four out of six engines and invisible in the other two. If you only measure one engine, you will draw a confident conclusion from a quarter of the evidence.
The four numbers that matter
Every credible AEO measurement reduces to four questions, and it is worth being precise about what each one answers.
- Visibility Score (or Presence Score): of all the tracked prompts, in what share of the generated answers were you mentioned at all? This is the headline. It answers 'do the engines know we exist for this topic'.
- Share of Voice: of all the brand mentions across those answers, what proportion were yours? Visibility can rise while Share of Voice falls — that means the category is getting more crowded faster than you are growing.
- Average Position: when you are mentioned, how early in the answer do you appear? First-named brands carry disproportionate weight, exactly as the first result on a SERP always did.
- Sentiment: when you are mentioned, is the mention favourable, neutral, or a caveat? A mention that reads 'cheaper but harder to set up' is not the same asset as a recommendation.
Track all four or you will optimise the wrong one. Visibility with poor sentiment is a reputation problem wearing a growth-metric costume.
What actually moves the numbers
In our own audits, the causes of low citation are unglamorous and repeat constantly:
- Throat-clearing. Pages that open with 'In this article we will explore…' give a model nothing quotable in the first 200 words. Lead with the direct answer instead.
- Answers buried below the fold, under a narrative build-up, where an extraction pass will not reach them.
- Missing or wrong structured data, so the page's claims cannot be read as facts.
- Thin internal linking, which leaves strong pages orphaned and out of the crawl paths that matter.
- No third-party corroboration. Engines lean heavily on sources that are not you — comparison pages, forums, docs, reviews.
None of these need a rebrand. They need someone to open the file and change it — which is exactly the step where most AEO programmes stall.
Where measurement stops and work begins
The awkward truth about the AEO tool category is that almost all of it stops at the dashboard. You get a score, a competitor chart, and a list of issues. Then the list goes into a backlog, and six weeks later the score has not moved, because nobody had a spare afternoon to rewrite the intro of eleven pages.
This is why Fenn drafts the change as a real diff and opens the pull request against your repository, behind a human approval step. The measurement is table stakes. The fix is the product.
Where to start this week
- Write down the 10 questions a buyer would actually type into ChatGPT before choosing something in your category. Those are your prompts.
- Run them against every engine, not one, and record who gets cited.
- Rewrite the top three pages so the direct answer is the first sentence.
- Re-measure after a fortnight. Answer drift is real; a single reading is an anecdote.
Measure this on your own domain.
Fenn tracks all four metrics across every engine, and opens the pull request that fixes what it finds. Free for one domain, indefinitely.