Answer engine optimization (AEO) is the practice of structuring a website so that AI assistants can read it, trust it, and cite it in the answers they generate. Where traditional SEO competes for a click on a results page, AEO competes to be the source quoted before a results page ever appears.
That distinction sounds academic until you watch someone actually search. They open ChatGPT, or they type into Google and read the AI Overview at the top, and they get a paragraph with three or four sources named underneath. Then they act. For a large and growing share of queries, no one scrolls to the blue links at all.
You can be ranked first and still lose that customer. That's the problem AEO addresses.
How is AEO different from SEO?
SEO optimizes for position. AEO optimizes for citation. Same website, different finish line.
A search engine returns a list and lets the human choose. Your job is to be high on the list and interesting enough to click. An answer engine reads a handful of sources, synthesizes one response, and hands it over finished. Your job is to be one of the sources it reads — and to be phrased in a way it can safely reuse.
Three practical consequences follow from that:
- The unit of competition shrinks. A results page shows ten links. An AI answer typically draws on three to five. The funnel is narrower, so partial credit is worth less.
- Clarity beats persuasion. A model won't repeat a claim it can't parse or verify. "Industry-leading solutions for modern brands" is unquotable. "Builds start at $6,500 and take two to four weeks" is extremely quotable.
- Structure is no longer optional. A human skims past a bad heading hierarchy. A model relies on it to work out what a passage is even about.
What makes a page quotable by an AI?
After building and rebuilding for this for a while, I've found it comes down to five things. In rough order of impact:
1. Your content exists in the HTML
This is the one that silently kills more sites than everything else combined. If your copy is injected by JavaScript after page load, a meaningful number of AI crawlers will never see it. Google's own crawler renders JavaScript; many AI crawlers fetch the raw HTML and move on.
Test it the crude way: view source — actual source, not the inspector — and search for a sentence from your homepage. If it isn't there, no amount of clever wording will help you.
2. Headings are shaped like questions
People ask assistants in full sentences. "How much does a website cost for a small business?" not "website cost small business."
When your <h2> matches the shape of the question, you've handed the model an obvious mapping between what was asked and where the answer lives. It's a small change with a strangely large effect.
3. The answer comes first
Under every question-shaped heading, answer it in the first sentence or two. Completely. Then elaborate.
This runs against how most of us were taught to write — build context, then reveal. But an assistant extracting a passage will grab the opening lines. If those lines are throat-clearing, you've given it nothing worth quoting. Notice that this article's first sentence is a complete definition of AEO. That's deliberate.
4. Passages stand on their own
A model may lift a single paragraph away from everything around it. If that paragraph says "as mentioned above, this approach works well," it's useless in isolation — and slightly risky to quote, so it gets skipped.
Write so that any given paragraph would still make sense if it were the only thing someone read. Restate the subject instead of relying on "it" and "this."
5. Facts are specific and consistent
Models are cautious about repeating claims they can't corroborate. Specific, checkable, internally consistent facts get reused; vague superlatives don't.
Consistency matters as much as specificity. If your homepage says you serve the Northeast, your about page says nationwide, and your Google Business Profile lists one city, you've given three conflicting answers to the same question. The safe move for a model is to say nothing about you at all.
Where does structured data fit in?
Schema.org markup is how you state facts about your business in a format that requires no interpretation. Instead of hoping a model infers your service area, opening hours or pricing from prose, you declare them.
The types that matter most for AEO are Organization or LocalBusiness, Service, FAQPage, Article, and Product if you sell things. Wire them together with @id references so the markup describes a connected entity rather than a pile of unrelated blocks.
I wrote a longer piece on getting this right: schema markup that actually does something.
Should you let AI crawlers access your site?
If you want to be cited, yes. Blocking GPTBot, ClaudeBot, PerplexityBot or Google-Extended in robots.txt removes you from consideration in the systems doing the citing.
It's a real trade-off and reasonable people land differently on it. Publishers with subscription revenue often block deliberately — an AI summary substitutes for the visit they're monetizing. But if you're a business whose website exists to generate enquiries, being absent from AI answers costs you far more than being summarized in them.
A growing convention is llms.txt: a plain-Markdown file at your site root that describes what your site is and points to your most useful pages. Support is still uneven and it's not a standard yet — but it costs ten minutes and it's a clear, honest signal. This site has one.
How do you measure whether any of it worked?
AEO has no Search Console, which is genuinely annoying. Two things work in the meantime:
- A prompt panel. Write twenty to thirty questions a real customer would ask. Run them across the major assistants monthly. Record whether you appear, how you're described, and who appears instead. It's manual and it's the most honest signal available.
- Server logs. Filter for AI crawler user agents. You'll see exactly which pages are being fetched and how often — which tells you what these systems consider worth reading on your site.
The short version
AEO isn't a replacement for SEO and it isn't a new discipline requiring new software. It's mostly the discipline of being legible: putting your content in HTML, structuring it around real questions, answering them directly, stating facts you'd be comfortable seeing quoted back at you, and marking it all up so a machine doesn't have to guess.
Which, conveniently, also makes the site better for the humans.