What is Answer Engine Optimization (AEO)?

Answer Engine Optimization (AEO) is the practice of structuring a website so that generative AI search engines — ChatGPT, Claude, Perplexity, You.com, and Google’s AI Overviews — can crawl your pages, understand them well enough to summarize, and cite them as sources in their answers.

If SEO is “how do I rank in the blue links,” AEO is “when an AI gives a one-paragraph answer with two citations, am I one of the citations?”

Why it matters now

ChatGPT alone reaches more than 800 million weekly active users. Roughly 17% of Google searches now show an AI Overview at the top of the page. Inside those AI answers, the model has to pick 1–3 sources to cite out of the entire web. The sites that get cited share a small, knowable list of properties — and almost all of them are things you control.

The tricky part: many of the things AI engines need are not the same as what traditional SEO optimizes for. A page can rank #3 in Google and still be invisible to ChatGPT, because:

The HTML loads behind a Cloudflare challenge that bots can’t pass.
The content is rendered client-side by React or Vue and the raw HTML is empty.
robots.txt blocks OAI-SearchBot even though Googlebot is allowed.
There’s no JSON-LD structured data telling the model what kind of page this is.
The article buries the answer 800 words in, when the model is reading the first 200 words to extract a summary.

The four AEO pillars

Out of all the things you could change, four categories of signals do most of the work:

1. Fetchability

Can the AI crawler load your page at all? This sounds trivial but it’s the single biggest delta between sites that get cited and sites that don’t. The relevant questions:

Does a plain GET with a normal User-Agent return a 200 and the actual content?
Is your robots.txt allowing the AI bots that matter — OAI-SearchBot, ChatGPT-User, Claude-User, Claude-SearchBot, PerplexityBot, Perplexity-User, Google-Extended?
Is the content present in the server-rendered HTML, or only after JavaScript executes? AI crawlers don’t run JS reliably.

2. Core SEO foundations

AEO inherits the SEO foundations — title, meta description, canonical URL, OpenGraph, sitemap, html lang. AI search engines use these as the same kind of “page identity” signals that Google does. None of this is new. The difference is that AEO is less forgiving: a missing title tag costs you in both worlds, but a missing JSON-LD block usually only hurts AEO.

3. Semantic HTML and structured data

This is where AEO starts to diverge from classic SEO. The model is reading your page to extract facts. It needs:

A clear single <h1>, hierarchical headings (h1 → h2 → h3), and proper landmarks (<header>, <nav>, <main>, <article>, <footer>).
JSON-LD for the page type — Article, Organization, FAQPage, Person, LocalBusiness, Product, RealEstateListing. Each gives the model a typed view of the page.
Author bylines linked via sameAs to LinkedIn, GitHub, ORCID — provenance signals that tell the model “this human wrote this.”
A dateModified that’s actually recent. Stale content is silently down-weighted.

4. Content quality (the GEO signals)

The 2024 Princeton paper GEO: Generative Engine Optimization (Aggarwal et al., arXiv:2311.09735) measured the actual citation lift you get from specific content patterns. The biggest wins:

Statistics density (+41% citation rate). Numbers and percentages get extracted into the AI’s summary almost verbatim.
Quotations (+28%). Cited blockquotes from named sources increase your selection probability.
Front-loaded answers. The model usually reads the first ~200 words. If your direct answer is in paragraph 1, you’re vastly more likely to be cited than if it’s in paragraph 7.
Question-shaped headings. “What is X?” and “How do I Y?” headings match the way users phrase queries.

How AEO is different from SEO

	SEO (Google blue links)	AEO (AI citations)
Goal	Rank #1–10 in SERP	Be 1 of 1–3 cited sources
Crawler	Googlebot (renders JS)	GPTBot, Claude-User, etc. (mostly don’t render JS)
Content read	Often just the first scroll	Often just the first 200–500 words
Page identity	Title, description, schema	All of that + JSON-LD types + llms.txt
Forgiving of JS-only?	Mostly yes	No
Counts external links to you?	Yes (PageRank)	Less; provenance via `sameAs` matters more

The biggest practical difference: AI engines almost always read raw HTML, not rendered HTML. If your page needs JavaScript to show content, you can be invisible to ChatGPT while still ranking on Google.

How to start

Audit your current state. Run your URL through the AEO Site Checker — you’ll get a 0–100 score with a per-check breakdown.
Fix Fetchability first. It’s worth 43 of 100 points. Make sure the page loads server-side, AI bots aren’t blocked in robots.txt, and you’re not behind a WAF that fails on plain fetches.
Add JSON-LD. Pick the right schema for the page (Article for blog posts, Organization for the homepage, LocalBusiness for a service business) and fill it in fully — author, dates, address, contact info.
Front-load your answers. Re-read your top pages: is the direct answer in the first paragraph?
Publish an llms.txt. A 200-line markdown index of your most important pages, written for LLMs to read. Cheap to produce, worth 5 points.

Once you’ve done all five, re-audit. The score should move 30+ points on most sites.

Common AEO myths

“Just give the AI good content. It’ll figure it out.”

It will not. A great article hidden behind a Cloudflare challenge is invisible. Ranking-quality content is a necessary but not sufficient condition.

“My SEO score is high so AEO is high.”

No correlation guarantee. The biggest sites in the world (Amazon, NYTimes) often score poorly on AEO because they’re behind aggressive bot protection or they JS-render their content. Smaller, plainer, well-structured sites often outscore them on AEO.

“It’s just llms.txt.”

llms.txt is one signal worth 5 points out of 100 in our rubric. Useful, cheap, but it won’t carry a site that fails on Fetchability.

Ready to see your score? Run an audit →