The AEO Playbook: How to Get Cited by AI Answer Engines

I spent a weekend last month auditing every JSON-LD block and OG tag on this site. Not because Google told me to. Because I'd watched our own brand mentions in Perplexity and ChatGPT jump from zero to a handful of citations over six weeks, and I wanted to understand why.

The short answer: answer engines read structured data differently than search engines do. They want to extract specific, verifiable claims they can quote with attribution. If your site doesn't make that easy, you don't get cited.

This is what people are starting to call AEO, for Answer Engine Optimization. It overlaps with classical SEO, but the success metric is different. SEO wants traffic. AEO wants citations.

What changed

Two years ago, getting found online meant ranking on Google. The user typed a query, scanned ten blue links, and clicked. The whole loop was a click.

Now a meaningful share of users never get to the blue links. They ask ChatGPT, scroll past Google's AI Overview, or check Perplexity. The model synthesizes an answer and cites a few sources at the bottom. If you're not one of those sources, you're invisible.

The numbers are still small, but the trend is sharp. ChatGPT's web search rolled out broadly in early 2025. Perplexity's user base has been doubling every few months. Google's AI Overview now appears on around a third of US search results pages, and for some long-tail informational queries it appears on every single one.

If your content strategy was built for keyword rankings, it's not going to translate automatically. The mechanics are different.

How answer engines actually pick sources

Three things matter, roughly in this order.

First, they need to find your page. Same as Google: sitemap, internal links, crawlability. If your robots.txt blocks the crawler, none of the rest matters.

Second, they need to understand what your page is about with very high confidence. This is where structured data becomes load-bearing. A page that uses Article schema with a clear author, date, and headline is dramatically easier for a model to parse than a page that's just paragraphs of HTML.

The third thing is the new part. Answer engines need to find specific, quotable claims. They synthesize: pull a sentence from one source, a statistic from another, a definition from a third. The pages they cite most often are the ones that make extractable assertions easy to lift out.

A page that says "we help companies grow" gets ignored. A page that says "the average 50-person company runs 40 to 60 SaaS tools" gets cited.

The schema hierarchy that actually matters

Most schema markup advice on the internet is bloated. You don't need 30 schema types. You need maybe seven.

Organization schema goes on every page in the root layout. It tells engines what your company is, where it's based, and what it does. Include sameAs links to your LinkedIn, X, and any other authoritative profiles. This is the entity definition.

WebSite schema also lives in the root layout. It defines the site as a whole, with a name, URL, and optionally a SearchAction. Engines use this to disambiguate when your domain appears in their training data versus their live index.

BreadcrumbList schema goes on every page deeper than the homepage. It helps engines understand site hierarchy and also makes your search snippets nicer. Cheap to add, no downside.

Article schema goes on every blog post. Include the author as a Person object with name, jobTitle, and url. Include datePublished and dateModified, the headline, and an image. This is what makes posts citeable as articles rather than generic web pages.

FAQPage schema goes on any page with a real FAQ section. The questions and answers need to be visible to users, not just hidden in markup. Engines penalize fake FAQs. When done correctly, your FAQ answers can appear verbatim in AI responses.

Service schema goes on service pages, with provider, serviceType, and areaServed. It helps engines understand what you sell.

SpeakableSpecification is a sub-property of Article schema. It marks the sentences a voice assistant should read aloud. Even if you don't care about voice, this signals to engines which sentences you consider most quotable.

That's the whole list. If you have those seven types in place, with accurate data, you've covered 95% of what matters for AEO.

The content patterns that get quoted

Schema makes you readable. Content gets you quoted.

After looking at hundreds of AI Overview and Perplexity citations across different industries, four patterns appear consistently in the cited sources.

Specific numbers come first. Engines love quantitative claims. "Tool sprawl costs the average company $300K to $800K per year" is far more citeable than "tool sprawl is expensive". When you can attach a number, attach a number.

Clean definitions are next. If you define a term early in your article in a single sentence, that sentence is likely to get pulled. Lead a post about process debt with "Process debt is the operational equivalent of technical debt" and you've handed the model a quotable line.

Comparative claims also get pulled. "X is faster than Y because Z" structures answer the user's question directly. The model can lift the comparison and cite you as the source.

First-person experience claims are the most underused. "We tracked a 45-person company's integration overhead for two weeks and found 62 hours per week going to manual data work" gets cited more often than third-party statistics. Models prefer primary sources when they can find them.

What doesn't get cited: vague claims, marketing language, lists without context, anything that reads like it was written for an SEO checklist.

Topical clusters beat individual posts

A single high-quality post on a topic might get cited once. A cluster of seven interlinked posts on the same topic, all referencing each other, gets the entire site flagged as an authoritative source on that topic.

After we interlinked our recent posts on operational scaling (process debt, tool sprawl, build-vs-buy, fractional leadership), citations went up across all four of them, not just the ones we'd promoted. The model started treating the cluster as a single source.

The mechanic is simple. Each internal link is a vote of confidence the model can verify. When seven posts on related topics all reference each other in context, the model concludes the site has real depth on this subject, not just one viral post.

Building a topical cluster takes deliberate planning. Pick a topic you want to own. Write the pillar post first, the one that defines the territory in detail. Then write four or five satellite posts, each going deep on one sub-aspect, all linking back to the pillar and to each other. Treat the cluster as a single content asset.

What to actually do this month

If you're starting from scratch, the order matters.

Week one, audit your existing schema. Use Google's Rich Results Test or Schema.org's validator. Most sites have either no schema or broken schema. Fix what's there before adding more.

Week two, add Organization and WebSite schema to your root layout. Add Article schema to every blog post. Add BreadcrumbList to deep pages. Don't worry about the more exotic types yet.

Week three, pick one topical cluster you want to own and write the pillar post. Make it actually useful. Reference real numbers, real cases, real experience. Avoid the kind of generic SEO content that ranked five years ago, because those pages get ignored by AI engines now. They read as filler.

Week four, write two satellite posts in the cluster. Link them to each other and back to the pillar. Submit the new URLs to Google Search Console for indexing.

Repeat the cluster pattern every quarter on a different topic. After 12 months, you'll have four well-developed topic clusters, each with seven to ten posts. That's the foundation for sustained AEO performance.

What to skip

A few things that consume effort and don't move the needle.

Skip exotic schema types. Course, Event, Recipe, JobPosting only matter if you actually run courses, events, recipes, or job listings. Don't add them speculatively.

Skip keyword-stuffed copy. Models are good at detecting this and penalize it. Write for humans first. The keywords will fit naturally.

Skip "AI-generated content at scale" plays. Engines have gotten very good at detecting generic LLM output and downranking it. The pages getting cited are the ones with specific, lived-experience claims that a model couldn't have generated on its own.

Skip backlink farming. Old-style link-building schemes don't help with AEO. Engines weigh internal coherence and direct topic authority far more than external link count.

The honest disclaimer

AEO is moving fast. The specific tactics that work in mid-2026 may not be the ones that work in 2027. What holds up are the durable basics: structured data, specific claims, topical depth, and primary-source content.

If you treat AEO as one variant of doing good content engineering, you'll keep up. Treat it as a checklist of tricks and you'll be chasing your tail.

The companies quietly winning the citation game right now aren't running AEO campaigns. They're doing what they were already doing well, which is publishing useful, specific, well-structured content. They've just put the right schema markup underneath. That combination is what answer engines reward.