The Seven Rules of AEO: What Gets Cited by AI

A page can be the best thing written on its subject and never get cited once, because no single section of it stands on its own. Seven rules fix that, and none of them cost anything.

By Jason Vanderheyden, Director of Web Development and Product Engineering, Hahn

What are the seven rules of AEO?

Seven rules decide whether an AI engine cites your page:

Answer the question in the title.
Put the answer first.
Make every section stand alone.
Name things, never “it” or “this.”
Be specific, and show your work.
Keep it fresh, and show the date.
Never hide it.

These are the universal rules from Hahn’s AEO Best Practices Field Guide. They hold for every medium, every industry, and every engine we have measured. They are also, individually, unglamorous. Almost nothing on this list requires new budget, new headcount, or a platform migration. Most of it is a decision made differently at the moment of writing.

Why do these seven rules work?

Two gates decide every citation, and each of the seven rules serves one or the other.

The first gate is trust. The engine decides whether it takes you seriously enough to consider quoting you at all. This is where E-E-A-T lives (experience, expertise, authoritativeness, trustworthiness), the framework Google built for its human search-quality raters, which research consistently finds AI engines applying with the same logic. Google’s own rater guidelines call trust “the most important member of the E-E-A-T family,” because an untrustworthy page scores low no matter how expert it looks. Rules 5 and 6 serve this gate.

The second gate is extraction. An AI engine does not read your page the way a person does. It breaks the page into passages, scores each passage independently against the question it was asked, and cites the strongest individual passages. Your opening paragraph, each section under a heading, each FAQ answer, and each row of a table are all competing for citation separately. Rules 1, 2, 3 and 4 serve this gate.

Rule 7 sits underneath both. If a page cannot be reached at all, neither gate ever applies to it.

Here is each rule, what it looks like when someone breaks it, and what it looks like when someone follows it.

Rule 1. Answer the question in the title

Shape your headline like a question a real person would type. It does not need to match the query word for word. It needs to be recognisable as the same question.

Instead of: Pricing
Write: How much does it cost to charge an EV at home?

Instead of: Our Story
Write: What does grass-fed mean on a beef label?

Instead of: Q3 Industry Trends
Write: What are the biggest commercial solar trends in 2026?

Why it works. The engine scores each passage against a question. A heading shaped like that question is the strongest possible match; a heading shaped like a filing-cabinet label is no match at all. In our research, the pages that dominated a question were the pages whose titles and URLs were that question.

The practical note. On a blog post, your title tag and your H1 can usually be the same sentence, because blog titles are naturally question-shaped. On a services page they should differ: the navigation label stays short for wayfinding (“Services”), the title tag carries your keyword phrase and brand for search results, and the H1 carries the question for extraction. Those three names do different jobs and were never meant to match.

Rule 2. Put the answer first

Lead every section with the direct answer in roughly 40 to 60 words, then expand. The wind-up goes after the answer, or nowhere.

Instead of: “Choosing the right kitchen layout is one of the biggest decisions in any renovation, and there is a lot to weigh. Every home is different, and costs depend on your goals. We have been designing kitchens in Austin since 2004.”
Write: “A full kitchen remodel in Austin typically costs $40,000 to $75,000.”

Instead of: “Understanding warranty coverage can feel overwhelming. Let’s walk through what you need to know so you can make an informed decision.”
Write: “The standard warranty covers parts and labour for 24 months from the install date.”

Why it works. The engine scores the passage on whether it answers the question. Everything before your answer is dead weight in that scoring, and a passage that spends its opening on scene-setting is competing against a competitor whose opening is the answer.

Most wind-ups should be cut, not moved. Sentences like “there is a lot to weigh” and “let’s walk through what you need to know” are not context, they are throat-clearing, and they carry no information to relocate. If a piece of context genuinely earns its place, it goes after the answer as the expansion. If it does not, delete it.

The tell. If your first sentence could open a section on any subject, it is not an answer.

Rule 3. Make every section stand alone

One idea per section, and no callbacks. Each chunk has to survive being lifted out and read on its own, because that is exactly what happens to it.

Instead of: “As we mentioned above, this makes crawling easier.”
Write: “Server-side rendering makes a page easier for AI crawlers to read.”

Instead of: “The third option is usually the cheapest of the ones we just covered.”
Write: “Ground-mounted solar is usually cheaper per watt than roof-mounted solar.”

The test. Copy any one section of your page into a blank document. Does it still make sense with nothing around it? If it opens with “as we saw above,” or leans on “it” and “this” instead of naming things, the answer is no. The engine will skip it and cite a competitor whose section stands alone.

This is also an architecture rule, not just a writing rule. A page trying to answer six questions loses to six pages answering one each, because each competitor offers the engine a whole URL purpose-built for the question while your page offers a fraction of one. One page, one job. That decision comes before anyone writes a word.

Rule 4. Name things, never “it” or “this”

Write the company’s full name. Write the product’s actual name. Every time, even when it feels repetitive.

Instead of: “It reduces peak demand by 12%.”
Write: “The GridWise time-of-use plan reduces peak demand by 12%.”

Instead of: “This is included in every tier.”
Write: “Overnight battery monitoring is included in every Sunfield service tier.”

Why it works. When a passage is extracted, its pronouns lose their subject. A paragraph that made perfect sense on your page becomes a floating claim about an unnamed thing, and an unnamed thing cannot be attributed to you.

Why it matters more than it sounds. This is the cheapest fix on the list and almost nobody does it, because good human prose actively avoids repetition. You are writing for two readers with opposite preferences, and the machine’s preference is the one nearly everyone is currently ignoring.

Rule 5. Be specific, and show your work

Every claim that could be checked should carry the number, the source, and the date. Vague claims get skipped in favour of specific ones.

Instead of: “Our beef is raised responsibly.”
Write: “Our cattle rotate pasture every 21 days across 4,000 acres in Hill County, Texas.”

Instead of: “We’re the region’s leading installer.”
Write: “We completed 340 commercial installations across 12 states between 2011 and 2026.”

Why it works. On some engines, unsupported superlatives actively read against you: the passage gets classified as marketing rather than reference, and reference is what the engine is looking for. “Industry-leading” is noise. A number with a source is a citation candidate.

The highest-leverage version of this rule is owning a number. In our data center research, individual figures published by Lawrence Berkeley National Laboratory each appeared in 150 to 240 separate AI answers. Nobody outranked them. They published the number first, sourced it properly, and now they are simply the answer. If you hold proprietary data, that is the most valuable asset you own. Publish the headline figure publicly, on one page, with a stated methodology and a date.

Rule 6. Keep it fresh, and show the date

Put a visible “Last updated” line on every priority page, and put a real refresh behind it. Evergreen content is largely a myth in AI search.

Instead of: an undated 2023 roundup, still accurate, quietly aging
Write: the same page, reviewed, figures updated, stamped “Last updated: August 2026”

Why it works. One crawl-log analysis of AI bots across more than 5,000 URLs found 94% of AI crawler visits land on content published within the last five years. Recency is not a tiebreaker in AI search; it is close to a filter.

The discipline this implies. A post that earns citations is an asset with a decay rate. Revisit it, update the facts, and move the visible date only when a genuine refresh sits behind it. A date bump with nothing behind it erodes the trust gate rather than serving it.

Rule 7. Never hide it

If it is behind a login, a paywall, a form, or a PDF, no engine can crawl it, retrieve it, or cite it. Gated content is invisible content.

Instead of: a grid-reliability study available only as a PDF behind a form
Write: a public page carrying the headline finding and the key statistic, with the full report behind the form

Why it catches everyone. It catches them at their strongest asset. Gate your best original research and you have created the single most citable thing you own, then hidden it from every engine on earth.

The fix is not to abandon lead capture. It is to split the asset. The public page earns the citations and drives people to the gate. You get AI visibility and lead capture instead of choosing between them.

The same logic applies to anything that requires a click to see. Tabs, accordions, carousels and “read more” expanders frequently hide their contents from crawlers. Anything you want cited lives in the initial rendered HTML, visible without interaction.

We will note the obvious: the guide these rules come from is a gated PDF, and this page exists because we read our own rule 7 and took it.

Do these rules apply to every page?

No. Apply them to pages that answer questions, and leave the rest alone.

Four of the seven cost nothing visually. Putting the answer first, naming your subject, sourcing your numbers, and showing a date constrain a designer in no way at all, and there is no good reason to break them anywhere. The other three can genuinely collide with a creative idea: a question-shaped headline is wrong for a brand campaign, one page per question fights a deliberately expansive landing page, and keeping everything out of tabs and accordions rules out a pattern that sometimes solves a real density problem.

When they collide, the useful question is not how strictly to follow the rule. It is whether that page is trying to be found. A pricing page, a service page, a how-to article and an FAQ all exist to answer something a person asked, and on those the rules should win every time. A brand film, a portfolio piece, a recruitment page, a campaign microsite: nobody is asking an AI engine for any of those, so optimising them buys nothing and costs you the work. Decide which group a page belongs to before you brief it and most of the tension disappears.

One part stays non-negotiable regardless. Whatever you most want cited cannot be gated. A design argument for an accordion is a real argument; a design argument for putting your original research behind a form is not.

What do these seven rules not cover?

They are the universal layer. They do not tell you what changes when the medium changes, and the medium changes a great deal.

Video is a text game. AI engines never watch a video, they read the transcript, the description, and the chapters, and roughly 94% of AI citations to YouTube go to long-form rather than Shorts. Earned media is the only reliable route into ChatGPT and Gemini, and no amount of publishing substitutes for it. Podcasts are completely invisible until the transcript exists on a page you own. Social is mostly reach, not citation, with one significant exception. And your third-party entity footprint (Wikidata, LinkedIn, directories, review platforms) determines whether an engine is confident enough about who you are to quote you at all.

The AEO Best Practices Field Guide covers all eight channels, with a 60-second pre-publish checklist for each and the research behind every claim. It is free.

Get the full playbook

Free, and it adds eight channel checklists, a 60-second test to run before you hit publish, and the research behind every claim on this page.

Get the free field guide

Frequently Asked Questions

Does writing for AI make content worse for people?

No, with one exception. Answering the question first, naming your subject instead of using pronouns, and putting a source on every number are standard technical-writing practice, and a reader who skims a page benefits from all three as much as an engine does. The one real cost is repetition: naming your company or product in every section reads as slightly redundant to someone reading top to bottom. That is a small price for being quotable.

What should a company fix first to get cited by AI?

Start with access. Check whether anything you want cited currently sits behind a form, a login, a PDF, a tab, or an accordion, because content an engine cannot crawl cannot be cited, and no other change helps until that is solved. Then rewrite your page headlines as the questions those pages answer. Those two changes take an afternoon and move more than everything else combined.

Should every heading on a page be a question?

Most subheadings should be, but not mechanically. A heading works when it is shaped like something a person would actually ask, and forcing a question onto a section that is genuinely a statement produces awkward copy that helps neither readers nor engines. The page's main headline is the one to get right every time: make it either the question the page answers or the answer itself.

Do these principles apply to service and product pages, or only to blog posts?

They apply everywhere, and service pages are where they get broken most often, because design pressure is highest there. The one-word navigation label becomes the headline, the answer sits below a hero banner, and the substance ends up inside tabs. A service page built like documentation, with clear sections and the facts in the open, gets cited. One built like a brochure does not.

More About AEO

How do AI engines choose what to cite? We tracked 2,429 answers to find out

AI engines do not cite the biggest brand, the highest domain authority, or the most popular page. They cite whoever published the clearest answer to the question being asked, and most of the time, nobody credible has. We know because we spent three months logging where four AI systems get their answers about one contentious industry: 2,429 answers, roughly 19,100 cited links.

How do you get YouTube videos cited by AI? Best practices for video AEO

AI engines never watch your videos. They read them: the title, the description, the transcript, the chapters. That single fact decides which videos get cited in AI answers, and it explains why a 200-view video with clean metadata can beat a 500,000-view video without it. Here is what the evidence says actually works.