By Mehdi HassanHow should you structure content so AI engines cite it?
Structure content so that any passage an AI engine lifts out still answers a question on its own. In practice that means four habits: open with the answer, write self-contained sections under question-style headings, use real tables and lists for anything comparative or sequential, and keep FAQ answers visible. Then make sure crawlers can read the HTML without running JavaScript.
None of this is a trick. AI Overviews, ChatGPT search and Perplexity all retrieve pages, pick passages that answer the prompt and cite the source. Pages that state clear answers in clear places are easier to retrieve and quote. The same pages also tend to rank and convert better, which is why we treat AI search structure as part of building the site, not a separate content project.
This guide is the hub for our content-structure series. Each section links to a deeper article, and the last section groups them as a reading list. If you want the site-level view (crawlability, rendering and schema hygiene), read our guide to building an AI-search-ready website alongside this one.
What does Google say about optimizing content for AI search?
Google says optimizing for its AI features is the same as optimizing for search. Its AI features documentation and its May 2026 guidance call llms.txt files, content chunking, AI-specific schema and rewriting pages for AI unnecessary. The pages that appear in AI Overviews and AI Mode are pages Google can crawl, index and judge helpful.
That matters because a lot of AEO advice sells exactly those things. The honest reading is that structure helps, but only the structure that also helps human readers: clear answers, logical headings, scannable comparisons. A page rewritten into short, robotic fragments to please a model is worse for people and gains nothing Google has said it rewards.
Other engines aren't Google, and they differ. Semrush's AI Visibility Index, built on 126 million prompts, found ChatGPT cites about 15 sources per answer against about 3 for Gemini. More sources per answer means more chances to be cited, but the same pages win in both: specific, well-structured, trustworthy ones. Our explainer on how AI search engines choose sources goes deeper on the differences.
Why should every page open with the answer?
Every page should open with the answer because both readers and AI engines weight the top of a page heavily. Reported analyses of ChatGPT citations, one covering 1.2 million responses and Kevin Indig's study of 18,000 citations, say about 44% of citations come from the first 30% of a page. If your answer sits under three paragraphs of preamble, it's in the part least likely to be quoted.
An answer-first opening is a 40–60-word paragraph that names the subject, answers the page's main question directly and makes sense quoted alone. It's not a teaser and it doesn't say "in this article we'll explore." It states the conclusion, then the rest of the page earns it with detail, evidence and nuance.
Write the question the page answers
One sentence, phrased the way a buyer would type or speak it: "How much does a Webflow to Framer migration cost?"
Answer it in the first sentence
Name the subject and give the answer: a range, a verdict, a definition or the key steps.
Add the condition that matters most
The one factor that changes the answer, stated specifically, not "it depends."
Keep it standalone
No pronouns pointing to earlier text, no "as mentioned above." Quoted alone, it should still be correct.
Repeat the pattern per section
Each H2 opens with one or two sentences that answer its own heading before adding detail.
We've written a fill-in version of this as the answer-first article template. For marketing pages rather than articles, such as a homepage, pricing page or service page, see answer-first page structure.
How long should each section be for AI search?
Each section should be long enough to fully answer its heading and short enough to stand alone: roughly 120–250 words under a question-style H2. AirOps reports more citations for sections of about 120–180 words, though that's a vendor heuristic, not a rule. The principle behind it is sound: a self-contained section is a clean, quotable unit.
Self-contained means the section makes sense lifted out of the page. It restates its subject instead of saying "this" or "it", it includes the specific numbers or conditions it relies on, and it ends with a conclusion. A reader who lands on that section from a jump link should get a complete answer.
Don't over-correct into tiny chunks. Google advises against splitting content into fragments for AI, and pages made of 40-word sections under 30 headings read like a broken FAQ. Sections that are too short lose the reasoning and evidence that make a source worth citing. Group related points under one heading and use H3s for sub-points.
For example, suppose a section answers "Is Framer good for SEO?"
A weak version: "It depends on your needs. Framer has a lot of features, and many people use it. Let's look at the details below." Lifted out alone, it says nothing.
A self-contained version opens: "Framer is good for SEO on marketing sites of up to a few dozen pages: it outputs server-rendered HTML, lets you set titles, descriptions and canonical tags per page, and generates a sitemap." Then it adds the limits (CMS structure, programmatic pages), when another platform fits better, and a one-line verdict.
The second version can be quoted by an AI engine, read from a jump link, or skimmed by a founder, and it works the same in all three.
Which content formats do AI engines quote most?
AI engines most readily quote definitions, direct answers, comparison tables, numbered steps, checklists and cost ranges, because those formats state facts in a structure that's easy to extract. Long narrative passages get quoted less, unless they contain a clear claim near the top.
Format | Use it for | Structure rule |
|---|---|---|
Definition | "What is X" queries | First sentence: "X is ..." with the category and purpose |
Comparison table | Options, platforms, plans, alternatives | Real HTML table, short cells, a verdict below it |
Numbered steps | How-to and process queries | One action per step, a verb first, 3–8 steps |
Checklist | Audits and readiness questions | Each item checkable as yes or no |
Cost range | Pricing and budget questions | Range by scope, what drives it, a fixed anchor if you have one |
Verdict | "Should I" and "which is better" queries | A clear recommendation and who it's not for |
Tables deserve special attention. Put comparisons in real HTML tables, never in images, and keep cells short enough to read as data. Our guide to comparison tables for SEO covers column choice and verdict placement.
Be careful with self-promotional listicles. Peec's analysis of 232,000 citations found self-promotional "best X" lists were only about 4% of ChatGPT citations, and Seer reported listicle citations falling 30% month on month. A "best agencies" page that ranks yourself first is a weak bet. An honest comparison with a verdict that names who each option suits is a much stronger one.
Should FAQ sections be visible on the page?
Yes. FAQ answers should be visible on the page, not hidden in collapsed accordions that need a click, and they should answer real buyer questions in 40–80 words each. Visible answers are easier for readers to scan and for crawlers to treat as primary content.
- Pick questions buyers actually ask: from sales calls, support tickets and "People also ask", not ones you wish they asked.
- Don't repeat your H2s: the FAQ should catch the follow-up questions the body didn't cover.
- Answer in the first sentence, then add the condition or detail.
- Keep each answer standalone, so it works quoted without the question.
- Add FAQPage markup as hygiene, not as a citation strategy.
Our full how-to on writing FAQ sections for AI search covers question sourcing, answer length and the design patterns that keep answers visible without making the page feel long.
Does schema markup or llms.txt help AI citations?
Neither schema markup nor llms.txt has shown a measurable lift in AI citations. Ahrefs' controlled 2026 schema test found no improvement: AI Overview citations fell 4.6%, while AI Mode and ChatGPT were flat. Ahrefs also found that 97% of llms.txt files got zero requests. Google says neither is needed for its AI features.
- Drop in AI Overview citations after adding schema in Ahrefs' 2026 controlled test
- 4.6%
- Share of llms.txt files that got zero requests, per Ahrefs
- 97%
- Correlation of web mentions with AI Overview visibility (Ahrefs, 75,000 brands)
- 0.664
- Correlation of backlinks with AI Overview visibility in the same study
- 0.218
Schema still has a job. Article with a named author, Organization with sameAs links and BreadcrumbList are basic hygiene that help search engines understand the page and keep it eligible for the rich results Google still shows. Just don't expect markup to make a weak page citable. We break down the test in does schema help AI citations, and give a verdict on the file in should your website have an llms.txt.
What makes a page trustworthy enough to cite?
A page is trustworthy enough to cite when it's clearly authored, current, specific and backed by evidence, and when the brand behind it is mentioned elsewhere on the web. Structure makes a page easy to quote; trust signals make it worth quoting.
- Named authorship: a byline linked to an author page with real credentials and links. See author pages and E-E-A-T.
- Freshness: a visible updated date and real updates. AirOps says over 70% of cited pages were updated within 12 months, and Ahrefs found AI-cited content is about 26% fresher. Our guide to content freshness for AI search covers how often to update.
- Original data: numbers nobody else has are what engines quote. See original research content.
- A clear entity: consistent company name, description and profiles so engines know who you are. See entity SEO for startups.
- Off-site mentions: in Ahrefs' study of 75,000 brands, web mentions correlated far more strongly with AI Overview visibility than backlinks did.
The mentions finding changes priorities for small teams. A good page on your own site is necessary, but being named in other people's articles, comparisons and communities is what moves AI visibility most. Pair each important page with something worth mentioning: data, a tool or a template.
Which pages should you structure first?
Structure your homepage, pricing, service and comparison pages before the blog. TripleDart's study of 28 B2B SaaS companies in Q1 2026 found homepages earned 31.9% of LLM citations and 81.7% of AI-attributed leads, while general blog posts drove sessions but no leads. The pages that close deals are the ones buyers ask AI about.
That's where structure meets conversion. Clicks are scarcer than they were: Ahrefs found the top result's click-through rate is 58% lower when an AI Overview appears. The visitor who does arrive, often from ChatGPT, has already read a summary and comes with a specific question. Your page has to answer it fast and make the next step obvious. Our article on how AI Overviews affect click-through rate covers what that means for page design.
Homepage
A plain statement of what you do, for whom and what it costs or how it's priced, near the top, in HTML text.
Pricing page
Prices or ranges in text and tables, what's included, and a visible FAQ on billing, contracts and limits.
Service or product pages
One page per offer, each opening with a direct answer to "what is it and who is it for."
Comparison pages
Honest tables against alternatives with a verdict that says who each option suits.
Then the blog
Articles that answer buyer questions and link to the money pages they support.
How do you make sure AI crawlers can read your content?
Make sure AI crawlers can read your content by server-rendering it, allowing the crawlers you want in robots.txt, and keeping key information in HTML text rather than images, PDFs or client-side widgets. Many AI crawlers don't run JavaScript, so content that only appears after scripts run may be invisible to them.
- Server-render pages: Next.js, Webflow and Framer all output HTML on the server for standard pages. Check that tabs, accordions and pricing toggles include their content in the initial HTML.
- Allow the crawlers you want: Googlebot, OAI-SearchBot and PerplexityBot, for example. Blocking a search crawler removes you from that engine's citations.
- Keep facts in text: prices, specs and comparisons belong in HTML, not screenshots.
- Protect citations during redesigns: keep URLs or redirect them one to one, and keep the answering passages intact. See migrating without losing AI citations.
If you're weighing a platform move for better control of rendering and structure, our breakdown of Framer to Next.js migration cost shows what that kind of project involves. For a complete pre-flight list, use the AEO checklist.
What should you read next in this series?
The rest of this series takes each structural habit and goes deeper. Start with the template if you're writing new pages, or with the schema and llms.txt pieces if someone has pitched them to you as an AEO fix.
If you want to... | Read |
|---|---|
Write a page that opens with the answer | |
Build comparison tables that get quoted | |
Write FAQs that get cited | |
Decide whether schema is worth the effort | |
Decide whether to add llms.txt | |
Make engines recognize your company | |
Strengthen bylines and E-E-A-T | |
Plan an update schedule | |
Publish data engines want to cite | |
Get cited in Google specifically |
