Do headings and bullet lists help AI quote my pages?

Headings and bullet lists are not a lever for AI citations. The best available evidence says formatting on its own barely moves which sources an AI answer uses.

Two things point the same way:

• The experiment. The 252,000-trial factorial study, "What Gets Cited" (arXiv:2605.25517), tested many page edits against each other. It reports that "formatting-only edits have little impact." Topical relevance and the document's position in the candidate list were the biggest drivers of being cited first.
• The operator. Google's guide to optimizing for generative AI features (July 2026) says: "You don't need to write in a specific way just for generative AI search." It also rejects chunking as a requirement and states there's no ideal page length.

What this means: reformatting a page into short blocks, adding more H2s, or converting prose into bullets is unlikely to change whether ChatGPT, Perplexity or Google AI Overviews quote you. The next move is to keep structure where it helps people and stop treating it as an AI project. The full study breakdown is in our digest of the 252,000-trial GEO factorial.

The formatting finding sits inside a wider pattern. Almost every published GEO technique has been tested on pages already inside the model's context, and the results are weak.

• C-SEO Bench (NeurIPS 2025) found most conversational SEO methods "not only largely ineffective but also frequently have a negative impact on document ranking."
• A July 2026 critical survey of 45 GEO studies (arXiv:2607.14035) warns that "citation-oriented rewrites can impair retrieval."
• Both the survey and the factorial name topical relevance and context position as the most reproducible drivers. Position is set by the retrieval system, not by you. See position bias for why that matters.

Read the factorial with its limits in view. It was written by Sprinklr for an industry track, used a synthetic corpus of 100 e-commerce blogs, gave the model two candidate documents per trial, and never called a live search engine. So it shows what happens once your page is already a candidate. It doesn't test whether formatting helps a page get retrieved in the first place. Nobody has tested that, so the honest label is untested, not proven useless.

That's also why heavy rewriting carries risk without a proven upside. Our comparison of on-page rewrites vs access fixes lays out the trade.

There is one case where page structure affects AI directly. It's when the structure keeps text out of the HTML that crawlers read.

Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but "don't execute them. They can't read client-side rendered content." An independent re-test in June 2026 found the same for GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot and PerplexityBot, among others. Googlebot renders JavaScript, so this mostly affects non-Google engines.

Common layouts that can trigger it:

• Tabs or accordions that fetch their content only when clicked
• "Load more" buttons and infinite scroll
• Pricing tables built entirely in client-side JavaScript

If your key facts live inside one of these, the problem isn't the formatting. It's that the text isn't there for the crawler. A related trap: a data-nosnippet attribute or a nosnippet robots meta tag inherited from a template will keep a section out of AI summaries on Google and Bing, however well it's formatted.

The one on-page change with positive experimental support is content, not layout. Stating a specific fact (a price, a number, a one-line definition) on a page that can be fetched gives an engine a sentence it can quote. The 252,000-trial factorial also reports that "including explicit price information and a recent timestamp also helps consistently."

A bullet list can carry those facts well, because it's easy to read. But the fact is doing the work, not the bullet. Say you run a bookkeeping firm and your pricing page says "plans from $200 a month" in a tab that loads with JavaScript. Moving that sentence into the server-rendered HTML matters. Turning it into a styled bullet doesn't, as far as anyone has measured.

The same logic applies to a summary at the top of a page. We cover that in whether a summary box helps AI quote you. On timestamps, be careful: a changed date with no changed content proves nothing, as explained in whether the last updated date helps.

1. Keep your headings and lists. They help readers scan, and Google says existing structure does no harm. Don't rip them out.
2. Stop reformatting for AI. Don't schedule a project to "chunk" pages or rewrite them in an AI style. The evidence doesn't support it.
3. Check what a crawler sees. Load your key pages with JavaScript disabled. If prices, product names or definitions disappear, server-render that content.
4. Check access before content. Confirm your robots.txt allows the search bots (OAI-SearchBot, PerplexityBot, Claude-SearchBot) and that your firewall doesn't block them. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.
5. State your facts once, clearly, and consistently. One accurate price on the pricing page beats three conflicting versions in nicely formatted lists.

For a broader checklist built on the same evidence, see the one-hour AI SEO fix list. Nothing here guarantees a citation. These steps remove the blocks you control.

Should I remove headings or bullet lists from my pages? No. The evidence says formatting has little effect on AI citations, not that it hurts. Headings and lists make pages easier for people to read and navigate, which is reason enough to keep them. Removing structure would cost you readability and gain nothing measurable. Leave sensible formatting in place and spend your time on crawler access and on stating your key facts in plain, fetchable text.

Does breaking my content into short chunks help LLMs? There's no evidence it does. Google's AI optimization guide rejects chunking as a requirement and says there's no ideal page length. The academic chunking research measures how a retrieval system splits documents inside its own index, which is a setting no publisher controls. Write sections at the length the topic needs for a human reader.

Do FAQ sections help AI answers more than regular headings? A question-and-answer layout is just another format, and the formatting evidence applies to it. Google deprecated FAQ rich results in May 2026, and FAQ markup isn't a lever for AI answers either. If your customers genuinely ask those questions, answering them clearly is useful. We cover this in detail in [whether you need an FAQ page for AI](/questions/do-i-need-faq-page-for-ai).

Are tables better than bullet lists for AI search? Nobody has published evidence that one format beats the other for AI citation. What matters is whether the table's contents are in the HTML before JavaScript runs, since most non-Google AI crawlers don't execute scripts. A server-rendered table with clear prices or specs is readable to crawlers. A table built client-side may be invisible to them, however tidy it looks.