Put the reader's question in the main heading, answer it in the first two sentences, and give each follow-up question its own section under a plain heading. Keep every fact that matters as text in the HTML your server sends, and use real tables for comparisons.
Google says its AI features need no special optimizations, markup or "chunking". So this is simply a layout for a reader in a hurry. Each step below comes with what an engine maker documents or what our scans measured. For what to write inside the layout, see how to write content that AI engines cite.
The page, top to bottom
Here is the skeleton. The bracketed parts are yours to fill.
<article>
<h1>How much does payroll software cost for a 10-person company?</h1>
<p>[The answer in one or two sentences, with the key figure.]</p>
<p>By [name], [role] at [company]. Prices checked [date].</p>
<h2>What the main plans cost</h2>
<table>
<caption>Monthly price for 10 employees, as published on [date]</caption>
<tr><th>Plan</th><th>Price</th><th>What it includes</th></tr>
<tr><td>[Plan A]</td><td>[$X a month]</td><td>[Features]</td></tr>
</table>
<h2>What pushes the price up</h2>
<p>[Answer, then detail.]</p>
<h2>Which plan fits which team</h2>
<p>[Answer, then detail.]</p>
<h2>How we checked</h2>
<p>[Sources, with links and the dates they were read.]</p>
</article>
Why the opening lines carry the most weight
An engine rarely uses a whole page. In our study of 3,531 answers from seven engines, Perplexity cited 20.2 third-party sources per answer, ChatGPT 4.7 and Gemini 3.5. When an answer draws on twenty sources, each one is a small part of it. When it draws on three or four, each one carries far more of what the reader is told.
Either way, the sentences that get used have to make sense on their own. A heading that names the question, followed by a sentence that answers it with a figure and a date, still reads correctly when it is lifted out of the page. A sentence that depends on the paragraph before it ("As mentioned above, this varies") does not.
Build it step by step
-
Make the main heading the question. In our scans of seven engines, a cituna.com guide headed "Which AI visibility tool includes Google Search Console?" was cited 21 times by ChatGPT and Gemini. Google's helpful content checklist asks the same of every page: does the main heading give a descriptive summary of the content?
-
Answer in the first two sentences. State the answer, with its key figure, before any background. A reader who stops there should still leave with the answer. Put the same answer in the meta description: Google says it sometimes uses that tag for the snippet when it describes the page more accurately than the page text would.
-
Give each follow-up question its own heading. Google recommends organizing pages into paragraphs and sections, with headings that give a clear structure. It also says AI Overviews and AI Mode may run several related searches on subtopics, which it calls query fan-out. A page that answers each subtopic under its own heading has a relevant section for more of those searches.
-
Put comparisons in an HTML table. Google lists making sure important content is available in textual form among its advice for AI features. An image of a table is not text. A
<table>with a<caption>and header cells is. Microsoft goes further for Copilot and Bing: its February 2026 post introducing AI citation reporting names clear headings, tables and FAQ sections as ways to make key information easier for AI systems to reference accurately. -
Keep the main content in view. For the "read more" deep links in its results, Google asks for content that is visible on the page without opening an expandable section or a tab. Keep the answer and the table outside accordions.
-
Render it on the server. Vercel and MERJ found that the crawlers from OpenAI, Anthropic and Perplexity do not run JavaScript, while Gemini uses Googlebot's rendering. When we fetched nine vendors' pricing pages without JavaScript, one showed plan names but loaded every price by script. JavaScript and AI crawlers covers the fixes.
-
Show the author and the date the facts were checked. Google asks whether pages carry a byline where readers expect one. It also lists changing a page's date without substantially changing the page as a warning sign, so change the date only when the facts change.
-
Do not hide the answer from snippets. Google says a page must be indexed and eligible to show with a snippet before it can appear as a supporting link in AI Overviews or AI Mode, and its newer guide to generative AI features adds that the site must be included in the Search generative AI setting in Search Console. Its
nosnippetrule applies to both features and stops the page being used as a direct input to them.max-snippetlimits how much can be used, anddata-nosnippetkeeps a marked section out of snippets. Check that no template wraps your answers in any of them by accident. -
Mark up only what the page shows. Google's structured data guidelines say markup must describe content that is visible to readers. Our guide to schema markup for AI search covers which types are worth adding.
What you can leave out
Google's guide to its generative AI features names three things that are not needed:
- Chunking. Google says its systems can understand several topics on one page and show the relevant part, so there is no need to cut content into tiny pieces.
- An ideal length. Google says there is none. Some subjects need a short page and some a long one.
- Perfect markup. Semantic HTML is not required, though Google calls it a good idea because it helps screen readers and other users navigate.
Check a page in five minutes
-
Confirm the main heading is in the server's HTML:
curl -s https://www.example.com/page | grep -o "<h1[^>]*>[^<]*</h1>" -
Confirm the answer is there too, replacing the phrase with the opening words of your own answer:
curl -s https://www.example.com/page | grep -c "Most plans for 10 people cost"A count of 0 means a crawler that runs no JavaScript never sees the answer.
-
Look for snippet controls that could hide it:
curl -s https://www.example.com/page | grep -Eo "nosnippet|max-snippet:[^\"]*|data-nosnippet" -
Turn off JavaScript in your browser's developer tools and reload the page. What remains is close to what a crawler that runs no scripts receives.
What to do first
- Rewrite the main heading of your ten most important pages as the question each one answers.
- Move each answer into the first two sentences.
- Replace every image of a table with an HTML table.
- Search your templates for
nosnippetanddata-nosnippetaround key content. - Run the curl checks on one page per template.