How to write pages that AI answer engines cite.
Answer engines cite the pages their searches find and can quote. How to write one, from choosing the question to the HTML a crawler receives, with what Google itself says.
Published · 6 min read
What an answer engine does with your page.
An AI answer engine is an assistant that answers a question in its own words and links to the pages it drew on: ChatGPT with search, Gemini, Perplexity, and Google's AI Overviews and AI Mode. Before it writes, it usually searches the web. Google describes its version as a "query fan-out": several related searches across subtopics, run to build one answer (Google Search Central).
So a cited page has passed three tests. The engine's search found it. The engine could read it. And it held a passage the engine wanted to use. Everything below serves one of those three.
Google is plain about what it does not require. On the same page it says: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." A page has to be indexed and eligible to show in Google Search with a snippet. The rest is writing a page worth quoting.
We asked Gemini this question, "How do I write pages that AI answer engines cite?", on 27 September 2026. It searched for "how to write content for ai answer engines to cite" and "seo for ai answer engines", cited 12 guides on the subject, and advised leading with the answer, question-style headings, standalone sections and concrete data. Most of that advice matches Google's guidance and our own records. One piece does not, and it is covered further down.
Start from the question the engine searches.
A page gets cited on a question, not on a topic. Pick the question first, in the words a buyer would type or say, and write the page for that question alone.
The engine's searches usually stay close to the question. When Gemini answered "How do I track AI recommendations for my business?" ten times, its most frequent search, in 8 of the 10 answers, was "tracking ai recommendations for business". A page titled and opened around that phrase is a page the search can find.
Then match the page's shape to the kind of question:
| The question | The page that answers it |
|---|---|
| "What is the best X for Y?" | A list of the options, yours among them, each described fairly with its price and who it suits |
| "X vs Y", "alternative to X" | A comparison on the points a buyer weighs, with a clear verdict for each kind of buyer |
| "How do I…?" | A guide with the steps in order, the answer summed up in the first paragraph |
| "What is…?" | A definition in the first sentence, then how it works and an example |
This is what the engines cite. Across 84 Gemini answers to our tracked questions, 17 to 27 September 2026, 79 cited at least one page whose address reads like a list, and the how-to question went to guides whose addresses read almost exactly like it. How to find which websites AI search engines cite has the full breakdown.
Put the answer first, then the facts.
Answer the question in the first paragraph, in two or three sentences a reader could stop after. An engine that lifts one passage from your page should be able to lift that one.
Then give the facts a buyer compares, stated plainly:
- Numbers with their context. "$49 a month for 15 prompts, checked daily" beats "affordable". A statistic needs its sample and its date: "cited in 25 of 84 answers, 17 to 27 September 2026".
- Names instead of pronouns. Write "Otterly.AI's Standard plan", not "their mid plan". A passage lifted out of context still has to say who it is about.
- Tables for comparisons. Prices, limits and features across several options read better in a table, for a person and for an engine.
- Dates on anything that changes. Say when you checked a price, and update the page when it moves.
Specific facts are what an answer can quote. A paragraph of general praise gives an engine nothing to repeat, and nothing for a reader to check.
Make each section stand on its own.
An engine may use one section of your page and ignore the rest. Write each section so it makes sense read alone.
- Headings that name what the section answers. "How much does it cost?" or "Pricing for small teams" tells a reader, and an engine, what follows. "Setting expectations" does not.
- The section's answer in its first sentence, and the detail after.
- One idea per paragraph, and no "as mentioned above".
- Follow-up questions answered on the same page. A buyer who asks for the best tool next asks what it costs and whether it covers their case. A short FAQ at the end catches those.
Say where your facts come from.
Link each claim you did not measure yourself to the page it came from: the vendor's pricing page, the official documentation, the study. Readers can check you, and an engine comparing sources finds your figures agreeing with the primary ones.
Where you did measure something yourself, say how: what you asked, how often, over which dates, and how many answers the figure rests on. Original numbers with a stated method are the one thing on your page no other page has.
Make sure a crawler gets the text.
None of the writing matters if a crawler receives an empty page. Check the HTML your server returns, not the page your browser shows.
- Allow the crawlers. OpenAI says a site has to allow OAI-SearchBot in robots.txt to appear in ChatGPT's search answers (OpenAI). For Google, the page must return a 200, be indexable, and not block its snippet with
nosnippetor a tightmax-snippet. - Render the text on the server. A page assembled in the browser by JavaScript can reach a crawler as an empty shell. Frameworks that render on the server, or a pre-rendering service, fix that.
- Keep the answer in the HTML. We found our own FAQ answers missing from our pages: the accordion component left closed answers out of the HTML entirely, so a crawler saw all the questions and none of the answers. We rebuilt it so the answers are always in the page.
- Keep it public. A page behind a login or a paywall is invisible to the crawlers that feed these answers.
What not to spend time on.
Gemini's answer to this question recommended FAQ schema. Google's own page says the opposite for its AI features: "You don't need to create new machine readable files, AI text files, or markup to appear in these features." It adds that no special schema.org structured data is required. Structured data you already use should match the visible text, and that is all Google asks of it here.
Three other habits waste effort:
- Repeating the engine's name. Writing "ChatGPT" ten times on a page does not make ChatGPT cite it.
- Hidden text and keyword blocks. An engine quotes what a reader can see.
- Many thin pages on one question. One page that answers a question fully does more than five that each answer part of it.
Check whether it worked.
Track the question the page was written for, from before you publish, and read the change over two weeks or more: whether the engine searched, whether it cited your page, whether it named you. How to measure whether AI recommendations are working covers the numbers.
If your site lives in a repository and a coding agent writes its pages, outrankfast's MCP server hands the agent each suggested page with the exact searches it should answer and the competitor page it is up against, so the agent writes the page with the questions in front of it.