How to write a page AI search can cite

Start from the person who asked

An answer engine quotes a page because the page already answered the question. It does not quote a page for using the word “AI” near the word “SEO”. Google’s helpful-content guide and the July 2026 generative AI guide describe the same test in plainer clothes: would a visitor who already knows the topic feel the page taught them something, or would they feel they had read it somewhere else?

Google’s example of the weak version is a list of tips any site could have published. The stronger version is a specific account only the writer had: what they did, what it cost, what went wrong. On a technical topic the equivalent is a procedure you can follow, with the limit stated. “Create great content” is not a procedure. “Put the answer in the first paragraph, then link the source under the figure” is.

That standard is uncomfortable for a product blog, including this one. The pages here are useful when they report a primary source, a crawler token, or a price printed on a vendor’s own page, and when they say what DiscoSEO will not promise. They are not useful when they restate a slogan. If a paragraph could sit on any other GEO site unchanged, it does not belong.

Put the answer where it can be lifted

The first paragraph should be the answer to the H1. A model that is willing to cite still has to find a passage. An opening about the company, the year, or why the topic is exciting gives it nothing to lift. The lede on this page is the answer to its H1. The rest of the page is here so the answer is not a slogan on an empty URL.

Google says there is no ideal length and no requirement to chop a page into chunks. Shorter and longer both work when they match the question. Extra words help when they are the rest of the answer: the contrast, the steps, the exceptions, and the sources. Extra words hurt when they are a second introduction.

Write the headings as the questions a reader would ask next. “Who has to be able to fetch the page?” is a heading. “Our philosophy” is not. Use one H1. Use H2 for the main questions and H3 only inside one of those questions. A heading outline that skips a level is harder for a person and for a parser.

Make the page fetchable

The answer has to be in the HTML, not in a picture of the HTML, and not only inside a widget that never writes the text into the response. Google’s AI features page says important content should be available in textual form. A canvas, a slideshow with no transcript, or a PDF that the HTML never links is a page an answer engine may never read.

JavaScript is allowed. Google says it can process content in JavaScript when the script is not blocked, and that a JavaScript site is simply harder to get right. If the answer appears only after a click that a crawler does not make, assume the crawler did not see it. View the page with scripts disabled, or read the HTML response, before you decide the paragraph exists.

One canonical URL. If the same answer lives at three addresses, you are asking the engine to guess which one to name. Pick the URL, point the others at it, and link that URL from the homepage and from any guide that assumes the reader already knows the point. A file nothing links to is easy to miss.

Sign it, date it, and show the source

A citation needs a source a person can check. Name the author. Link the name to a site that is clearly theirs. Show a date a person can see, and put the same date in the article metadata. When you rely on a number, link the page that published the number. When the number moves, update the page and the date.

This is also the defense against the failure mode of AI writing. A page that restates other people’s posts, with no source and no author, is the page a model can replace with a sentence of its own. A page that says “Pew measured this in March 2025, on 68,879 Google searches, and here is the write-up” is a page the model has a reason to point at. The figures live on GEO statistics, each one next to its link.

Do not mark up a claim the prose does not make. Google’s structured-data guidelines say the HTML body has to describe the same thing the JSON-LD describes. A FAQ in the script, with no questions on the page, fails that test. Schema is covered on Schema markup for AI search.

One question, not a fan-out farm

Google describes query fan-out on the July 2026 guide. While an AI answer is built, the system issues related searches. The guide’s example starts from “how to fix a lawn that’s full of weeds” and fans out into herbicides, non-chemical removal, and prevention. Google then says not to create a separate page for every variation in order to manipulate those answers. That pattern can fall under the scaled-content spam policy, and a high count of URLs does not make a site more relevant.

Publish the page the question deserves. If a second question needs a different answer, a different procedure, and different sources, give it a URL and link the two. If the second question is the first question with the words rearranged, it does not need a URL. The Chat GPT version of one question as one page is How to show up in ChatGPT.

What still will not get you named

A clear page is eligible to be quoted. Eligibility is not a citation. Chat GPT, Claude, Perplexity, and Google each decide whether to name a source, and the decision changes. A firewall can block a crawler that robots.txt allows. A newer page can answer the question more specifically. The product you described can stop being the product you sell.

DiscoSEO does not guarantee rankings or citations. The way to check, without pretending one chat is a ranking, is How to check AI search mentions. The way the crawlers differ is AI crawlers and robots.txt.