What does an article that ChatGPT actually quotes look like?

Your firm and a competitor both have a competent article on the same subject, and both rank somewhere on Google. When a buyer asks ChatGPT, Perplexity or Google's AI Overviews the question, the competitor is quoted and you are not, and it happens across the site.

The difference is rarely the writer's talent. It is how the piece is built. In Kevin Indig's analysis of 1.2 million ChatGPT answers, reported by Search Engine Land in February 2026, 44.2% of citations came from the first 30% of a page, and cited passages were nearly twice as likely to use clear definitions of the "X is" kind. Assistants do not read your article the way you wrote it. They look for a claim they can lift, near the top, stated plainly.

This guide takes the article apart element by element, with the evidence for each, a real example and an annotated page to brief a writer against.

#What is an assistant doing when it reads your page?

Three things, in order: finding candidate pages, reading them and citing. Most advice skips the first step, where many good articles fail.

Finding. ChatGPT search, Perplexity, Copilot and Google's AI features run web searches, and Google's often run several related searches for one question, which Google calls query fan-out. Your page has to turn up in those results. In an AirOps study reported by Search Engine Land in April 2026, pages in the top search position were cited by ChatGPT 58.4% of the time, against 14.2% for pages in position ten. Ranking helps without guaranteeing anything: Ahrefs found in March 2026 that only 38% of AI Overview citations came from top-ten pages, because the fan-out searches reach pages that rank for the sub-questions. Crawlers must also be let in: ChatGPT search needs OAI-SearchBot, and Perplexity asks sites to allow PerplexityBot.

Reading and citing. The assistant scans the candidates for a clear statement, evidence that it is reliable, and something the other pages lack, favouring the top of the page and plain sentences. When it names a source, it is telling the buyer where a claim came from. So ask of every article: can an assistant find it, and can it lift a reliable, distinctive claim and attribute it to us?

#The nine elements and the evidence behind each

Most of these studies are American, several are run by vendors that sell AI-visibility tools, and all are correlational. Treat them as direction, not law.

#Should the title and headings be questions?

They should use the buyer's words, and a question is often the most natural way to do that: "How long does a commercial lease renewal take?" rather than "Lease renewals: our guide". The evidence supports matching over punctuation: AirOps found pages with the strongest heading-to-question match were cited 41.0% of the time against roughly 30% for weaker matches, while an SE Ranking study of 216,524 pages, reported in November 2025, found question-style headings on their own averaged fewer citations than plain ones.

So write titles and headings in the words buyers type, as questions where that reads naturally. A title that promises "everything you need to know" promises nothing in particular.

#What should the first paragraph do?

Answer the question in the first two or three sentences, then spend the rest of the piece on why, how and when not. The most common failure we see is an article that takes four hundred words to reach the point, pushing its best sentence out of the part of the page assistants quote most.

Say the question is how long a lease renewal takes. The first paragraph says: typically four to six months from first notice to signing if both sides engage, longer if the landlord disputes terms, and here is what controls the timing. That is quotable. "Lease renewals are an important part of commercial property" is not.

Follow the opening with a short list of key points, and end each section with one sentence that states its point cleanly, in the definitive form the Indig study found assistants favour.

#How should the headings be organised?

One topic per heading, with the answer in the section's first sentence: "What delays a lease renewal?", "Can the landlord refuse?" Each becomes a unit an assistant can use on its own.

Keep the hierarchy clean and sections to a few hundred words; a two-thousand-word section under one heading is as hard to use as no headings at all. AirOps also found pages that answered the main query narrowly outperformed broad, comprehensive guides, so resist cramming three questions into one article.

#How much depth and length does an article need?

As much as the question needs. Depth is how expertise shows: a piece that says renewals take four to six months is useful; one that explains which steps run in parallel, what the landlord's surveyor can and cannot do, when the timetable slips and when relocating is cheaper is far more likely to be named. In our experience depth has four parts: the mechanism, the exceptions, the trade-offs, and more than one angle.

Length is where the studies disagree. AirOps found pages of 500 to 2,000 words performed best for ChatGPT citations, and pages over 5,000 words were cited less often than pages under 500. SE Ranking found articles over 2,900 words averaged 5.1 citations against 3.2 for those under 800. Neither supports a single word count, so do not set one. Whether ChatGPT prefers long articles or short ones goes through the evidence in full.

#What makes an article distinctive enough to quote?

Something the other candidate pages do not have. When two pages are equally well built, the one with a finding, pattern or comparison of its own wins. In a Princeton and IIT Delhi study presented at KDD 2024, adding statistics raised a page's visibility in AI-generated answers by about 34% and adding quotations from relevant people by about 41%. It was tested on a research engine, so treat it as directional.

It needs no research budget. A pattern from your own work, stated with its basis, is enough. Say where every figure came from and when: "in the trade body's 2025 survey" or "across the renewals we handled over the past three years". Does publishing our own research help ChatGPT recommend us? covers how to produce it from what you already have.

#Does the author matter, and how many sources should we cite?

Yes, and as many as your claims need. A byline with a name, a role and a line of experience signals that someone accountable knows the subject; "by the team" signals nothing. So do the marks of experience in the text: examples only a practitioner would know and what usually goes wrong. What makes ChatGPT treat our content as expert rather than marketing sets out those marks.

On sources, the Princeton and IIT Delhi study found citing credible sources raised visibility by about 30%, but it set no ideal number, and neither do we. Cite, link and date a source for every figure and every claim a sceptical reader would question.

#Should an article end with FAQs?

Yes, for readers, and do not expect a citation boost from the format itself. SE Ranking found pages with FAQ sections averaged 3.8 ChatGPT citations against 4.1 without, and Google stopped showing FAQ rich results on 7 May 2026. The case for a closing set of questions is that buyers have follow-ups, and Google's AI features run related searches for them. A page that answers the next three questions a buyer asks, in their words, serves the reader and gives the fan-out searches something to find. Do we need schema markup for AI assistants to understand our site? gives the practical view on marking them up.

#Does the date matter?

Yes. Ahrefs found in July 2025 that pages cited by AI assistants were on average 25.7% fresher than Google's organic results, with ChatGPT showing the strongest preference for recent pages. Show a publication date, a "last reviewed" date and a line on what changed, and change the content when you change the date. Does ChatGPT prefer newer content? covers how often to refresh.

#What does a real cited page look like?

Our work for ID-Pal, an identity verification firm, is the clearest example we can share. The starting point: "Across 36 recorded AI answers to the questions ID-Pal's buyers actually ask, ID-Pal was named three times, and only for its brand or its Salesforce integration. AI knew who it was, and had no reason to recommend it."

The rebuilt pages follow the shape in this guide: "Five industry pages rebuilt: fund administration, wealth management, legal, accountancy and estate agency. Each opens with the question a compliance lead types, names the regulator, and answers inside the first ninety words." That is element 1 (the buyer's question), element 2 (the answer first) and element 7 (the authority named) in one sentence.

The result: "Gemini now names ID-Pal for UK wealth managers. Claude names it for UK fund administrators, on a page rebuilt in August, and links NorthRow to ID-Pal in the same answer." Overall, "UK buyer questions where AI names ID-Pal: none in May, two in August, four in September 2026."

#An annotated page, element by element

For example, a commercial property firm's article on lease renewals, built to the nine elements:

#How do I get my article cited in Google AI Overviews?

Be indexed, eligible for a snippet and useful for the question or one of its sub-questions. Google says its AI features need no special markup or files, and that separate pages for every variation of a query, made mainly to manipulate rankings, break its scaled content abuse policy. So build one strong page per real question, cover the follow-ups, and put the answer where it can be lifted. What Google AI Overviews mean for your firm covers the wider picture.

#What makes Perplexity choose a source to cite?

Perplexity says it searches the internet in real time when asked a question and gives each answer numbered citations. Its crawler, PerplexityBot, "is designed to surface and link websites in search results on Perplexity", so Perplexity recommends allowing it in robots.txt. Beyond that, the same elements apply, and community sites weigh heavily: in Profound's 2025 study of 680 million citations, Reddit made up 46.7% of Perplexity's top ten sources, so a firm discussed on relevant forums and review sites has more ways in.

#What to do this month

1. Score your three most important service articles against the nine elements.
2. Search Google and Bing for each article's question and two follow-ups, and note where each page ranks.
3. Rewrite the first paragraph of each to answer the question directly, with the key figure inside the first third.
4. Add one specific figure or example from your own work to each, with its source and date.
5. Add a byline, a last-reviewed date and the three follow-ups buyers ask most.
6. Add each article's question to a fixed list of buyer questions and check it monthly in ChatGPT, Perplexity and Google AI Mode.

#Frequently asked questions

#Does the article need original research to be quoted?

No, but it needs something distinctive. A pattern from your own work, a comparison nobody else has assembled or a clear statement of where standard advice fails is enough. What does not work is a page that could sit on a competitor's site unchanged.

#Do AI assistants prefer lists and tables over paragraphs?

They prefer whatever states the point most clearly. Use a list for list content, a table when options are compared and prose where the reasoning is the point. Google says there is no need to break content into tiny pieces for AI, and we have seen no reliable study showing a blanket preference for lists.

#Should we write differently for AI assistants than for people?

Mostly no. What helps an assistant helps a reader: the answer first, one topic per section, plain headings, examples, honest limits. The one difference is placement. People may read to the end, but assistants quote the top of the page most, so nothing important should wait until the final third.

#Can a mediocre article on a big firm's site beat our excellent one?

Sometimes. SE Ranking found links from other websites were the strongest predictor of ChatGPT citations, which favours established sites. But a more specific page on a smaller site can win on specialist questions, and the Princeton and IIT Delhi study found lower-ranked pages gained most from these changes. Why most blog posts never get quoted covers the common failures.

#How long before a rewritten article gets cited?

Assistants that search the web live can pick up a rewritten page within weeks; answers from a model's memory change only when it is retrained, which takes months. Count a win only when it repeats.

If you would like to know how your articles compare with your competitors' today, book a free Growth Review: 30 minutes with David Adams and three fixes in writing. Writing the expert content that gets a firm named is part of Reputation That Wins.

#Sources