Does ChatGPT prefer long articles or short ones?

Somebody on your team has told you that nobody reads long articles any more. Somebody else has told you that ChatGPT only quotes long, detailed pages. The budget is finite, the people who could write are busy with client work, and you need to decide whether the next twelve pieces should be 600 words or 3,000.

The studies do not settle it, and they disagree with each other. What they do agree on is where the answer should sit. A 2026 analysis of 1.2 million ChatGPT answers by Kevin Indig, reported by Search Engine Land in February 2026, found that 44.2% of citations came from the first 30% of a page, 31.1% from the middle and 24.7% from the final third. Google, for its part, says in its guide to generative AI features (updated July 2026): "There's no ideal page length".

So the decision is less "long or short" than "how much does this question need, and is the answer near the top". This guide gives you a way to decide per piece, sets out what the studies found and why they conflict, and explains how to write sections that can be lifted on their own.

#Do AI assistants favour longer pages?

Not on their own. The studies disagree, and Google says there is no ideal length. What counts is whether the page answers the question completely and early.

Longer pages, when they are good, contain more of what an assistant can use: a clear definition, the reasoning, the exceptions, a worked example and answers to the obvious follow-up questions. For example, when a buyer asks what should be in a managed IT support contract for a 40-person accountancy firm, a page that names the things that belong in the contract, explains why each matters and notes the two that firms usually forget gives an assistant something to build on. A page that says "a good contract should cover response times, scope and costs" is not wrong, but gives it little to use.

Specific sentences take more words. "Response time targets should distinguish between one user's problem and a whole-site outage" is a sentence an assistant can lift. "Make sure response times are clearly defined" is not. That is why good pages on big questions tend to be longer, and why padding, which adds words without adding specifics, does not help.

#What do the studies say about length?

They disagree, mainly because they measure different things. The table sets out the three most useful recent studies and Google's position.

Why do they conflict? SE Ranking counted citations per page across many prompts, and a long page covers more sub-questions, so it can be cited for more of them. AirOps measured how often a page was cited when it was retrieved for one specific query, and a focused page that answers that query directly wins more often than a long guide that answers it somewhere in the middle. Both are correlational, and both come from firms that sell search tools.

Our reading: long pages can collect more citations overall, but each citation goes to a passage that answers one question well. Google's guidance on creating helpful content makes the same point from the other side, asking whether you are writing to a particular word count because you heard Google prefers one: "No, we don't."

#How long should our articles be?

As long as the question needs. The ranges below are what we work to, by type of page. They are working ranges, not rules.

The floors in our other guides describe the same pattern from different angles. What an article that ChatGPT quotes looks like treats a broad B2B subject as usually under-treated below about 1,200 to 1,500 words, and our content audit method scores a page of under about 800 words that covers a subject at one level as shallow. A 300-word glossary entry is not shallow, because its question is narrow. The test is always the question, never the count.

For example, an employment law firm's page on "what is a settlement agreement" can be well under 1,000 words and be better for it. "Should we offer a settlement agreement or go through a formal process" is a big buying question, because the buyer needs the trade-offs, timings and risks before they can act.

#Does ChatGPT read the whole page or only the top?

It can read the whole page, but its citations cluster near the top. In the Indig analysis, 44.2% of ChatGPT citations came from the first 30% of the page, with a sharp drop near the footer. Cited passages were also nearly twice as likely to use clear definitions ("X is", "X refers to").

So put the direct answer in the first paragraph, the key figure in the first third, and a plain definition wherever you introduce a term. ID-Pal, an identity verification firm that had bought NorthRow, rebuilt its industry pages on this pattern. Its case study records: "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." And: "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." The ID-Pal case study has the detail.

#Why does every section need to answer its own question?

Because assistants and Google's AI features often lift one passage, not a whole page. Google's AI Mode and AI Overviews use query fan-out, which Google describes as "a set of concurrent, related queries generated by the model", and each related query can match a different section of your page.

That is also why ranking alone is not enough. Ahrefs found in March 2026 that only 38% of AI Overview citations came from pages in the top ten, down from 76% in July 2025. A section that answers a follow-up question clearly can be cited even when the page does not rank first for the main one.

• Make each heading the sub-question in the buyer's words. AirOps found pages with the strongest match between headings and the query were cited 41.0% of the time, against roughly 30% for weaker matches.
• Answer the heading in the first two sentences, then explain.
• Name the subject at the start of a section instead of "it" or "as mentioned above", so the passage makes sense on its own.
• Keep sections to a few hundred words. SE Ranking found sections of 120 to 180 words between headings performed best.

Google's guide says there is no need to break content into tiny pieces for AI, and we agree. The aim is sections that stand on their own inside a complete page.

#When is shorter content the better choice?

When the question is narrow, factual or time-sensitive, and extra context would only get in the way.

• Definitions. "What is Cyber Essentials certification" needs a clear definition, a few paragraphs of explanation, one example and a note on who needs it.
• Single-task instructions. "How to register for the Construction Industry Scheme as a contractor" is a list of steps and the common mistakes.
• News. A change in regulation, a new appointment, a new service: say what happened, what it means and what to do.
• Tool-specific help. Setting up one feature in one system rarely needs more than a short page.

Short wins when the question has one answer and no real trade-offs. Long wins when the buyer is deciding something and needs the reasoning. Most pages that get a firm named fall into the second group, but the short pages are the supporting cast, and they link to the main pieces.

#Should we split a long guide into several pages?

Usually, when it answers several distinct questions. AirOps found that pages answering the main query narrowly outperformed broader, more comprehensive guides, and a 6,000-word "ultimate guide" hides the answer to any one question somewhere in the middle.

Break it into a short overview page that links to the separate pieces, keep the original address for the overview, and give each new page its own question as its title and a direct answer at the top. You keep the content and gain several quotable pages. Covering the follow-up questions buyers ask matters because Google fans out into them; organising content so AI assistants see you as the expert explains how to map them before you write.

#How do we tell whether a long article is earning its length?

Ask five questions of the draft. If it fails any of them, it is too long, or it is two articles.

1. Can you say what the article argues in one sentence?
2. Can you name the point of each section in a few words?
3. Would removing any section leave the reader worse off?
4. Is there a shape: the question, why it matters, how it works, where it goes wrong, what to do?
5. Does every claim have an example, a figure or a reason behind it?

A piece that passes all five at 2,400 words is finished. A piece that fails two of them at 4,000 words needs an editor, not more words.

#What to do this month

1. List the ten questions that most decide whether a buyer shortlists you, and mark each as a big question or a narrow one.
2. Check the first 30% of your five most important pages: if the direct answer and the key figure are not there, move them up.
3. Read every heading on those pages and check that the first two sentences under it answer it.
4. Run the five-question test on any page over 4,000 words, and split those that answer several questions.
5. Write two short pages for narrow questions that link to your strongest big-question page.

#Frequently asked questions

#What is the ideal article length to get cited in Google AI Overviews in 2026?

There is none. Google's own guide says "There's no ideal page length", and the studies disagree. What helps is a direct answer early, headings that match the questions people ask, and sections that answer the follow-up questions Google's fan-out searches for.

#Will anyone actually read a 2,500-word article?

The people who matter will, if it answers their question properly. What loses readers is padding, poor structure and no answer at the top. Give the answer first and a reader, or an assistant, can take what it needs and stop.

#Is a 500-word answer page enough to get cited by Perplexity?

For a narrow, factual question, it can be. AirOps found pages under 500 words were cited more often by ChatGPT than pages over 5,000; we know of no published length study for Perplexity itself. For a buying question with trade-offs, 500 words usually leaves out the reasoning, and a fuller page will serve the buyer better.

#How long should a B2B service page be for Google and AI?

Long enough to say who the service is for, what problem it solves, how it works, when it is not the right fit, and what proof you have, with the answer to "what do you do" in the first paragraph. In our experience that is usually shorter than a guide, and the guides do the deeper explaining and link to it.

#We have a 6,000-word guide that took weeks to write. Should we cut it?

Probably split it. A very long guide usually contains three or four questions that each deserve a page with its own title. Keep the original address for a short overview that links to them.

#Is it better to publish often or to publish long?

Publish steadily, at the depth each question needs. In our experience a regular rhythm of well-structured pages builds more than an occasional very long one. How often to publish for AI assistants sets out a realistic cadence.

#Does word count matter for videos and podcasts?

Not in the same way. A recording helps with assistants once it is written up, and the written version needs the same structure as any other page. Our piece on turning webinars and podcasts into written content covers the process.

Write as much as the question needs, put the answer where it will be found, and stop. If you want a second opinion on which of your pages are earning their length, the free Growth Review is where we usually start: 30 minutes with David Adams and three fixes in writing. If the site's structure is the problem, that is Website That Sells.

#Sources