Which page format do ChatGPT and Claude cite most: guide, documentation or comparison?
Which type of page won the 20 questions Peaje measures for its own site, what each format is with an example and how to choose the format of your next page.
It depends on the question, and the safest way to know is to look at which pages ChatGPT and Claude cite today for your business's questions. In the measurement Peaje ran on its own site on November 7, 2026, technical documentation and guides won, followed by a pricing page, and no comparison made it into the four most cited sources. It is a small sample from a single business, so it shows the method, and its results may not repeat in your niche.
Updated: November 2026
Across those 20 questions, usepeaje.com had 0% of citations in ChatGPT and Claude, and the source that showed up most in its place was developers.google.com, with technical documentation, in 4 answers. We explain how that measurement works in how do I know if ChatGPT, Claude or Perplexity cite my business.
Who won the 20 questions Peaje measures?
These are the four sources that showed up most in the answers that did not cite usepeaje.com:
| Site | Answers it appears in | Type of page cited |
|---|---|---|
| developers.google.com | 4 | Technical documentation |
| help.openai.com | 3 | Guides |
| semrush.com | 3 | Pricing page |
| help.ahrefs.com | 3 | Guide to its Brand Radar feature |
Added up by format, guides appear in 6 answers, documentation in 4 and the pricing page in 3. The winning pages answer something concrete, like how a feature is used or what a plan costs.
Before you copy this, keep its limits in mind. It is a single business that sells to technical people, which partly explains the weight of documentation. It is a snapshot of one day. And Peaje asks each assistant through its API with web search turned on, so the app people use may cite other pages. Our blog is written as guides and still gets almost no bot visits, so we do not know yet whether that format will work for us.
Which formats does Peaje tell apart, and what is each one?
Peaje assigns a format to every page ChatGPT and Claude cite, based on its URL and title. It uses rules first (a docs. or developers. subdomain is documentation, a "vs" in the title is a comparison, "best" or "alternatives" is a roundup), and an AI model classifies whatever the rules cannot settle. Since it does not read the whole page, a page with an ambiguous title can end up in the wrong format.
| Format | What it is | Example |
|---|---|---|
| Guide | Explains how to do something or what something is | "How do I add llms.txt to my site?" |
| Roundup | A list of options, "the best", alternatives | "The best AEO tools in 2026" |
| Comparison | One option against another | "x402 vs MPP: which should I use?" |
| Research | Original data, a study or a benchmark | "How many stores accept agent payments: a study of 500 sites" |
| Documentation | Reference for a product or an API | The developers.google.com documentation that won 4 answers |
| Forum | Discussion threads, like Reddit or Stack Overflow | A thread asking whether anyone uses llms.txt |
| Video | YouTube, TikTok or Vimeo | A five-minute tutorial with chapters |
| News | A dated announcement | "Tempo launches its main network" |
| Product | A product or pricing page | The Semrush pricing page that won 3 answers |
How does Peaje choose the format of each page it suggests?
It gathers the pages cited in the last three measurement rounds, leaving out your own site, and groups them by format. When the agent suggests a new page, it favors the questions where a format already wins, picks that format and hands you a page AI already cites as a model, so you can follow its structure without copying its text. Each task carries a Format: line with how that type of page is built:
- Guide: the answer in the first paragraph, question-led headings, steps and a visible last-updated date.
- Comparison: who each option is for, a side-by-side table of what decides the choice (price, limits, coverage) and a verdict per use case.
- Roundup: the year in the title, clear inclusion criteria, a short verdict per option and your business included honestly.
- Research: one original data point with the method and the date it was measured, and findings written as sentences with numbers.
- Documentation: what it does, a working example with its response, and the limits and errors on the same page.
When there are no measurements yet, it picks the format that fits the question. In page series it does not suggest video or forum, because everything it drafts goes on your own site.
What do studies say?
The study that named GEO, from Princeton and IIT Delhi (KDD 2024), found that some changes to a page's text raised its visibility in generated answers by up to 40%, and the ones that worked best were citing sources, adding quotations and adding statistics, while repeating keywords did not help. They measured it on a simulated engine built on GPT-3.5 and on Perplexity, and not on current versions of ChatGPT or Claude. We did not find a study with a primary source that compares format by format in ChatGPT and Claude, which is why we give no percentages of that kind. For background on the term, see what is GEO and how is it different from SEO.
Which format should I choose for my next page?
- Type your question into ChatGPT and into Claude, and look at what type of page they cite.
- Use that format. If the question is about price, put your prices in text on a page of your own; if it is about how to do something, write a guide; if it is between two options, a comparison with a table.
- Add a number of your own with its source, which is what repeats most in cited pages and in the study above.
- Measure again in a few weeks, because citations change (how often to check).
If you want someone to do it with you, Peaje's agent (Pro plan) follows 20 of your business's questions in ChatGPT and Claude every week, suggests each page with its format and a model to follow, and measures each verified page again at 2 and 6 weeks. It costs US$30 a month for the first 50 businesses and US$69 after that. Peaje cannot promise you will get cited, and our own 0% today proves it.
Sources
- Peaje: how it measures whether ChatGPT and Claude cite a business (the November 7, 2026 measurement of usepeaje.com is internal)
- GEO: Generative Engine Optimization (Aggarwal et al., KDD 2024)
- Peaje pricing