Docs beat vendor sites in ChatGPT's software citations: a field note
Promptwatch's October 2026 data report puts documentation ahead of vendor sites for software prompts in ChatGPT. What that does and doesn't mean, and what to check on your own docs.
In Promptwatch's data report on software prompts in ChatGPT, documentation and help centers took 2.64% of all citations. Vendor sites took 2.01%. Documentation came out ahead of the product websites that pay for the design, the copywriters and the ad budget.
That is the finding we keep coming back to in software audits, so this note covers what the number says, what it can't say, and the checks we'd run on a docs site before touching a homepage.
What the report measured
The window is September 28 to October 4, 2026. The prompts are organic software prompts, run in ChatGPT's real interface. The report charts 23 domains, which together account for 7.89% of all software citations. Everything else sits in a long tail the chart doesn't show.
Inside that 7.89%, the report groups domains by source type:
- Documentation and help centers: 2.64%
- Vendor sites: 2.01%
- Review sites and analysts (G2, Gartner, Capterra, Software Advice): 1.57%
- GitHub: 1.2%
- App stores: 0.47%
Vendor pages and docs together reach 4.65%, about three times the share of review sites and analysts. The top individual domains are GitHub at 1.2%, learn.microsoft.com at 0.98%, G2 at 0.73%, AWS docs at 0.41% and Google Cloud docs at 0.35%.
What it does not tell you
One week, one vendor's dataset, one engine. The report describes a pattern in that window. It doesn't prove that docs cause citations, and it doesn't say a page called "docs" is cited because of anything you can copy.
Two other limits matter for a smaller software company.
First, the documentation domains named in the chart belong to Microsoft, AWS and Google. Big platforms have docs that developers search for by product name and error message. A mid-size SaaS vendor shouldn't read 2.64% as its own expected share. The report doesn't split small vendors from large ones.
Second, the report doesn't tell you whether a homepage in particular is cited less than a docs page. It compares source types, not individual pages. So "docs before your homepage" is our reading of the gap between docs and vendor sites, and we'd hold it loosely.
What we'd take from it is narrower and more useful: for software questions, ChatGPT seems to lean on pages that explain how a thing works, and the docs are often the only place on a vendor's domain that does so in plain terms.
What to do without any tool
The report's own advice is short: publish crawlable docs and help pages, keep a public GitHub presence if you sell to developers, and keep G2 and Capterra profiles complete. Here is how we'd turn that into a week of work.
Open your docs logged out. Use a private window with no cookies. If you hit a login wall, a cookie gate or an empty shell that only fills in after scripts run, assume a crawler sees something similar. Fix that before anything else.
Check robots.txt and any bot rules for the docs subdomain. Docs often live on docs. or help. hosts with their own robots.txt and their own CDN rules, set up by a different team years ago. A rule written for scrapers can quietly shut out AI crawlers. Our notes on robots.txt rules for AI crawlers cover how to loosen it carefully.
Write page titles that match how buyers ask. Docs titled "Overview" or "Getting started" tell nobody what the page answers. A title like "How to export reports to CSV" does. This is ordinary information design, nothing exotic, but old docs are full of vague titles.
Put the commercial facts where the docs can reach them. Limits, supported integrations, security basics and plan differences are things buyers ask about. If those facts live only in a sales deck, the docs can't carry them.
Keep docs current. An outdated page that says a feature doesn't exist will be read as true by anyone, human or model, who finds it. Date your pages and retire the stale ones.
Look at the review sites too. The report's 1.57% for G2, Gartner, Capterra and Software Advice is larger than any single vendor in the chart, Microsoft included. Complete profiles are cheap. We wrote about whether review sites help a software brand get recommended by ChatGPT.
Measuring it on your own prompts
The report is an average. Your question is whether your docs, your vendor pages or someone else's pages get cited for the prompts your buyers use. That needs your prompts, tracked over time, with the cited URLs recorded.
This is where we'd use Promptwatch. Its citation analytics work at page and domain level, so you can sort the cited URLs for a prompt set and see how many belong to your docs host, how many to your main site, and how many to a competitor or a review site. Page Tracker lets you follow specific URLs, yours or a rival's, for example your most important docs page against the competing page that keeps being cited instead.
The diagnostic step after that is Agent Analytics, the IP-verified AI crawler logs. If a docs page is never cited, the logs tell you whether the crawlers are even fetching it, and what status code they get. Our write-up on the crawl-to-citation path explains how to read a page that is fetched often but never cited.
One plan note, because it changes the workflow. Crawler logs start on the Professional plan at $245 a month. Essential, at $95 a month, has none, so on that plan you'd lean on citation analytics and check your CDN logs by hand.
A reasonable first month
Week one: log out, open the docs, fix whatever blocks a crawler. Week two: rewrite the vaguest titles, starting with the pages people reach most. Week three: tighten the review profiles. Week four: pull the citation data for your tracked prompts and see which of your pages show up.
Then compare against the report's pattern, but only as a sanity check. If your docs aren't cited and the logs say crawlers never reach them, that's a fault you can fix. If crawlers reach them and the pages still aren't cited, the content is the question, and that's a longer conversation.
We wouldn't promise any of this moves a citation. We'd only say the cost of checking is low, and the docs are the cheapest place to start.