Docs vs vendor sites vs G2: which source type ChatGPT cites most for software
Promptwatch's October 2026 data report splits ChatGPT's software citations by source type; here is the comparison, what it does not prove, and how to check your own category.
Quick verdict
For organic software prompts in ChatGPT, documentation and help centers took 2.64% of all citations in one week, ahead of vendor sites at 2.01% and review sites and analysts at 1.57%. That comes from Promptwatch's data report on software citations, which covers September 28 to October 4, 2026. Vendor pages and docs together reach 4.65%, about three times the review-site and analyst group.
So if you have to rank where a software company's effort goes first, the report points at what you publish yourself: crawlable docs and help pages, then product pages. Review profiles still matter, but they are a second job, not the first.
What the report measured
The dataset is organic software prompts run against ChatGPT in the real interface during a single week. Twenty-three domains appear in the chart, and together they hold 7.89% of all software citations. The rest of the citations are spread across a very long tail of other domains that the chart does not show.
That framing matters. A 2.64% share sounds small, and it is, because software answers cite a wide field of sources. The useful signal is the relative size of each group, not the absolute number.
The comparison table
| Source type | Share of all software citations | Domains the report names for the group |
|---|---|---|
| Documentation and help centers | 2.64% | Not itemised in our source; the report's top domains include learn.microsoft.com (0.98%), AWS docs (0.41%) and Google Cloud docs (0.35%) |
| Vendor sites | 2.01% | Not itemised in our source |
| Review sites and analysts | 1.57% | G2 (0.73%), Gartner (0.34%), Capterra (0.32%), Software Advice |
| GitHub | 1.20% | github.com |
| App stores | 0.47% | Not itemised in our source |
Source: Promptwatch data report, September 28 to October 4, 2026, ChatGPT, organic software prompts. The five groups add up to the 7.89% total for the 23 domains.
Two other points from the same report. GitHub alone, at 1.2%, tops the named domains, ahead of learn.microsoft.com at 0.98% and G2 at 0.73%. And the four review and analyst names together (G2, Gartner, Capterra, Software Advice) hold 1.57%, which is more than any single vendor including Microsoft. The report also groups big tech across several domains: Microsoft at 1.35% over three domains, Google at 0.98% over four, AWS at 0.69% over two and OpenAI at 0.42% over two.
What this does and does not mean
It is one vendor's dataset, for one week, in one engine, in English-language organic software prompts as far as we can tell. It describes where ChatGPT's citations went, not why. A page that sits in a cited group is not thereby cited, and a review site that is cited rarely may still shape the buyer's shortlist through other channels.
Do not read "docs beat G2" as a law. The documentation group includes the giant cloud vendors, whose docs answer a huge number of how-to questions. A ten-person SaaS company does not have that footprint. The honest reading is narrower: ChatGPT, for software questions, leans heavily on material that explains how a product works, and less on third-party opinion about it.
What to do without any tool
The report suggests three practical moves, and they are cheap enough to start this week.
- Publish docs and help pages that a crawler can read. If your documentation sits behind a login, in a JavaScript app that renders nothing for bots, or in PDFs, an engine has little to cite. Check that the pages load as plain HTML and that robots.txt does not block the bots you care about.
- Keep a public GitHub presence if you sell to developers. A well-maintained repository README is a page about your product that you control.
- Keep G2 and Capterra profiles complete. Fill in categories, descriptions and pricing notes. The report's point is that the review group collectively out-cites any single vendor, so an empty profile is a gap.
Beyond the report, one manual test is worth an afternoon. Take ten prompts a buyer might type in your category, run them in ChatGPT, and write down every cited URL. Sort the list into docs, vendor pages, reviews and others. You will quickly see whether your category looks like the report's average or not.
Measuring it for your own category
Sorting citations by hand does not scale past a few prompts. Platforms with citation analytics do it continuously. In our catalog, Promptwatch is the one that documents domain-level and page-level citation analytics, Citation Trends that classify each citation by content type and source type, and a Page Tracker for specific URLs, yours or a competitor's. Citation Trends also reports first cited, ramp-up days, peak rate and decay half-life per page, which tells you whether a docs page is a durable asset or a short burst.
If you want to know whether your docs are reachable at all, the missing piece is crawler logs. Promptwatch's Agent Analytics shows IP-verified AI crawler requests, a crawl-to-citation path and error tracking. Be clear about the plan tier: crawler logs start on Professional at $245 a month, and Essential at $95 a month has none. The free Explore plan is limited to ChatGPT and 10 prompts, which is enough to try the prompt-and-citation side against the very engine this report studied. There is a 7-day free trial on monthly billing for Essential and Professional.
For the other tools in our catalog, such as AthenaHQ and Scrunch AI, we found no documented source-type classification of citations, so we cannot say they offer one. Check in a trial.
Who should pick which
- A developer-tool or infrastructure company: start with docs and GitHub. The report gives GitHub 1.2% of all citations and the top documentation domains are cloud vendors, so audit crawlability first.
- A mid-market SaaS company with thin documentation: write help-center pages for the questions buyers ask, then complete your G2 and Capterra profiles. Do both, in that order.
- A team that already has docs and profiles: move to measurement. Track which of your pages ChatGPT cites and which competitors share the same answers. Promptwatch is the tool we would test first for that, mainly because citation analytics and crawler logs sit in one product.
- A team with no budget: run the ten-prompt manual test monthly and keep a log.
For more on checking what the engines actually cite, see our guides to tracking citations in ChatGPT with a weekly audit, reading a crawl-to-citation path and monitoring third-party sites AI uses to recommend brands.
Bottom line
The software citation report is a useful map of where ChatGPT looked for one week, and it favors sources that explain products over sources that judge them. No change to your docs or profiles guarantees a citation, and the shares will move. Measure your own prompts, then decide.