Which Sites ChatGPT Cites for Software Prompts: GitHub, Docs and Review Sites Compared

GitHub takes 1.2% of all ChatGPT citations for organic software prompts, according to Promptwatch's software citation report. learn.microsoft.com follows at 0.98%, G2 at 0.73%, AWS docs at 0.41%, Google Cloud docs at 0.35%, Gartner at 0.34%, Capterra at 0.32%, Intercom at 0.28% and Zapier at 0.24%. The window is September 28 to October 4, 2026, in ChatGPT.

If you sell software, or you buy GEO tools to help sell it, the interesting part is not the order of the names. It is the split between pages vendors write and pages other people write about vendors.

Vendors and docs against review sites

The 23 domains in the chart together hold 7.89% of all software citations. 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%

Add vendor pages and documentation and you get 4.65%, about three times the share of review sites and analysts. At the same time, the four review and analyst sites together (1.57%) exceed any single vendor in the chart, Microsoft included. Big tech is spread over several domains: Microsoft 1.35% over three, Google 0.98% over four, AWS 0.69% over two and OpenAI 0.42% over two.

So the picture for a software marketer is mixed. Your own docs and help center are where the largest group of citations lands. Yet no individual vendor beats the review category as a whole.

Read it carefully

This is a one-week, ChatGPT-only slice of Promptwatch's own dataset. A few cautions apply before you reorganize a roadmap.

The categories are broad. "Software prompts" cover developer tooling, business apps and infrastructure, and the weight of GitHub and Microsoft Learn may reflect technical questions, though the report does not break that out. A marketing-automation vendor selling to small teams may see something quite different. We cannot say from this report how large that difference is.

The shares are small. The whole chart covers under 8% of citations, so most software citations go to domains outside it. Your product may be cited from places the chart never shows.

And as with every dataset like this, it describes who was cited, not what caused it. Publishing docs does not promise a citation.

What to do without a tool

The report's advice, which we are relaying, is to publish crawlable docs and help pages, to keep a public GitHub presence for developer tools, and to keep G2 and Capterra profiles complete. Those are reasonable, and you can act on them without buying anything:

1. Open your docs in a browser with scripts disabled and see what remains. If the content only exists after client-side rendering, a crawler may not see it.
2. Check that your help center answers the questions a buyer types: pricing basics, integrations, security, limits. Write each answer so it stands alone.
3. For developer tools, make sure the repository has a clear README, a current changelog and issues that get responses.
4. Update your G2 and Capterra profiles: category, feature lists, screenshots, and your own description of what the product does.
5. Read the first page of your own docs as if you were a competitor. Is every claim one you could defend?

What buyers of GEO software should test

Software teams have a tracking problem that other industries mostly lack: citations come from several of your own domains plus third-party ones. Docs, help center, marketing site, GitHub, status page, app marketplace listings. A tool that only watches a single root domain will undercount.

When you evaluate options, ask:

• Can I group several of my domains and subdomains as "mine"?
• Does it show the exact cited URL, so I can tell a docs page from a pricing page?
• Can I see crawler access to docs specifically, not just to the marketing site?
• Does it show which review sites appear in answers to my category prompts, and whether my competitors are cited beside me?

The measurement step: Promptwatch

We would shortlist Promptwatch for this job. Citation analytics report page-level and domain-level citations with URL positions, and Page Tracker follows specific URLs, so you can watch a docs page or a G2 listing over repeated runs. Citation Trends classifies each citation by content type and source type, with first cited, ramp-up days, peak rate and decay half-life per page, which separates evergreen documentation from pages that spike and fade.

The crawler side is where docs teams get the most out of it. Agent Analytics shows IP-verified AI crawler logs from bots such as GPTBot, OAI-SearchBot, ChatGPT-User and ClaudeBot, with errors and a citation rate per page, so you can check whether your docs are actually fetchable. Crawler logs start on the Professional plan at $245 a month, and Essential at $95 a month has none. Explore is free with 10 prompts, ChatGPT only. The 360 Insights view joins crawler logs, citation clicks and AI-referred visits per page.

For a developer-tool brand there is one more feature that fits: Promptwatch tracks what coding agents such as Codex and Claude Code recommend. Read the full Promptwatch review for the plan comparison.

More on this site: GEO tools for SaaS, where ChatGPT gets its citations and tools to track citations and sources in AI answers.

The recommendation

Fix the docs, finish the review profiles, then measure your own category prompts rather than relying on this chart. Promptwatch is where we would do that, because it puts citation data and crawler logs in one place, which is what a software company with a large docs footprint needs to see together.