How We Plan Citation Sources for SaaS Clients in 2026
GitHub, Microsoft Learn, G2 and vendor docs all show up in Promptwatch's software citation data. Here is how we turn that source mix into a plan for a SaaS client.
In ChatGPT's answers to organic software prompts, GitHub took 1.2% of all citations and learn.microsoft.com took 0.98%. G2 followed at 0.73%. Those are figures from Promptwatch's software citation share report, published on October 6, 2026, covering September 28 to October 4, 2026. A SaaS founder reading that list tends to notice one thing: the vendor's own marketing site is nowhere near the top.
That is a useful reaction, but it needs handling carefully. We use this report as the starting point for the source-planning conversation with SaaS clients. This is how that conversation goes, and what we do after it.
What the report shows
The ranking in the chart: GitHub 1.2%, learn.microsoft.com 0.98%, G2 0.73%, AWS docs 0.41%, Google Cloud docs 0.35%, Gartner 0.34%, Capterra 0.32%, Intercom 0.28%, Zapier 0.24%. Grouped by company, the report puts Microsoft at 1.35% across three domains, Google at 0.98% across four, AWS at 0.69% across two and OpenAI at 0.42% across two.
The 23 charted domains together account for 7.89% of all software citations. Split by source type: documentation and help centers 2.64%, vendor sites 2.01%, review sites and analysts 1.57% (G2, Gartner, Capterra, Software Advice), GitHub 1.2% and app stores 0.47%. Vendor pages and docs together reach 4.65%, about three times the share of review sites and analysts. The four review and analyst sources, combined at 1.57%, still exceed any single vendor in the chart, Microsoft included.
What it does not show
One engine, one week, one vendor's prompt set. The report covers organic software prompts in ChatGPT and says nothing about Perplexity, Google's AI answers, or non-English markets, and we do not stretch it that far. It also does not explain why GitHub or Microsoft Learn rank where they do. Our best reading is that developer-facing documentation is plentiful, structured and public, but that is our inference, not something the report claims.
There is also a scale issue. The biggest single domain holds 1.2%. Software citations are spread thin, and the 23 named domains hold under 8% between them. A mid-size SaaS vendor is therefore not fighting for a slice of a few dominant sites. It is fighting for a place in a long tail.
The plan we build from it
The report's advice is plain: publish crawlable docs and help pages, keep a public GitHub presence for developer tools, and keep G2 and Capterra profiles complete. We attribute that to the report, and then we sort the client's work into three source groups.
Owned documentation. For most SaaS clients this is the highest-leverage group because it is entirely under their control. We check that docs and help pages are server-rendered, not hidden behind a login, and not blocked for AI crawlers by robots.txt or a firewall rule. We read the docs for plain statements of what each feature does, what it integrates with, and what it costs to use. A help center that answers the question directly is easier for anyone to quote.
Third-party review profiles. G2 and Capterra are in the chart, so a half-finished profile is a visible gap. This is housekeeping: accurate category, current feature list, real screenshots, a recent product description. We do not suggest nudging customers into reviews in ways the platforms forbid, and we do not write reviews.
Public code and community presence. If the client sells to developers, GitHub is the largest single domain in the data. That means a public repository with a clear README, examples, and an SDK is worth maintaining. If the client has nothing public, we do not invent a repo for the sake of a chart. It is a product decision for their engineering team.
Vendor marketing pages sit in the middle. They hold 2.01% as a source type in this report, so they are worth keeping accurate and specific, but we do not treat them as the only lever.
Sequence matters
We start with access, then content, then third parties. If the docs cannot be fetched, rewriting them is wasted effort. If the docs are fine, the profile cleanup comes next because it is quick. Only after that do we suggest anything that needs outreach or engineering time.
This order also keeps the plan honest about uncertainty. Nothing here guarantees a citation. We are removing obstacles and filling visible gaps, and then we look at what the engine does.
How we measure it
The measurement step is where we use Promptwatch, the platform we run client programs on. We agree a set of tracked prompts with the client, in categories like comparisons, integrations and how-to questions, and then read the citation analytics at domain and page level. That shows whether the client's docs, a G2 profile, or a GitHub repo is the source that ChatGPT draws on for each prompt, and which competitors appear in the same answers.
Page Tracker lets us follow a specific URL, theirs or a competitor's, so we can see which responses cite it. For docs, that is the clearest view of which pages are working as sources. When a doc page is never cited, the query fan-outs view shows which sub-queries ChatGPT ran, and we check whether the page answers any of them.
Agent Analytics adds the crawler side. Crawler logs start on the Professional plan at $245 a month, and Essential does not include them. If a client's docs are not being fetched, the logs show the failing status codes and which bots were turned away. Then the fix is infrastructure, not copywriting.
The takeaway for a SaaS client
The report suggests that docs, review profiles and GitHub carry weight for software prompts, and that the vendor homepage is not the only way in. We take that to mean a SaaS source plan should cover three places, in order of control. Then it needs a measurement loop that tells us, prompt by prompt, whether any of the work moved where ChatGPT looks. One window of data is a starting hypothesis, and we treat it like one.