Developers stopped starting with a search box some time ago. The question goes to an assistant, and the answer uses your library or somebody else's — usually before anyone visits your homepage.
Who is asking
Two moments matter, and only one of them looks like buying. The first is selection — which library, which platform, which managed service. The second is implementation, where the assistant writes code against whatever it believes your API looks like.
The second moment is the one most vendors ignore, and it is the one that compounds. A model that reliably produces working examples against your SDK is a distribution channel; a model that produces plausible code against an API you deprecated is a support queue.
Questions buyers actually ask
Not keyword variants — whole questions, the way a person types them into an assistant. These are the shapes we build tracking around.
- what is the best way to add authentication to this framework
- which queue should I use for background jobs in Node
- how do I paginate this API
- does this SDK support streaming responses
- why am I getting a 429 from this endpoint
- is there an open-source alternative to this managed service
Which surfaces matter here
Developer questions are unusually likely to be answered without a search at all, because many of them are stable knowledge the model already carries. That single fact reorganises the priorities: your documentation has to be right in the live index and right in whatever the model learned, and you control neither of those directly.
When a search does run, the sources are narrow and familiar — official docs, the repository, question-and-answer sites, changelogs. There is less roundup content between you and the answer than in almost any other category, which cuts both ways.
Reference pages for these surfaces
Where these answers come from
The pages an assistant leans on in this category are rarely all yours. Each block below is a source family worth checking before you write another article.
Official documentation
The primary source for everything an implementation question needs. It is also frequently on its own subdomain, with its own robots.txt, its own sitemap, and a client-side renderer that returns an empty shell to anything that does not execute JavaScript.
The repository
READMEs, examples, issues and release notes are public, heavily linked and heavily read. An issue thread documenting a workaround is often the most retrievable answer to a real error message.
Question-and-answer sites
A decade of accumulated answers, many of them about versions you no longer ship. Correcting the record is slow; leaving it alone is worse.
Changelogs and migration guides
The only pages that state what changed and when. Without them, nothing has a way to distinguish your current API from the one it learned two years ago.
Developer community writing
Tutorials, newsletters and posts on developer platforms. These carry the idiomatic usage patterns that end up reproduced in generated code.
What the documentation says
Each line below is read from the linked document — a vendor's own guidance, a published standard, or a public register. Nothing in this section is our estimate.
OpenAI documents separate crawlers with separate jobs — OAI-SearchBot for surfacing sites in ChatGPT's search features, GPTBot for collecting content for model training — and states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. Blocking training and blocking search are two different decisions.
Source: OpenAI — Overview of OpenAI Crawlers (opens the source document in a new tab)Anthropic's developer documentation states that Claude searches when a request depends on information that is current, changing, or outside its training data, and answers directly when the request draws on stable knowledge. A large part of your documentation is treated as the second kind.
Source: Anthropic — Web search tool (developer documentation) (opens the source document in a new tab)The same documentation states that citations are always enabled for web search, and that each cited passage carries the source URL, the page title and a short excerpt of the text that was cited.
Source: Anthropic — Web search tool (developer documentation) (opens the source document in a new tab)robots.txt, standardised as RFC 9309, matches rules per user-agent and per path. A docs subdomain publishes its own file, so a permissive marketing site tells you nothing about what your documentation allows.
Source: IETF — RFC 9309: Robots Exclusion Protocol (opens the source document in a new tab)
What is not documented
Open questions, not omissions. Anyone answering these with a precise number for your category is guessing.
- No vendor publishes how official documentation is weighted against a question-and-answer site or a repository issue. When an answer prefers a five-year-old thread to your current docs, there is no documented lever.
- Training data is not observable. We can tell you what an assistant says about your API; we cannot tell you which version it learned, or when.
- Whether a search runs at all is a per-turn model decision, so an answer with no citations is not evidence that your documentation is missing from anything.
- Generated code is not a mention. Tracking reads answers, so a model that quietly writes correct calls against your SDK without naming you is invisible to this — and, as far as we know, to every other measurement method on the market.
What GetCited.me measures here
We run your selection and implementation questions on a schedule and record whether your project was named, which alternatives were named beside it, which URLs the answer cited, and how the mention was framed.
Because assistants often answer developer questions from stable knowledge without searching, the split matters more here than elsewhere: a mention with no citation is a signal about what the model already carries, and a citation is a signal about what your live pages can support.
Cited hosts are grouped, so you can see when the answer came from your documentation, from your repository, or from a third-party thread that outranks both.
Launch (Free) tracks Gemini only, with a one-time grant of answer credits and weekly automatic tracking. ChatGPT, Claude, Perplexity and Google AI Overviews are on Growth and Scale. Card payments are temporarily disabled while payment-provider onboarding completes, so every account currently runs on Launch — the pricing page lists what each plan includes.
What you can actually do
Concrete steps, each one checkable with a free tool or explained in the glossary.
Free tools for this industry
No account, no card — results in the browser.
Related terms
Keep reading
The reference sets behind this page.