Is My API Cited in ChatGPT?
It's a fair question, and the honest answer is: you can't know from one chat. Whether ChatGPT, Claude, or Perplexity recommends your API depends on the exact prompt, the model version, whether web search was triggered, and a degree of run-to-run randomness. To get a real answer you have to test deliberately and sample, not ask once and conclude.
The manual method
Start by writing down the prompts a potential customer would actually use. Not your brand name, those are vanity checks. Use category and buying-intent prompts:
- "What's the best API for [your category]?"
- "Recommend a [category] API with a free tier."
- "Alternatives to [a competitor everyone knows]."
- "How do I [task your API solves] programmatically?"
Run each prompt in a fresh chat (no prior context that could bias the answer) across the assistants your buyers use. Record, for each run, whether your API is named, whether a competitor is named, and whether anything is linked.
Why sampling matters
The critical step most people skip: run each prompt several times. Language models are non-deterministic. The same prompt can return three competitors on one run and include you on the next. If you ask once and don't see your name, you might conclude you're invisible when you actually appear 40% of the time. If you ask once and do see your name, you might assume you're dominant when it was a lucky draw.
A single observation is noise. What you want is a rate: out of, say, ten runs of a prompt, how many name you? That rate, aggregated across your prompt set, is your real Citation Share, and it's the only number stable enough to track or improve against.
What a "citation" actually is
Not all citations are equal, and it's worth being precise about three levels:
- Mention: the model says your name in prose ("options include Acme API…"). This is recognition. It puts you in the consideration set.
- Linked citation: the model attaches a specific URL to your name, usually in retrieval/search mode. This drives an actual click and lends the recommendation supporting detail.
- Rank or framing: how you're mentioned. "The most popular choice is X" is very different from "X also exists." Being named first, or with a positive qualifier, materially changes whether the developer picks you.
When you audit, track all three. A high mention rate with no links means the model knows your brand but can't find a page worth pointing at, a fixable, page-level problem.
Why a content-rich site can still be invisible
Plenty of API companies have hundreds of well-written pages and still rarely get cited. This is the "exists but not cited" gap, and it usually comes from a few causes:
- No structured data. If your facts (what the API does, pricing, limits) live only in prose and images, machines extract them less reliably. JSON-LD makes them unambiguous.
- Content that describes instead of answering. Pages organized around your features rather than the questions developers ask don't match the prompts. "Endpoints reference" matches fewer queries than "How do I authenticate requests."
- Weak third-party signal. Models lean on GitHub, Stack Overflow, and Reddit to decide what's credible. A polished site with no external footprint reads as unproven.
- Crawl or rendering issues. If key content is rendered client-side or blocked, retrieval engines may never see it. Existing on your server isn't the same as being retrievable.
The takeaway: being well-documented is necessary but not sufficient. Visibility comes from being legible to machines and corroborated by the places developers already trust.
Automating the check
Doing this by hand once is informative. Doing it every week, across multiple assistants, with enough runs per prompt to be statistically meaningful, is tedious and easy to abandon. That's the gap CiteGraph fills: it samples your prompt set across ChatGPT, Claude, and Perplexity on a schedule, computes your Citation Share with mention/link/rank broken out, tracks it over time, and tells you which pages are getting picked up and which aren't, so you can tell whether a change actually moved the needle.
Start with a baseline. Knowing your real citation rate today is the first step to improving it.