Does ChatGPT mention you?
And does it cite you?

FetchLayer's AI Visibility API returns ChatGPT answers, the brands they mention, the pages they cite and your rank among the recommendations as JSON over REST or MCP. A request is one ChatGPT answer, and each request costs one credit: $1.99 per 1,000 requests pay-as-you-go, or $25 a month for 15,000, with 30 free requests and no card to start. A request that fails is not billed. The same key covers 36 other platforms, and every route is also a tool on FetchLayer's MCP server at https://mcp.fetchlayer.dev, for Claude, ChatGPT, Cursor and other MCP clients.

It answers the question SEO tools answer for Google, but for AI assistants: when someone asks ChatGPT the question your buyers ask, does the answer name you, does it cite your pages, and where do you sit among the options it recommends? Send a prompt and your brand, and you get that back as structured JSON — mention counts, the sentence around each mention, the citations, your rank among the recommendations, and the competitors the answer put forward instead. This is the measurement side of what SEO now calls answer engine optimization or GEO.

They are two different outcomes with two different fixes, and most AI brand monitoring collapses them into one number. An answer can name your brand in its prose while none of your pages are among the sources it links — you are part of the consensus but not the evidence. Or it can cite one of your pages without recommending you at all — your content is being read, but someone else gets the recommendation. Every route reports both separately: mentioned beside citedAsSource, and in a rank check, rank among the recommendations beside citationRank among the cited sources. There is a third state worth watching, retrieved: ChatGPT read one of your pages while searching and then cited someone else.

No. You send your FetchLayer key and a prompt, and you get back what ChatGPT answers a visitor with no account — the real consumer-facing answer, not an API completion. There is no model, temperature or system prompt to choose, because a visitor has none of those either. FetchLayer is not affiliated with, endorsed by, or sponsored by OpenAI.

By measuring the variation instead of pretending it is not there. Ask the same prompt once and you have an anecdote; ask it several times and you have a rate. /ai-rank takes a runs count and reports mentionRate, averageRank, bestRank, worstRank, citationRate and averageCitationRank across those runs, with the same figures for each competitor. A brand mentioned in 5 runs out of 5 is in a different position from one mentioned in 1 — and a single answer cannot tell those apart. Keep the prompt, country and web-search mode fixed when comparing week to week, and read the aggregate rather than any one run.

Start with /prompt-discovery. Give it a topic or your brand name, and it returns the prompts it asked, the web searches ChatGPT actually ran while answering them, and prompts it suggests people use. The two lists are labelled and kept apart, because they are different kinds of evidence: observed_search_query means ChatGPT really ran that search, so those are the queries your pages have to rank for to be readable by it; generated_prompt means ChatGPT wrote that prompt when asked, which is a suggestion, not an observation. Feed the ones that matter into /ai-rank.

Yes — /visibility-report takes a list of prompts and a list of brands and returns share of voice per brand, plus a per-prompt breakdown of mention rate, average rank and citation rate. Extra brands are free, because every brand is read out of the same answers; it is the prompts and runs that cost credits. The request is bounded by answers: prompts × runs must be at most 10, so ten prompts at one run each is fine. For wider tracking, run several smaller reports on a schedule and store the results.

The answer as Markdown and as plain text, the pages it cites — each with position, URL, title, domain, the label shown on the citation, and the sentence that cites it — every inline citation, the web searches it ran, and every page it looked at while searching, each flagged with whether it was cited. That last list is useful on its own: a page can be read and never cited. Null means the answer did not show a field, not zero.

One credit per answer ChatGPT generates, which is exactly what pagesScraped reports. /ask and /mentions cost 1; /ai-rank costs its runs; /prompt-discovery costs one per prompt asked, plus one more while it also asks for suggested prompts; /visibility-report costs prompts × runs. Matching, ranking and aggregation cost nothing extra, and an answer that could not be produced is never billed.

Every answer is a real ChatGPT session, so a single answer takes a few seconds rather than milliseconds. A request’s answers are generated alongside each other rather than one after the other, so asking for more runs costs more credits but little extra time — a five-run rank check and a full ten-answer visibility report both finish well inside a minute. Give your HTTP client a generous timeout anyway.

You cannot turn web search off: a visitor without an account has no such setting, so webSearch is auto or on, and usedWebSearch reports what it did. Some answers state which model produced them and some do not, so model is null rather than guessed. Some answers do not show the searches behind them, and then searchQueries and retrievedPages are null rather than empty, with notes saying so. And nothing here measures traffic, spend or clicks — it measures what ChatGPT answers, not how many people saw it.

Yes. All five routes are MCP tools on the same FetchLayer MCP server, named chatgpt_ask, chatgpt_mentions, chatgpt_ai_rank, chatgpt_prompt_discovery and chatgpt_visibility_report. Tool calls use the same key and are billed exactly like REST, so an agent running a five-run rank check still pays five credits.

ChatGPT today. Every route takes a provider field, which is chatgpt by default and currently its only value, so the same requests will serve further assistants as they are added rather than needing a different shape.