What does Cite42 actually return?

Structured JSON. Every endpoint takes a prompt set, an optional brand and competitor list, and a list of AI models (ChatGPT, Claude, Perplexity, Gemini). It returns mention rate, average position, share-of-voice, citations, and the raw model responses. You decide what to do with it: chart it, alert on it, feed it to your own LLM, or stick it in a spreadsheet.

How is this different from a SaaS dashboard like Profound or Otterly?

Those tools sell a UI. Cite42 sells the underlying data. If you want a chart, build it. If you want alerts, wire them up. We do not compete with you on the surface, we expose the data layer that surface needs. Most of our customers are agencies and builders who want the data inside their own tools, not yet another login.

What does a single API call cost?

AI-model endpoints bill per model queried, so you pay for exactly the models you ask for: from ~$0.05 per model per call (a full four-model sweep is roughly four times that), and usage is free. Classic data endpoints (keywords, trends) are a flat per-call rate. The catalogue lives in apps/web/src/shared/lib/pricing.ts and the cost of every call is returned in the response so you can budget per request. Credits never expire.

Why pay-as-you-go instead of monthly seats?

Because AI search visibility is a workflow you run intermittently, weekly reports, ad-hoc audits, an MCP query inside Claude. Paying $99 per seat to make a call once a week is silly. You top up $25 and call the API as much as you actually need. If you need a few hundred calls a month, you will pay tens of dollars; if you need ten, you will pay a couple of dollars.

Can I use Cite42 inside Claude Desktop?

Yes. Cite42 ships an MCP server (@cite42/mcp) you install with one command. Once it is connected, Claude can call any Cite42 endpoint directly inside a conversation. Ask "How is my brand ranking on ChatGPT this week?" and Claude calls the rankings endpoint, formats the result, and shows it. The same credit balance powers MCP and direct API calls.

Do I need to bring my own OpenAI / Anthropic keys?

No. Cite42 manages the upstream model accounts, rate limits, and prompt formatting. You pay one rate per call regardless of which model answered. That is the whole point: we run the messy bit so the data shows up as JSON in your codebase.

Why we built Cite42: an API for AI search data

A year ago I started keeping a spreadsheet of every time a customer or a friend asked ChatGPT or Perplexity a buying question in front of me. The list got long. The pattern was always the same: the person typed the question, scanned the answer, and acted on the three brands the model named. Whatever was in those three brands controlled the rest of the conversation.

That is a new distribution channel. It is also a measurement nightmare. Nobody is going to log into a SaaS dashboard every week to find out whether ChatGPT recommends them. They are going to want the answer in the tool they already use, Slack, a weekly report, their own internal app, and they want it as data, not as a chart somebody else designed.

In one sentence

Cite42 is one REST API and one MCP server for AI search data. You pay per call, top up $25 at a time, and get clean JSON back. No seats, no dashboard you have to scrape, no minimums. Stripe-style billing for the AI search era.

The problem: AI search data is locked inside chat UIs

AI models now sit on top of a meaningful slice of the buying journey. ChatGPT handles roughly a billion queries a day. Perplexity is the default research tool for an entire generation of analysts. Gemini intercepts the long-tail of informational queries that used to feed your blog. Inside all of these, the question buyers ask sounds like "what is the best X for Y," and the answer cites three brands.

The problem is that nobody built these systems to be measured. ChatGPT does not expose a ranking endpoint. Perplexity does not publish citation logs. Gemini renders answers server-side and only partially appears anywhere you can audit. Every team that wants to know how they show up has had to roll their own scraper, deal with rate limits, deal with prompt drift, and rebuild it every time a model updates.

The work is mechanical and unrewarding. It is also exactly the kind of work a small API can absorb on behalf of everyone who needs it.

What Cite42 does

Cite42 exposes one HTTP endpoint per question your team actually asks about AI search. Every endpoint takes the same shape: a prompt set, an optional brand and competitor list, and a list of models. It returns structured JSON.

Brand rankings
Mention rate, average position, and share-of-voice for your brand across ChatGPT, Claude, Perplexity, and Gemini. On-demand or scheduled.
Citation tracking
Which URLs each model cites for your prompt set, and how often. The first step before you optimise content for AI search inclusion.
Competitor compare
Side-by-side rankings for your brand plus up to 10 competitors, with a co-mention graph, in a single /v1/compare call across all four models.
Brand sentiment
How each model describes you, positive, neutral, negative, with the exact phrases, in one /v1/sentiment call. Re-run it on a schedule to track tone over time.
Prompt monitoring
There's no separate monitoring product to buy: cron a /v1/rankings or /v1/compare sweep and diff the runs, or tell your MCP agent to watch a prompt set and ping you when results move. You own the trigger; you pay per call for the data.

Every response includes the raw model output, the parsed structure, and the cost of the call to five decimal places. You can stream the same data into a dashboard, a Slack alert, or your own LLM agent. We do not care which.

Why an API, not a dashboard

There are already three or four well-funded companies selling AI search analytics dashboards. They are fine. They are also the wrong product for everyone we talked to in the first month.

The pattern was consistent. Agencies wanted to embed AI search data inside the reports they already deliver to clients. Founders wanted the data inside their own admin UI so their team could see it without learning another tool. Builders wanted the data inside the agent they were building, so the agent could reason about it. Nobody wanted to log into a fifth dashboard. They wanted the data, and they wanted it as JSON.

That is the bet. The next layer of AI search tooling is not another dashboard. It is the infrastructure that the dashboards, agents, and reports sit on top of. We would rather be the Stripe of AI search than the Mailchimp.

The honest comparison:

	Dashboard product	Cite42 (MCP + API)
Surface	Their UI. You log in.	Your UI. You call the endpoint.
Pricing model	Seats, monthly minimums, annual contracts	Pay-as-you-go from $25. Credits never expire.
Data ownership	Their database. Export sometimes available.	Your codebase. Every call returns full JSON.
Best fit	Marketing manager who wants one chart	Builder, agency, or team wiring AI search into their own tooling

How pricing works

1 USD = 100,000 micro-credits. Every endpoint has a per-call cost, billed to five decimal places, debited from your balance the moment the call returns 2xx. The minimum top-up is $25. Credits never expire.

We picked this model because AI search visibility work is bursty. You run a competitive audit before a quarterly board meeting. You set up weekly prompt monitoring for your three priority queries. You spend $4 in March, $40 in April when you launch a campaign, $1.20 in May. Paying a $99 monthly seat for that workload is absurd. Paying for what you actually call is not.

The full catalogue lives in our pricing page. The short version is that AI-model endpoints bill per model queried, from ~$0.05 per model per call depending on the endpoint, with cached repeats free. The cost of every call is returned in the response so you can budget per request, set per-tenant ceilings, and show your end users what their query is going to cost before they run it.

MCP for Claude Desktop

The same product ships as a Model Context Protocol server. One install command in Claude Desktop, paste your Cite42 API key, and Claude can call the endpoints directly inside a conversation.

The use case that sold us on MCP was a Friday afternoon question. A founder we were working with asked Claude, "has my brand position on ChatGPT changed this week?", and Claude, with the Cite42 MCP installed, called the rankings endpoint, diffed the result against the previous run, and answered in two sentences. No dashboard, no SQL, no Slack alert. Just the question, the answer, and the data in between.

MCP and the REST API share the same credit balance and the same API key. If you call one, you have already integrated the other.

Who this is for, and who it is not

Cite42 is built for builders. Founders wiring AI search into their own admin UI. Agencies embedding the data into client reports. Indie hackers shipping AI-search-adjacent products who want the underlying signal without rebuilding the scraping layer. Internal teams at SaaS companies who want a few hundred calls a month inside their existing dashboard, not another login.

Cite42 is not the right fit if you want a polished dashboard with charts, alerts, and a team seat for every marketer. There are good products that do that, and we will happily tell you which ones. Cite42 exists for the layer underneath those products.

You can grab an API key with $1 free and call any endpoint in the next ten minutes. If the JSON looks like what you want in your codebase, the rest is just a matter of how many calls you make.