You can rank #1 and still not be in the answer.
Buyers ask an assistant before they ask Google. Ranking and citation have come apart — the pages a model names are often nowhere near the organic top ten. This checks the only thing that matters now: when someone asks ChatGPT, Claude, Perplexity or Gemini the question your buyer actually asks, does your product get named, where in the list, and with a link back to you?
no account · no server · zero API keys in paste mode · MIT licensed, source on GitHub
Who the model should be naming
Aliases matter. Models write “Acme Billing”, “Acme.io” and “acme” in the same answer.
The questions your buyer asks before they ever see your pricing page
Keywords are short. Prompts carry the buyer, the constraint and the stack. Fill the blanks and generate a starter set, then edit freely.
Pick how the answers get collected
You ask the question in the real consumer apps — the same ones your buyer uses — and paste the answer back. Nothing is sent anywhere, and you get the personalised, logged-in answer an API would never show you. Open the grid below, click any square, copy the prompt, paste the reply.
The answer grid
One square per prompt per engine. The number is a 0–100 visibility score: being named at all is most of it, then how early you appear, whether your domain is linked, and how the answer talks about you.
Add a brand and at least one prompt to build the grid.
What the grid says
Run a check to see the leaderboard.
Domains found inside the answers show up here — this is the third-party layer worth winning.
The fix, generated from what you entered
A missing square is usually one of three things: the crawlers are blocked, the page buries the answer, or nothing on the page is extractable as a fact. These are the files that fix the first two.
llms.txt
Fill in step 1 and 2.
robots.txt — let the answer engines in
Fill in step 1 and 2.
Schema for your comparison page
Fill in step 1 and 2.
Answer-first page skeleton
Fill in step 1 and 2.
Fair question: why would you trust an unknown page with an API key?
You shouldn't, on trust alone. So the answer here is that you don't have to take anyone's word for it — six ways down, in order of how paranoid you want to be.
Use no key at all
Paste mode scores the answers you collect by hand. Every number on this page works without a key — and it reads the logged-in consumer answer, which is closer to what your buyer sees than an API response.
Take the page off the internet
This is one HTML file with no build step and no backend. Save it, open it from your own disk, and the "site you don't know" is now a file you own. Nothing about it changes.
Read it in a minute
The repo is public and the whole app is one unminified file. Search it for fetch( — four call sites, all named below, no bundler hiding them. Diff the commit against what your browser received.
Let the browser enforce it
The file ships a content security policy that allows exactly four network destinations. If this page tried to post your key anywhere else, Chrome would refuse the request, not ask permission. Search the source for Content-Security-Policy.
Watch every request live
Every outbound call is logged below as it happens, with its destination and size. Calls to anything not on the allowlist are refused in code and logged in red before the network is touched.
Make the key worthless
Cut a fresh key with a $1 cap, run the check, revoke it. Ten minutes of exposure on a key that can spend a dollar is a different risk from your production key, and it costs you one click.
Network log
allowlist: 4 hostsIf your policy says keys never touch a browser
Deploy this on your own domain, put the real key in an environment variable, and point the endpoint field at your worker. The page then never sees a secret at all.
Help me build more free tools
AEO trackers charge $99 a month for a version of this. Cited is free, open source, and has no paywall coming — no upsell screen, no seat limits, no "contact sales" tier. If it saved you a subscription, a tip funds the next tool. If it didn't, use it anyway and tell someone it exists.
I build small, single-file tools that run without a backend — the kind you can read end to end, host yourself, and keep using after the company behind them loses interest. Cited is the first one. There will be more.