Prompt tracking — the buyer questions, asked properly
Everything in CiteGraph starts from the question set. We mine the buyer questions your category actually gets asked — from your own site and the way engines phrase them — then run each one repeatedly on every engine, every week. Each question carries its own score, its per-engine record, and the brands winning it, so the tracked list is a working asset, not a keyword dump.
How it works
Mined, not guessed
The first scan reads your site and drafts the questions a buyer in your category asks — comparisons, alternatives, pricing, use cases. You edit the set anytime; the next scan follows it.
Every question, repeated runs
Each question runs 3–5 times on each of 4 engines. One run is an anecdote; the spread across runs is what makes the score on each question trustworthy.
Scored where you manage it
The tracked list shows a score ring per question, a named ✓/✗ with run counts, and the exact brands the answers cite instead of you — so pruning and adding questions is an evidence-based edit.
In depth
Per-engine record on every question
Expand any question to see how each engine treated it: named 2/5 on Perplexity, 0/5 on Gemini, 0/5 on ChatGPT-family reads very differently from a flat 13% — and tells you which engine's sources to go work on.
Questions by plan, honestly priced
Every plan tracks 10 questions per project; Growth and Scale ask each one 5 times per engine instead of 3. The limit exists because every question costs real API money to ask 12–20 times a week — which is also why each one's number can be trusted.
Fresh angles, weekly and grounded
Every scan proposes a few questions your category's answers keep orbiting that you don't track — each with its stated basis in the actual answers (a repeated comparison, a use case the engines volunteer). It refreshes weekly because the answer corpus does. What it is not: a 'trending prompts' feed dressed in volume numbers nobody can verify.
Your edits drive the next scan
Add a question your sales calls keep hearing, drop one that stopped mattering — the change saves instantly and the next weekly scan measures the new set. History stays intact for questions you keep.
What we won't show: estimated “prompt volumes.” Nobody outside the engine companies has real AI query volume data, and a made-up volume score would just be keyword-tool numbers wearing a new hat. We rank questions by what we can actually measure — how often you win or lose them.
Common questions
How does CiteGraph choose which prompts to track?+
The first scan reads your site and mines the buyer questions your category gets asked — comparisons, alternatives, pricing and use-case questions — phrased the way people actually ask AI assistants. You can edit, add or remove questions at any time, and the next scan follows your set.
How often does each prompt run?+
Weekly, 3–5 times per engine on each of 4 engines — 12 to 20 answers per question per week. That repetition is what lets every rate print with a margin of error instead of pretending one answer is the truth.
Can I see how each engine answers a specific question?+
Yes — every question expands into its per-engine record (named and cited counts per engine), and every raw answer is stored verbatim in the Answers inspector, with citations.
Does CiteGraph have trending prompts?+
We ship the honest version: New angles this week — questions proposed from your latest scan's actual answers, each carrying the basis that suggested it, refreshed weekly because the answers are. What no tool can honestly ship is prompt 'traction': real AI query volumes aren't published by anyone, so a trending feed claiming measured demand is repackaging keyword data or guesses.
Why don't you show prompt search volume?+
Because real AI query volumes aren't published by anyone — any tool showing one is estimating from web search data that lags AI behavior. We'd rather rank your questions by measured wins and losses than decorate them with a number nobody can verify.
See it on your own domain — first dossier in about a minute.