Detect
Ask the engines what your buyers ask.
Every week, Ondit puts the questions a buyer would type to every answer engine we track, several times each, and records who they name and what they cite. Not a score from a model's opinion: the same question, asked again, until the answer is a measurement.
What it does
The instrument, and what makes it one.
Buyer-shaped questions
Written from the buyer's side — "which vector databases are easiest to self-host?" — from category templates and from your own brand description. Never a question that presupposes its answer.
Every engine, repeated
ChatGPT, Perplexity, Gemini and Claude, each with web search on, each asked five times. One answer is not a measurement; five is a distribution.
Mentions and citations
Who is named in the prose, matched on name and aliases with word boundaries; and who is cited, by domain and by page, from the links the engine actually used.
How it is built
What comes out
A table of samples: question × engine × replication, with the answer text, the names it used and the URLs it cited.
What happens next
Analysis turns the samples into shares, competitors and gaps — the same day.
- Weekly, on a schedule the plan setsIndustry questions run on Ondit's own account; your questions run on your plan's engines and replications, in your region and language.
- Generic names are not measured by nameA product called "App" would score every answer containing the word. Those are tracked by the domain they cite, never by the name.
- A closed gate is recorded, not retriedWhen an engine is priced above the ceiling or a notice comes back, the batch is marked skipped and says why. A permanent condition is not retried three times.
- Every call is pricedEach answer carries its token cost at the model's published rate, so the number you see for spend is one you can check.