Skip to content
← DocumentationSetting up a client

Prompt groups and paraphrases

Why breadth beats repetition, and how a group rolls up into one tracked question.

A prompt group is one intent expressed several ways. “best medical aid in South Africa”, “which medical aid should I get in SA”, and “top rated medical schemes South Africa” are one question a buyer has, and three inputs an engine treats as meaningfully different.

Why the platform models them

Brand-mention outcomes are surprisingly sensitive to phrasing: mention behaviour begins to break down once phrasing similarity drops below roughly 0.60, meaning a paraphrase a human would call identical can flip a brand from present to absent. Measuring one phrasing and reporting it as “how the engine answers this question” overstates what you know.

How it reports

  • The group is the tracked question, and its band aggregates every member phrasing.
  • Each member stays independently inspectable, so “we win on one phrasing and lose on the other” is visible rather than averaged away.
  • Language is a dimension of the group, not a separate prompt. An English and an Afrikaans phrasing of one question belong to one group and report separately within it.

Language explains roughly a third of the variance in AI answers — more than the choice of model, and more than which brand is being asked about. A market with two languages needs both phrasings tracked, not one translated at report time.

Writing a prompt setMarkets and languages
Next step

See it against your own client list

A working demo runs your prompts, in your market, on live engines — not a sandbox with seeded data. Bring one client brand and three competitors.