Pick a set on the left or start a new one. Six stages: brief, topics, distribution, generate, review, export. Rules stay code, buttons change counts.
The set is labeled "borrowed profile" until a client survey is attached.
Topics proposed from the brief and the corpus, ranked by platform demand when the volume lookup is on. Approve the table before any prompt is written. Twenty per topic is the floor.
| Topic | Includes | Excludes | A competitor could win it | Demand / mo | Keep |
|---|
Total promptsThe platform cap is 340. Sentiment and FactCheck count against a separate allowance.
VisibilityUnbranded, the asker's side. Whatever Sentiment and FactCheck do not take.
Rule · competitor-anchored prompts never go on Visibility
Top of funnelCategory questions, no brand in sight.
Middle of funnelComparing options, constraints, use cases.
Bottom of funnelBranded by definition, so it lives on Sentiment.
Rule · the named asset is always Sentiment
0
SentimentPrompts that name the owned asset.
Bottom of funnelNamed asset, buyer's question.
Middle of funnel, named
FactCheckOne prompt per checkable claim from the brief, bottom of funnel only.
Population ageShares pick which real verbatims the model sees, by age band. Typed in until a survey arrives.
Typed inPersonasTest harnesses, not biographies. Three fit the platform. Blank baseline stays between 35 and 45 percent.
Viewpoint carrierModes B and C can embed the viewpoint in the prompt text instead. One carrier, never both.
Each topic by funnel cell pulls its nearest real prompts as examples, the model writes candidates against the measured profile, the rules prune.
Accept, edit or reject with a reason, one row per prompt. Every decision is kept; rejections are what the generator learns from.
| Prompt | Topic | Funnel | Type | Persona | Decision |
|---|
The nine-column platform CSV, accepted and edited rows only, prefixed if the workspace is shared.
Set check
Voice comes from survey verbatims only. Built sets are history, never examples.Olympus · Agentik Momentum