AI NativeThe prompt library is the set of buyer questions that get asked to AI assistants on every scan. Everything the platform measures starts from this library.
The tree structure
The prompt tree is not a flat list. It is organised by persona, then funnel stage, then by prompt layers.
A Layer 1 (L1) prompt is an opening question that a buyer might actually type into an AI assistant. L2 and deeper prompts are follow-up questions that branch from L1 questions. Provenance probes are a special type: they ask where information comes from, which feeds the citation analysis on the provenance tab.
The tree header shows the total number of prompts and the prompt set version number. The version increments every time a prompt is added, edited, or deleted. This matters because version boundaries define separate measurement series.
Coverage status
At the top of the library page, a coverage block shows what percentage of persona-stage cells have at least one prompt. A cell with no prompts is called a blind spot. It appears in the coverage panel and is listed as a gap because AI can win those cells uncontested.
If the latest scan did not measure all prompts (for example, you added prompts after the last scan), a banner tells you how many are unmeasured and offers a link to run a scan that covers all of them.
Adding prompts manually
Click "Add" in the page header to expand the add form. You set the question text, whether it is branded (asks about your product by name) or unbranded (a generic category question), which persona it belongs to, and which funnel stage it sits in. The five funnel stages are awareness, consideration, comparison, decision, and retention.
Generating prompts with AI
Click "Generate with AI" to expand the generation form. You can optionally provide a steering instruction (for example, "more objection probes for rate shoppers") and choose how many prompts to generate (3, 5, 8, or 10). The engine LLM drafts demand-grounded questions based on your product's positioning and persona context. The new prompts appear in the tree for review before any scan measures them.
Importing a prompt list
Click "Import prompt list" to expand the import form. Paste CSV or TSV rows with columns: Product, Prompt, Stage Name, Type Name. A leading serial-number column is accepted. The importer maps your stage and type names onto the platform's funnel stages and branded/unbranded classification, and shows the mapped results in an imported prompts section.
Imported prompts initially have no persona assignment. A banner appears on the page offering to run a persona assignment pass across the unassigned imports.
Editing and deleting prompts
Hover over a prompt row to see edit and delete icons. Clicking edit expands an inline form with the current question text. Saving the edit increments the prompt set version. Deleting a prompt also increments the version.
You can expand an inline answer preview for any prompt that has been measured. Click "See what AI actually said" beneath a prompt to see the answers from the latest scan, including the engine, outcome state, and accuracy flag.
Prompt set versioning
The version number is shown on the tree page (for example, "prompt set v3"). Every time the set is changed, the version increments. This creates a version boundary in your scan history.
Scans run before a version change and scans run after are in different versions. The trend chart compares within a version, not across versions, so a prompt edit does not corrupt your historical trend. The platform keeps all historical scan data regardless of version.
Opportunity data per prompt
If the product has open opportunities, they appear attached to the relevant prompt in the expanded answer view. The opportunity block shows the lever type, the opportunity score (value times gap times winnability), and a plain-English content action.
Related docs
Questions
What is a "blind spot" in the coverage panel?
A blind spot is a persona-stage cell that has no prompts. Because no questions are asked for that cell, the scan produces no data for it. Competitors who do appear in answers to questions in that cell accrue advantage you cannot measure or counter.
Does adding a prompt change my historical data?
Adding a prompt increments the prompt set version but does not change historical scan results. Old answers are unchanged. The new prompt appears in the tree and is measured in the next scan. The trend chart continues from the new version boundary.
What happens to my data if I delete a prompt?
Deleting a prompt increments the version number. Answers collected for that prompt in previous scans are retained in the database and remain visible in the Answers tab filtered by the old scan. The prompt no longer appears in the tree or in future scans.
Can I import prompts from an existing research spreadsheet?
Yes. Use the Import prompt list feature. Paste CSV or TSV rows with product, prompt text, stage name, and type name columns. The importer maps your naming conventions to the platform's taxonomy and shows you the result before any scan runs. Imported prompts still need persona assignment.
What is a provenance probe prompt?
A provenance probe asks AI where it got its information: for example, "What sources do you use for that?" These prompts feed the citation analysis on the provenance tab. They are tagged in the tree with a "provenance" label.
How do branded and unbranded prompts differ?
Unbranded prompts are generic category questions a buyer might ask without knowing your brand, such as "which home loan has the lowest rate?" These measure your share of voice against competitors. Branded prompts mention your product by name and measure how AI represents you when asked directly.
What steering should I give the AI generation form?
Describe the gap you want to fill. For example: "add comparison-stage questions for first-time buyers" or "focus on retention questions about renewal and claims." The generation uses your product positioning and persona context, so specific steering produces more targeted questions.