3.1 — What the AI generates
On this page
3.1 — What the AI generates
VibeMap's generation pipeline produces eight artifacts. This page is a quick tour of each, with examples of what "good" output looks like and what to do if yours doesn't match.
1. Summary
A tightened-up restatement of your prompt that surfaces the product's positioning, target user, and main value proposition.
What good looks like: A reader who's never heard of your product can describe it in one sentence after reading the summary.
If yours is off: Edit your prompt to be more specific about the user and the value, and click Regenerate Summary from the Summary tab.
2. Personas
2-5 user types your product serves. Each gets a short bio, goals, pain points, and a primary motivation.
What good looks like: Personas feel like specific people, not categories. "Cost-Conscious First-Time Diver, 28, planning her first overseas dive trip on a £1,000 budget" not "Budget diver".
If yours is off: Edit individual personas in the Personas tab. You can also delete personas and regenerate just the ones you want.
3. Features
The product's major capabilities, grouped logically.
What good looks like: 5-12 features at the top level. Each feature is a coherent product capability ("Booking Engine") rather than a single screen ("Booking confirmation page").
If yours is off:
- Too many features (20+): Your prompt is probably trying to do too much. Trim it to the v1 surface.
- Too few features (1-2): Your prompt is probably too narrow. Describe more of the product surface.
- Wrong granularity: Edit a feature title and description, or delete it and add a new one from the Features tab.
4. User stories
Per-feature "As a … I want … so that …" statements, each tied to a specific persona.
What good looks like: Every user story has all three clauses filled in, references a real persona (not "the user"), and describes a concrete behavior.
If yours is off: Regenerate user stories for a single feature from the feature's detail view. You can also write user stories by hand if you have specific UX in mind.
5. Acceptance criteria
The atomic, testable, claimable work units. This is the most important artifact — your AI agent works directly from these.
What good looks like:
- Each AC is one specific behavior or constraint
- Each is independently testable
- Each is small enough to be one PR (~1-3 hours of work)
- Each has crisp, unambiguous wording
Bad AC: "Filter UI works well" Good AC: "Filter UI shows a price slider with min/max bounds matching the actual price range of the listings on the page"
If yours is off: Regenerate acceptance criteria per user story from the Criteria Breakdown tab. Or write them by hand — clear ACs are worth the time.
6. Pages
The page-by-page UX outline.
What good looks like: One page per major screen. Each has a clear URL slug, a one-line purpose, a list of sections, and auth rules (who can view it).
If yours is off: Edit pages individually, or regenerate the page list from the Pages tab. The Sitemap sub-tab gives you a visual of how pages connect.
7. Schema
A starter database schema fit for the feature set, with tables, columns, types, and relationships.
What good looks like: Tables map cleanly to features. Relationships look right. Column types are sensible.
If yours is off: The schema is the artifact that benefits most from human review. Open the Schema tab, switch to Tables view, and edit columns directly. The diagram view updates live.
8. Files
A starter folder/file structure for your codebase. Intentionally light — your agent will fill in the actual files.
What good looks like: A few top-level folders (app/, lib/, components/, etc.) with reasonable subdivisions. Not 50 placeholder files.
If yours is off: Edit it by hand. The files outline is a shared map between you and your agent; it doesn't need to be exhaustive.