How to use ClarifAI

ClarifAI interviews you about a product idea and turns it into a structured brief. Unlike tools that paper over what they cannot answer, it keeps the open questions visible — six coverage topics tracked as you talk, and an honest Gaps Remaining list in every brief. This page covers how the interview works, what each style does, and how to get the most from a session.

Getting Started

ClarifAI is a conversational tool that turns a rough product idea into a structured requirements brief. Instead of filling in a form, you have a real back-and-forth with an AI that knows how to ask the right questions at the right depth — and is honest about what it still cannot answer.

Every session works the same way: describe your idea and start talking — there is nothing to set up first. The AI tracks six coverage topics as you go — problem, users, solution, metrics, constraints, and scope — and keeps asking until it has enough to generate something useful. The chat never stalls on a form: you can watch the topics fill in the Topics panel, and every brief ends with an honest Gaps Remaining list of what is still open.

1Add your Anthropic API key in Settings (the gear icon, top right). Your key is stored securely in your browser and never saved to any server. In demo mode you can skip this and start without a key.
2On the welcome screen, describe your idea and send it — type, or use the microphone (or press ⌘⇧M) to speak. There is nothing to configure first; the interview starts the moment you send.
3The AI interviews you one question at a time, adapting its depth to your answers — it leads at first and quietly steps back as you volunteer detail. The style defaults to Curious Researcher; change it anytime in Settings (it applies to your next interview).
4Answer honestly and in as much detail as you have. The AI tracks six coverage topics as you go — you can watch them fill in the Topics panel. The more you give, the faster coverage fills and the better the brief.
5When all six topics are covered, a Generate Brief button appears. Pick an output format — your brief is generated and saved. You can generate additional formats from the same session without re-doing the interview, and switching between already-generated formats is instant and uses no tokens.
6Review the brief, edit any section inline, check off Gaps as you resolve them, and export to Markdown or PDF.

If you are a returning user, your previous session appears on the welcome screen. You can continue from where you left off or start a fresh one at any time.

Interview Styles

Your interview style — set in Settings, defaulting to Curious Researcher — determines the AI's personality for the session. Each one is tuned to draw out a different kind of thinking. There is no best style — the right one depends on where your idea is and what kind of pressure you want on it.

Best for mature ideas

Skeptical PM

Pushes back on assumptions, demands concrete evidence, and names contradictions out loud. If you say "users will love this," it will ask why and what evidence you have. Best when your idea is somewhat formed and you want it stress-tested before writing a single line of code.

Best for early-stage ideas

Curious Researcher

Open-ended questions, reflective follow-ups, explores threads without judgment. It does not challenge so much as expand your thinking. Best for early-stage ideas where the problem space is still fuzzy and you need help articulating what you are actually building.

Best for thorough coverage

Structured Interviewer

Works through all six coverage topics in a fixed sequence, adapting depth but never skipping ahead. Predictable and thorough. Best when you want to make sure nothing gets missed, or when working on a regulated product where completeness matters.

Best for final logic check

Contradiction Detector

Listens quietly and only intervenes when your answers are inconsistent with each other. If your target user shifts between answers, it will notice and ask you to reconcile. Best for ideas you have already thought through and want a final sanity check on.

Your preferred style lives in Settings and takes effect on your next interview; a session keeps the style it began with. Pick the one that fits where your idea is before you start.

Modes

Modes control who leads the conversation. The style controls how the AI asks questions; the mode controls whether the AI or you sets the pace. You never pick a mode — every interview starts in Interview mode and shifts on its own.

·Interview mode — The AI asks the questions and you respond. Every session starts here — you are guided rather than having to remember what to cover. The AI automatically shifts toward Collaborative if it detects you are volunteering detailed, structured answers on your own.
·Collaborative mode — You share information at your own pace and the AI listens, probes gaps, and fills in what is missing. The AI shifts here on its own once you have given it a lot of context; you can jump between topics in any order.

Mode switches happen silently — there is no notification; the conversation just becomes less question-heavy as you take the lead.

Output Formats

When the AI has enough coverage, you choose what kind of document to generate. The AI continuously judges the maturity of your idea based on what you have shared, and unlocks formats as your idea develops.

Two formats are always available regardless of maturity (One-pager brief and Assumption map). Three unlock progressively as you add more depth to the interview — each format card below shows exactly what is needed to unlock it.

One-pager brief

Always available

A concise, shareable summary covering problem, users, proposed solution, and key constraints. The right starting point for any idea. Use it to align stakeholders early or give a designer or engineer a quick overview before committing to a longer document.

Assumption map

Always available

A structured list of every assumption embedded in your idea, ranked by risk. This format surfaces the bets you are making so you can decide which ones to validate first. Particularly useful in early discovery, before committing to a direction.

Classic PRD

Mature ideas

A full product requirements document with sections for background, goals, non-goals, user personas, user stories, functional requirements, and success metrics. Requires depth across all six coverage topics. Best when handing off to an engineering team or entering a formal planning cycle.

🔓

Unlocks when your solution, metrics, and constraints are well-defined. Share concrete goals, measurable success criteria, and what is explicitly out of scope.

Sprint backlog

Dev-ready ideas

Breaks the solution into actionable work items ready to drop into a sprint or project board. Each item includes a user story, acceptance criteria, and a rough size indicator. Available only when the idea is scoped clearly enough for implementation.

🔓

Unlocks when the idea is scoped for immediate implementation. The AI needs to understand the technical approach, team constraints, and a clear definition of done.

Figma brief

Design-ready ideas

A handoff-ready document structured for designers, covering user flows, interaction notes, content requirements, visual direction, and accessibility considerations. Available when the UX direction is clear enough for the AI to draw meaningful conclusions about how the product should look and behave.

🔓

Unlocks when the UX direction is clear. Describe specific user flows, key screens, interaction patterns, and any visual direction or brand constraints.

After generating, you can edit any section directly in the app by clicking the pencil icon on each card. Edits are saved automatically. Use the “Generate in another format” button at the bottom of the brief page to add more formats from the same session. Each format is generated once and stored — switching between formats you have already generated is instant and uses no additional API tokens.

Tips and Tricks

A few patterns that consistently produce better briefs:

Describe the problem before the solution. The AI tracks problem coverage separately. If you jump to features, it circles back to the underlying problem anyway. Addressing it upfront keeps the conversation moving.

Be specific about your target user. "Small business owners" produces a weaker brief than "independent restaurant owners managing their own bookkeeping." The more concrete the user, the more precisely the AI can probe for constraints.

Give numbers when you have them. Target retention rates, conversion benchmarks, or growth goals fill the metrics topic far faster than qualitative descriptions.

Say what the solution is not. Explicit non-goals fill the constraints topic and save the AI from asking about them separately.

You can paste existing notes. Paste a Notion doc, Slack thread, or email directly into the chat. The AI extracts what it needs and asks only about the remaining gaps.

Edit the brief after generation. Click the pencil icon on any section card to edit inline. A five-minute edit pass consistently produces a sharper document than relying entirely on the AI output.

Generate multiple formats from one session. Each format is generated once and cached. Use the "Generate in another format" button at the bottom of your brief to add a second or third format. Switching between formats you have already generated is instant and uses no additional tokens — only the first generation of each format calls the AI.

Use voice input to answer faster. Click the microphone icon in the chat input, or press ⌘⇧M on Mac (Ctrl+Shift+M on Windows), to dictate your answer. The microphone stays active through natural pauses — click again or press the shortcut to stop. Particularly useful for long, free-form answers where typing slows you down.

Gaps Remaining

Every brief includes a Gaps Remaining section at the bottom with two to four open questions. These are things the AI could not resolve from your conversation — not because you did anything wrong, but because the answers depend on decisions not yet made, research not yet run, or constraints not yet known.

Gaps are not errors. They are the next set of questions you need to answer before the brief is complete. Treating them as a working to-do list is more productive than trying to eliminate them before generating.

·Review each gap after you generate. Some will be answerable immediately from context you did not include in the interview.
·Click the checkbox next to a gap to mark it resolved. The gap text stays visible but is marked done, so reviewers can see what has been decided.
·Resolved gaps appear greyed out in PDF and Markdown exports, giving your audience a clear picture of what is settled and what remains open.
·If a gap requires a decision, use it as a concrete agenda item in your next review. The Gaps section is designed to be shared as-is without additional formatting.
·You do not need to resolve all gaps before exporting. A brief with open gaps is still a useful brief — it makes the unknowns explicit instead of hiding them inside optimistic assumptions.

If you want fewer or more targeted gaps, re-run the session with more specific answers on the topics you know are incomplete. The gap count and content will shift accordingly.