How do I use Ask AI?

Marvin’s Ask AI helps you query your entire customer knowledge to get cited answers to any question. Empower everyone on your team to make decisions grounded in customer knowledge. Learn what features customers have requested over time, how prospects think about your competitors, and what they’ve said about pricing across studies and channels.

Ask AI has been rebuilt. If you've used it before, here's what's different:

Type your question into the Ask AI bar anywhere in Marvin. Where you ask determines what gets searched.

Some questions to start with:

You can also open Ask AI from the Discover Research tab.

If your question is ambiguous, very broad, or could be read more than one way, Ask AI asks a single clarifying question before searching, with a few interpretations to choose from and a Something else option for your own wording.

This isn't a step on every query; it only appears when the answer would genuinely be better for it.

A main agent sends out explorer agents that each search a different part of your repository at the same time — interviews, surveys, support tickets, imported files and insights. They work independently, then report back so the main agent can synthesize one answer across everything they found.

Ask AI shows a live, log of what it's doing, moving through stages like Found relevant material across your repository, Scanned your research, Scoped your answer, Explore agent, Understood your data, Building with what I have so far and Analyzing.

As soon as Ask AI has enough relevant material to say something useful, it posts an early answer labeled Early answer from a few initial sources. It's a real answer, not a placeholder, and it tells you what it's still working on.

⚠️ The Quick Answer is provisional. It won't include citations, and it hasn't seen everything in your repository yet. We recommend waiting for the full response before you act on it or share it.

The complete answer appears in place of the Quick Answer as it's generated; there's no separate "your answer is ready" step to watch for. The full answer adds:

You can also ask for a format up front; show this as a bar chart, give me a table, and Ask AI will produce it.

Use Standard unless you have a specific reason to go deeper. It searches your whole repository and handles the everyday questions: theme identification, quote retrieval, topic synthesis, VOC summaries, feature feedback.

Switch to Advanced when the question is complex, when your repository is large and spread across many projects and file types, or when the answer needs to stand up to scrutiny — an executive report, a strategic synthesis, a final VOC summary before a launch decision. The Quick Answer still appears early, so the longer run never means a blank screen.

Notifications are on by default, and a banner confirms You will be notified when the answer is ready. Navigate away, do something else, and come back when the full answer lands.

Before you begin: Form a hypothesis, scope the project, or find participants you've interviewed before who meet your criteria. Use what comes back to write project overviews and discussion guides.

Inside files: Query interviews, surveys, and usability tests directly. Marvin reads recordings, images, and documents for context, and uses your discussion guide and labels to build accurate summaries.

When you're ready to synthesize: uncover trends, find themes, prove or disprove your hypothesis, locate quotes. Use the Analyze tab for one project or Cross-project analysis for several, then add answers to Insights to build a report for your team.

Filter before you ask. You can filter by project, file tags, research areas, time period and file type.

Choose whose voice counts. Under the settings toggle, set Participants to Respondents so answers highlight what participants said rather than interviewer opinions. Marvin still reads interviewer questions for context. Respondents is the default.

Filter after the fact. Apply a filter once an answer exists, and Marvin adds to the answer in the same chat.

Refine any answer without starting over. A few things to try:

Not an exhaustive list. Keep going until you have what you need.

Marvin keeps a history of your Ask AI results on the advanced search page, grouped by last 7 days, 7–30 days, and older than 30 days. Get there from the Ask AI modal.

Your team can ask questions where they already work, without hunting through the repository.

Who can use it: all roles on Standard and Enterprise plans, in any channel or DM. Answers are based on the projects you have access to in Marvin.

How to ask: use /heymarvin or /research followed by your question.

You'll get an overview, key takeaways, and supporting quotes where they're relevant. Select Ask a Follow Up to go deeper — Marvin keeps the context of your previous question.

Enterprise accounts can extend Ask AI on Slack to anyone in the organization, even without a Marvin account. An Enterprise Admin needs to share research with the entire organization from Admin settings. This works for accounts with a single team only, not for accounts with multiple teams. Non-Marvin teammates are treated as guest users and see answers based on research that's public and shared with everyone in the team.

Because every claim carries a citation back to the source file, whoever you send it to can check the evidence in one click.

What happened to Think Fast mode? It's been retired. Standard mode is more thorough than Think Fast was, so the trade-off it offered no longer exists. Existing Think Fast users were moved to Standard automatically.

What's the difference between Standard and Advanced? Both use explorer agents, produce visualizations, ask clarifying questions, and show a Quick Answer. Advanced deploys more agents and runs deeper passes, which makes it the better choice for complex synthesis or very large repositories.

Does Ask AI really search every relevant document in my repository? Higher recall is the core improvement in this release. Ask AI runs multiple iterations of parallel search and spawns explorer agents specifically to surface documents that earlier passes may have missed.

Do I have to wait for the full answer? Not strictly — the Quick Answer is there to read immediately. But it won't have citations and may not have looked at every relevant source, so we recommend waiting for the complete response.

When does Ask AI ask a clarifying question? Only when a query is ambiguous, too broad, or likely to produce a less accurate answer without more context.

What do the thinking steps show? Each stage expands into real detail: the search terms Ask AI generated for itself with result counts, and the specific documents it's reading, by name.

Want to learn about the kind of prompts you can use on Ask AI? Check out our help center article.