Deep Research is an agentic AI tool in Marvin that runs exhaustive, project-level analysis on your raw data. It does the heavy lifting by reading every file from top to bottom, so you can spend your time spot-checking insights rather than producing them.
Unlike the Analyze feature, Deep Research doesn’t require notes, taxonomies, or coding. Bring in raw transcripts, surveys, or documents, pick a ready-made framework, and Marvin returns a structured report with citations that link straight back to the source.
When Deep Research shines:
Quick turnarounds: You have a large volume of data and need themes, summaries, or a holistic view in minutes.
The middle ground: You don’t know where to start. Use it as a kickstarter to surface top themes and pain points before you dig in.
Non-researchers: Built for PMs, designers, and cross-functional partners who want professional insights without diving into every detail.
Follow the steps below to get the most out of Deep Research:
Step 1: Gather your raw data in Marvin
Deep Research works directly on unstructured data, so there’s nothing to code or tag first. All you need is data in the project. The fastest path is to let Marvin collect it for you, so your transcripts, recordings, and responses land in the project already structured and searchable.
Collect data with Marvin
AI Moderated Interviewer. Run moderated interviews on demand from your discussion guide. The AI adapts to answers, asks dynamic follow-ups, and interviews in 40+ languages. Hundreds of interviews can run in parallel, and each completed interview is saved to your project automatically.
Record with the Marvin bot. Invite the Marvin bot to Zoom, Google Meet, or Microsoft Teams. It silently records, transcribes, and summarizes each call straight into your project.
Record calls and webinars in Marvin. Capture live sessions directly, no external tool required.
Pull in surveys and tickets. Connect integrations to bring responses and support conversations into Marvin automatically. Marvin integrates with Qualtrics, SurveyMonkey, Google Forms, Dscout, Pendo, Zendesk, and Intercom.
Optional: bring in data you already have
Already collected your research elsewhere? Upload it directly instead.
Upload files. Add audio, video, Word or Google Docs, Excel or Sheets, PowerPoint, PDFs, and text files.
Import surveys. Drop in a CSV or Excel export to turn open-ended responses into analyzable data.
Tips for success
Add a project overview and discussion guide. These help Marvin organize your data and noticeably improve the quality of analysis—and the discussion guide doubles as the script for the AI Moderator.
Mind the 75-file limit. Deep Research analyzes up to 75 files at a time. Larger datasets? Ask AI will be a better option in this case.
Learn more:
Step 2: Open Deep Research
Once your files are in a project, open Deep Research from the project’s research tools. Deep Research runs at the project level, so it analyzes the data inside the project you’re working in.
Set your expectations
Project-specific scope. Deep Research runs within a single project—not across your whole repository in one pass.
Set it and forget it. Once analysis starts you can’t ask follow-up questions or interact with it mid-run.
Exhaustive by design. Because it reads every file top to bottom, a run takes some time to complete.
Tips for success
Start it before you step away. Kick off a run right before another task or at the start of your day, and come back to a finished report.
Learn more: How does Deep Research work?
Step 3: Narrow your dataset
This is the most important step for quality. By default, Deep Research reviews everything in the project, including unrelated files. This can pull in off-topic quotes and make the AI response feel inaccurate. For the best results, narrow the dataset before you run it.
File tags and filters. Filter by methodology, quarter, product type, segment, or any tag so Marvin only reads what’s relevant.
Timestamps. Set date ranges to exclude outdated findings—especially after product or system changes that make older research invalid.
Tips for success
Always filter. A focused dataset is the single biggest driver of accurate, trustworthy results.
Tag as you go. Consistent file tags make every future Deep Research run faster and sharper.
Learn more:
Step 4: Choose a research intent
Research intents are prepackaged frameworks—prebuilt prompts so you get a structured, professional report without writing a single prompt. Pick the one that matches your goal, or write your own.
Pick a framework
Project Summary: Overall summary, key takeaways, and recommendations in minutes.
Personas: Auto-builds marketing-style personas with names, characteristics, goals, pain points, and citations.
Jobs-to-be-Done: Maps the jobs your users are trying to get done from unstructured data.
Hypothesis Testing: Proves or disproves a central belief across data with a conclusion, implications, and recommendations.
Question & Answer Tables: Synthesizes participant responses question-by-question across qualitative interviews.
Find Quotes: Surfaces specific quotes and links each one back to its source file.
Custom Prompt: Write your own instructions for a fully tailored report.
Tips for success
Match the intent to the question. Comparing usability responses? Use Q&A Tables. Validating a belief? Use Hypothesis Testing.
Learn more:
Step 5: Run it and step away
Once your filters and intent are set, you’re ready to run your report. Deep Research does a comprehensive review across every file, so it may take a little time to generate a thoughtful analysis. While it works, you can move on to something else and come back when your report is ready.
Make the wait work for you
Batch your runs. Queue several intents at once, such as a summary, hypothesis test, and quote finder, so you can gather a complete set of outputs in one session.
Plan around the run time. Deep Research is designed to do detailed analysis across your files, so you can start a run, move on to your next task, and come back when your results are ready.
Tips for success
Use the time productively. Since there’s no live interaction during a run, you don’t need to stay on the screen. Let Deep Research work while you focus on something else.
Learn more:
Step 6: Verify, combine, and publish
Every report comes with numeric citations that link back to the source files, so validating the AI takes seconds. From there, turn your outputs into a polished deliverable and your'
Finish strong
Spot-check the citations. Click the numbered references to confirm each finding connects back to a real source. This is the fastest way to build confidence in the results.
Combine outputs. Export multiple reports, such as a summary, hypothesis test, and quotes, then bring them together into one comprehensive deliverable.
Publish to Insights. Deep Research is a contributor feature. Publishing reports to Insights makes them visible and searchable for stakeholders and viewers.
Tips for success
Working across projects? Run Deep Research in each project using the same prompt, then combine the reports into one shared Insight for a complete view across your research.
Working with more than 75 files? Use Ask AI with the “Think Longer” option. Its multi-step analysis helps you get detailed answers across larger sets of data without the same file limit.
Learn more:
Which tool should I use?
Deep Research is just one part of Marvin's AI toolkit. Here's how to pick the right one for the job.
Ask AI (Think Fast): Your go-to for quick, straightforward Q&A. Optimized for speed, recency, and relevancy.
Ask AI (Think Longer): A multi-agent model built for comprehensive, defensible answers. Great at reasoning through conflicting signals across a larger sample, and ideal when you're over the 75-file limit.
Deep Research: Structured, exhaustive, project-level reports built on prepackaged frameworks. Perfect for repeatable, citation-backed deliverables.
Together, they give you the right depth for any question: fast answers, rigorous reasoning, or full research reports.
Learn more:
The new Ask AI experience in Marvin