The quality of your deep research analysis will depend on how you use the preset prompts or write a custom prompt. We’ve put together some best practices in this article, so you can get targeted answers for each of your queries.
When should you use deep research?
Marvin’s deep research analyzes the data in your project files. It will consider all types of files for this, including recordings, surveys and documents. However, it does not use the notes or project overview while answering your query. If you want to conduct analysis based on the notes you’ve created, we recommend you use the Analyze tab.
How do I use the prompts effectively?
You can use all the prompts mentioned in the table below, based on the type of research project. Before you begin, you’ll need to add more detail and context for Marvin to give you the right results. Each prompt may require a slightly different type of context.
Prompt | Required context |
Uncover information about a topic | Keywords/questions + file tags |
Prove/disprove a hypothesis | Hypothesis + file tags |
Find quotes | Keywords/questions + file tags |
Create insight report | Keywords/questions + file tags |
Custom prompt | Custom prompt + file tags |
Create a persona | File tags |
Question/Answer table | Discussion guide + file tags |
Quantified thematic analysis | File tags |
Here’s an example of how you would have to add the information when you’re trying to uncover information on a topic:
How do I add keywords, questions or a hypothesis to a prompt?
Keywords and questions tell Marvin which topics are top-of-mind for you. It will search across the file tags you specify to search for this information. Add keywords and questions that are specific but not overly complex.
If you’re using the prove/disprove the hypothesis prompt, please mention the exact statement. For example, “My hypothesis is that users pick our product because it streamlines their workflow”.
Best practices
Be specific and avoid terms such as “they”, “recently”, “that information”, etc. For example, instead of asking “What do they say about the feature?”, specify, “What do users of X say about ABC feature?”
Use complete terms instead of abbreviations. For example, “personally identifiable information” instead of “PII”
Don’t use synonyms for terms that have been used in calls. Use the exact term instead. For example, you cannot use “Ask AI” and “AI analysis” interchangeably without specifying that they may refer to the same thing
Assume that Marvin will not know about frameworks. Explain the framework in a few lines or ask for help with each aspect of the framework
If you're using the custom prompt, provide context about what the research will be used for
Things to avoid
Avoid complex prompts that require Marvin to account for multiple permutations. Instead, run separate analyses and combine the information later in a separate report
Avoid non-specific time frames and parameters, such as “What have been the top 10 pain points for users in the last two years. In this case, Marvin does not know which exact time period to consider as “the last two years”. It’s also not clear which criteria Marvin should consider for it to be in the “top 10”. Instead, you could ask “Which pain point was repeated most often by users between March 1, 2023 and March 1, 2025? Count multiple mentions by the same user only once.”
Avoid relying on deep research analysis for predictions. There are too many variables for Marvin to consider to make accurate predictions. Instead, you could ask what trends users have been mentioned across calls.
How should I use file tags for a prompt?
The file tags you choose will be key to surfacing the right data. Imagine Marvin diligently searching through all the files and folders in your project for the topics you specified in keywords. By picking the relevant file tags, you’re telling it to look in specific file cabinets.
Avoid using keywords to guide Marvin to a specific data set. For example, avoid framing questions such as “Why do enterprise users like our product”. Instead, you could ask “Why do users like our product?” and then add a file tag that corresponds to feedback from enterprise users. This way, you’re certain that all the enterprise users have been covered in the analysis while Marvin focuses on surfacing relevant information
Best practices
In case you’ve not added detailed file tags to your project, we recommend you do so before you begin deep research. This will also help you significantly when you’re using Ask AI and the Analyze tab
Consider adding file tags from within the file, as you can easily spot the categories a particular file corresponds to
Use tags that will add layers of analysis to your research. Some useful types of tags are time period (e.g., March 2025), company name (e.g., Marvin), user segment (new user), type of research (e.g., user feedback or industry report).
Add multiple file tags to get specific. Marvin will use both to filter the data. For example, if you choose file tags for “Survey” and “March 2025”, Marvin will only consider files that have been tagged as surveys as well as from March 2025.
Things to avoid
Avoid comparing more than three file tags at a time. When you choose large data sets with multiple permutations, Marvin may give you more generic answers.
Avoid using the same file tags when comparing two distinct types of data.
Points to remember while using Deep Research
Avoid large and complex prompts.
Marvin will not be able to use another deep research analysis as context
Deep research is a great place to begin your research, especially if you don’t have detailed notes for each file. It’s a great way to cut down time on collating and organizing insights into themes. You can then use these to go deeper into your data and build each part of your report from separate analyses.