Using AI search and analysis in Condens

Ask your Sessions, or simply put, AI search on raw data, lets you search across your raw session transcripts and notes. You can use it at three levels:

1. Within a single Session

When you're inside a session, you're already working with a focused, defined set of data. Search here is straightforward; you can look for specific keywords or ask a natural language question, and the results are scoped to that session.

2. Across a Project

Searching at project level covers all sessions within that project. Since the data set is wider, it's good practice to combine your search with filters to keep your results relevant. For example, by participant type, date, or any custom metadata fields you've set up.

3. Across your workspace

Workspace-level search is global AI search across all your raw session data. It's the most powerful scope and the one people find most useful for cross-study questions. Because the sample can be very wide, combining it with classic search filters makes a real difference to the quality of your results.

Tips on writing good prompts for AI searches on raw data

What works well are specific questions with concrete answers.

Within a single Session

Good questions at this level tend to be detail-oriented — drilling into a specific moment, sentiment, or topic within that conversation:

✅ "What did this participant say about the checkout flow?"

✅ "Were there any moments where the user expressed frustration?"

✅ "What was the overall sentiment toward the new feature?"

✅ "Did the participant mention any competitors?"

✅ "Where did the user struggle during the task?"

Across a Project or Workspace

At this level, good questions look for patterns, themes, and signals across multiple participants:

✅ "Who mentioned issues with the checkout flow?"

✅ "Which participants asked for a dark mode?"

✅ "Where did users describe feeling confused during onboarding?"

✅ "Which sessions included feedback about waiting for results to load?"

✅ "Did anyone talk about switching from a competitor, and what did they say?"

✅ "What recurring themes came up around the new dashboard?"

Questions that require interpretation or a generated conclusion won't yield the same kind of results:

❌ "What are our biggest usability problems?"

❌ "What do users think about our collaboration features?"

❌ "What should we improve first?"

You may still get some results for these. The AI can point you to relevant moments in transcripts, but you won't get a generated answer, ranking, or recommendation. We believe that kind of synthesis should remain in the researchers' hands, and the evidence AI search surfaces is what helps you do it.

Tips for better results

• Searching your entire workspace can work, but the more focused your data set, the more relevant your results. Before you search, filter by project, session, or custom metadata fields to scope it to what's actually relevant.

• Use specific feature, flow, or product names rather than general terms. "Profile setup" will surface more relevant results than "onboarding" if that's the language your participants used.

• If results feel thin or off, try rephrasing. A slightly different question can surface different moments, especially if your participants used varied language to describe the same thing.

• AI search surfaces as much relevant evidence as it can, and everything links back to its source. But it doesn't guarantee it will catch every instance. Use it to get oriented and find the key evidence, then follow the source links to verify.

Once you've tagged highlights and pulled them onto a Whiteboard, AI-powered affinity mapping helps you group and explore them by theme, sentiment, or any framing you define. This is where synthesis starts, but you're working with evidence you've already selected, not raw transcripts.

How to write a good clustering prompt

Clustering prompts can carry more context than a session search question. You're not trying to match keywords; you're giving the AI a frame for how to think about the material you've brought together.

1. Simple, open-ended prompts work well when you want to explore broadly:

   • "Group these by the type of problem described."

   • "I want to understand the main pain points customers have."

   • "What are the main suggestions for improvement and areas of satisfaction in this feedback?"

   • "Organize these by how users feel about the experience overall."

2. More detailed prompts work well when you have a specific research context:

   • "These highlights are from usability tests on our onboarding flow. Group the findings into areas of confusion, smooth interactions, and overall task success."

   • "These highlights relate to users' motivations, frustrations, and product interaction habits. I want to understand common behaviors, recurring pain points, and key motivations."

   • "This is feedback from enterprise customers on a B2B workflow tool. Identify user expectations, frustrations, and feature requests to find areas for improvement and drive feature adoption."

   • "These are highlights from discovery interviews with first-time users. Group them by what surprised people, what felt familiar, and what felt missing."

Tips for better results with AI-powered affinity mapping

• Try reapplying clustering with different prompts to see how the groupings shift. It's a good way to look at the same evidence from multiple angles.

• Move from broad to specific. Start with an open-ended prompt to get an overview, then reapply with a more focused one to dig into a particular theme or dimension.

• If groupings feel off, adjust the prompt rather than the highlights. The clustering reflects how you've framed the question as much as what's in the data. A clearer or more specific prompt usually gets you closer to what you're looking for.

• Use Find similar highlights to expand a cluster. Once a theme starts to take shape, use Find similar highlights inside that cluster to pull in related evidence from elsewhere in your repository that you may have missed or tagged differently.

The Insights Magazine is where your team publishes final, validated research findings. AI search here works differently from session search: you're searching across concluded, documented insights, not raw transcripts. It's designed to work for stakeholders who want to easily explore and surface insights, and may not necessarily know the exact keywords or metadata behind this information.

This is the right place for:

1. Exploring what your team already knows
   Ask broader questions about your accumulated research knowledge:

   1. "What have we learned about enterprise onboarding?"

   2. "Is there any research on our mobile checkout flow?"

   3. What do we know about user trust?"

2. Checking for research gaps

   The Magazine is also useful for understanding what you don't have documented yet:

   • "Do we have any findings on how power users customize their workflow?"

   • "Is there research on why customers downgrade their plan?"

   • "What do we know about how teams onboard new members?"

3. Preparing for stakeholder conversations

   • "What findings do we have about customer satisfaction with support?"

   • "What has research shown about the most requested integrations?"

   • "What do we know about how customers measure ROI from the product?"

These questions work here because the answers have already been synthesized and documented as findings. You're not asking AI to form a conclusion; you're asking it to find one that already exists in your knowledge base.

Tips for better results with the Magazine AI search

• Phrase questions as knowledge queries, not analysis requests. "What do we know about X?" and "What have we learned about Y?" tend to work better than "What should we do about X?" because you're looking for documented findings, not asking AI to form a new opinion.

• Search the Magazine before starting new research. It's a quick way to check whether a question has already been answered, or whether existing findings can inform your next study before you run it.

• If your search doesn't return what you're looking for:

  • The Magazine will suggest alternative phrasings to help you get there. Try them; often it's just a matter of how the question is framed, not whether the information exists.

  • If there genuinely is no relevant research published, that's useful to know too. It tells you where your knowledge gaps are.

You can query your Insights Magazine directly from Slack or Microsoft Teams without leaving the conversation. This is useful when a stakeholder asks a research question mid-discussion and you want to surface a validated finding on the spot, or when you want to make research accessible to teammates who don't work in Condens day to day.

The search works the same way as in the Magazine itself: it searches across your published findings and links back to the source. For setup instructions and full details, see the Slack integration article and Teams integration article.

If your team uses AI tools like Claude or ChatGPT, you can connect them to your Insights Magazine through the Condens MCP server. This lets you query your validated research findings directly from those tools, making it easy to bring research context into wherever your team is already working. Furthermore, you can combine Condens data with other sources, like Notion, Jira, HubSpot, and many more, all within the same conversation.

For a full list of MCP-compatible tools and best ways to use these connections, check out the article below:

Connect Condens to your AI assistants with Condens MCP

See how to bring customer insights into your AI assistants, custom company LLMs, and agentic workflows.

LEARN MORE ABOUT MCP