Research synthesis with AI, without losing the nuance
Use AI to move faster through research while keeping the surprises.

Zander Whitehurst
9 min read
Research synthesis has always been one of the slowest parts of product work. Hours of interviews turn into pages of notes, then into sticky notes, then into themes, then into insights. It is valuable work, but it takes time that teams rarely have.
Some of the best product decisions I have ever made came from one throwaway comment in an interview. A participant once said "oh, I just screenshot it and send it to my boss" and that single sentence reshaped an entire feature. No summary would have flagged it. It only came up once.
AI can make synthesis dramatically faster. It can also quietly remove the most valuable parts of research if you are not careful. Here is how to get the speed without losing the nuance.
What AI is good at
AI handles the mechanical parts of synthesis well. Given transcripts or notes, it can:
- Group similar comments into themes
- Pull out direct quotes that support each theme
- Count how often topics come up
- Summarise long conversations into key points
- Translate and tidy notes from different sources
These tasks used to take days of manual work. Now they can take minutes, which means you can spend more time on interpretation and less on sorting.
What AI tends to miss
AI summaries naturally gravitate towards the average. They highlight what most people said and smooth over what only one or two people said. That is exactly where many of the most important insights live.
A single participant describing an unexpected workaround, a moment of frustration that contradicts the rest of the data or a casual comment that reveals a hidden need can be worth more than the most common theme. These are the surprises that change product direction, and they are easy for a summary to lose.
AI also cannot see what was not said. Hesitations, body language and the tone in someone's voice often carry meaning that never makes it into a transcript.
Read the raw material first
Before you ask AI for any synthesis, spend time with the raw material yourself. Read at least a few transcripts in full, or rewatch key moments from interviews.
This does two things. It gives you your own sense of what matters, so you can judge whether the AI synthesis is accurate. It also makes it much more likely that you will notice the surprising details that a summary would hide.
Ask for outliers, not just themes
When you do ask AI to help, change the questions you ask. Instead of only asking for the main themes, ask for:
- Comments that contradict the main themes
- Things only one participant mentioned
- Moments where people described workarounds
- Places where what people said differs from what they did
Prompts like these push the analysis towards the edges, where new insights are more likely to appear.
Keep evidence attached to every claim
Every insight should link back to evidence. When AI suggests a theme, ask it to list the quotes and participants behind it. Then check a few of them yourself.
This protects against a common problem: AI confidently presenting a theme that is only weakly supported, or occasionally inventing details. Keeping evidence attached makes your synthesis trustworthy and makes it much easier to share with stakeholders.
Separate observations from interpretations
Good synthesis keeps a clear line between what people said or did and what you think it means. AI often blends the two, presenting interpretations as if they were facts.
Structure your synthesis in layers:
- Observations. What happened, in participants' own words.
- Patterns. What repeated across participants.
- Insights. What you believe the patterns mean.
- Opportunities. What you might do about it.
AI can help with the first two. The last two need your judgement and your understanding of the product and business.
Involve the team
Synthesis should not happen in isolation. Share early themes with designers, PMs and engineers and ask what surprises them. Different perspectives catch different things, and people are far more invested in insights they helped shape.
AI makes it easy to create a first draft quickly, which leaves more time for this kind of collaborative discussion. We cover AI assisted discovery like this in our AI Product Management Bootcamp.
Keep a research memory
Over time, teams build up large amounts of research that is rarely revisited. AI can help here too, by making past research searchable and easy to summarise when a new question comes up.
Store transcripts and notes in a consistent place, tag them clearly and use AI to pull relevant findings into new projects. Old research often answers new questions, if you can find it.
My rule of thumb is simple. If the AI synthesis contains nothing that surprises me, I have not looked hard enough. Real research is messy and full of contradictions. If the summary feels perfectly tidy, it is probably hiding the interesting bits.
Faster, not shallower
The goal is to use AI to move faster through the mechanical parts of research so you can go deeper on the human parts. If your synthesis becomes quicker but also flatter, something has gone wrong.
Treat AI as a capable research assistant: fast, tireless and good at organising. Keep the interpretation, the surprises and the final judgement firmly in your own hands.
Memorisely