FigJam’s AI can sort a selection of stickies into groups and write a summary of any pile you point it at. For anyone who’s spent an afternoon dragging 300 interview notes around a board, that’s a real time saver. It’s also exactly where research synthesis is most likely to go quietly wrong.

This walkthrough is synthesized from Figma’s own documentation, Miro’s competing clustering feature, a Nielsen Norman Group talk and an academic study on how well AI handles research synthesis, and a couple of practitioner guides, plus general FigJam knowledge. It isn’t any one person’s process, and we haven’t run it end to end on a study of our own.

The idea that holds it together: let the AI do the sorting, and keep the deciding for yourself. Nielsen Norman Group’s testing, presented at Advancing Research 2026, found that AI tools can speed up experienced researchers a little but hallucinate findings, miss meaningful patterns, and produce shallow summaries. They also found that the output looks polished enough to hide those flaws. The same talk noted AI does better at grouping data that humans have already coded.

  1. Write atomic notes. One observation, quote, or behavior per sticky, written plainly enough to make sense on its own. AI grouping works on the words in each note, so “P4: couldn’t find export, went to Help” sorts far better than “export??”.

  2. Encode the participant in color, not authorship. FigJam can sort by Author, but that means whoever typed the sticky, which in a team synthesis is the note-taker, not the participant. Give each participant or session its own sticky color and put an ID at the start of each note. That gives you a way to check later whether a theme comes from many people or from one talkative participant.

  3. Do a light coding pass first. Before you hand anything to the AI, skim the notes and add short tags or labels where the meaning isn’t obvious from the words alone (for example, “workaround” or “trust issue”). This matches what Nielsen Norman Group saw working best, and it gives the AI something firmer than raw phrasing to group on.

  4. Sort by Topic. Select the stickies, click Organize in the selection toolbar, and choose Sort stickies, then Topic. FigJam copies the selection into a new section and groups the copies, so your original notes stay where they were. That’s useful: keep the original wall as your source of truth.

  5. Treat every group as a guess. Go through each cluster and ask whether the notes actually belong together, or just share a word. A synthesis guide from Articos (a company that sells AI research tools, so not a neutral party) notes that AI clustering tends to stay surface-level and keyword-driven, so two notes that both mention “search” might land together even though one is a complaint and the other is praise. Move misfits by hand and rename each section in your own words.

  6. Check breadth with your colors. Look at each cluster’s mix of sticky colors. A cluster that’s mostly one color is one participant’s story, however big it looks. The same Articos guide’s rule of thumb is to treat themes with thin support as hypotheses rather than findings. Frequency isn’t the same as importance either, and that judgment stays with you.

  7. Summarize per cluster, then edit. Select two or more stickies in a cluster, click Organize, then Summarize. The summary appears as a new object on the board. You can’t edit it directly, so copy its text into a sticky or text box and rewrite it. Use the summary as a draft label, not an insight. Every sentence you keep should point back to specific notes on the wall.

  8. Write the insights yourself. The “so what” (which pattern matters, why, and what the team should do about it) is the part every source here says AI can’t do well. Draft it as a team. The AI-sorted board is the material, not the conclusion.

A different approach worth knowing. Miro clusters stickies by keyword or by sentiment rather than by topic, and it generates a keyword title for each cluster. It can also cluster by tag, which pairs well with the light coding pass in step 3. Keyword clustering is more literal and easier to predict, which makes its mistakes easier to spot. Sentiment clustering is handy for a quick split of usability feedback into positive, neutral, and negative groups. FigJam’s topic sort tries to read meaning, which can produce better first groupings but makes it harder to see why a note landed where it did. Either way, step 5 still applies.

Where the sources disagree. One guide suggests asking an AI to generate an affinity diagram from a description of your product before any research exists, then treating its themes as questions to investigate. That can help you plan interviews, but it isn’t synthesis. Nothing on that board came from a user, and it should never sit next to real findings looking like one.

Why it matters for designers

Plenty of product designers run their own research without a dedicated researcher or a research repository tool, and the synthesis step is where time runs out. Doing the mechanical grouping in seconds means more of that time can go to the interpretive work that actually changes a design.

It also helps with a team dynamic that a recent academic study picked up on. In interviews with UX researchers, product managers, and designers, researchers expressed limited trust in AI results, while product managers often overestimated what AI could do. A visible process, with AI sorting on one side and human-checked themes on the other, makes it easier to show a stakeholder what the AI did and didn’t decide.

What to watch out for

Polished output is the main risk. A neatly labeled cluster with a confident summary looks finished. Nielsen Norman Group’s warning is that closer inspection can reveal gaps that could mislead stakeholders. Budget real time for step 5.

Watch what’s in your notes. Research notes often carry names, job titles, and details participants shared in confidence. The sidebar covers where FigJam AI sends text. If your consent forms don’t cover third-party AI processing, anonymize before you sort.

Figma’s documentation on these features has changed over time, including how AI access is priced, and older pages describe them as a free beta. Check what your plan includes today.

The sources here cover the features and the research on AI synthesis in general. None of them is a hands-on accuracy test of FigJam’s topic sort specifically, so treat its groupings with the same skepticism you’d give any AI first draft.