← All articles

Research Synthesis with AI

Research · · 2 min read

This one is unusual because the dedicated tools are losing to the general ones, so the question is less “which product” and more “when do you need a product at all.”

What you’re synthesizing. Interview transcripts, usability session notes, survey open-ends, support tickets, sales-call recordings, app reviews. The common shape: lots of unstructured text, one question — what are people actually telling us?

The two options:

  • Dedicated research tools — Dovetail, Marvin, Condens, Notably. Built for research teams: transcript storage, tagging, highlight reels, a searchable repository across studies. Their AI features summarize and cluster within the tool.

  • Your reasoning partner plus a folder. Transcripts in a directory, one good prompt, outputs you control. In 2026 this produces synthesis at least as good as the dedicated tools for a single study, and it’s free beyond what you already pay.

When the dedicated tool earns its cost:

  • A research team of more than two people who need to find last year’s findings.

  • Video: if you need clipped highlight reels for stakeholders, the general tools can’t do that.

  • Compliance: consent tracking, participant data handling, retention rules. Dedicated tools do this properly; a folder doesn’t.

If none of those apply, you don’t need one yet. Most product leaders at startups don’t.

The test, whichever way you go. Take a study you’ve already synthesized by hand — you have these. Run the raw transcripts through the candidate and compare against your own findings:

  1. Patterns. Did it find the themes you found? Missing one is bad; finding a real one you missed is the point.

  2. Contradictions. Did it notice where participants disagreed, or did it flatten them into a majority view? This is where most tools fail and most value hides.

  3. Traceability. Every claim links to a quote, and the quote is real. Check three at random. If one is paraphrased into something the person didn’t say, the tool is not usable for decisions.

  4. Surprises. Does it flag the outlier who said something that fits no theme, or drop it?

  5. Speed to first read. Under an hour from transcripts to something you’d share, or it hasn’t changed the timing problem.

Two cautions:

  • Synthesis quality is mostly prompt quality. “Summarize these interviews” produces mush. “Find claims made by three or more participants, list where participants directly disagree, quote each with speaker and line” produces something you can act on. Write the prompt once, carefully, and reuse it.

  • Don’t let the tool replace listening. Read at least two transcripts in full before you look at any synthesis, so you have a feel for the people the numbers came from. Otherwise you’ll believe a tidy summary of a messy reality.

For you, this test is Part 3 of your series, and the prompt is most of the product. If you build the synthesizer well, the test above is its eval.