Skip to content

Skill

Synthesize Research

anthropics

Turns a pile of already-collected qualitative input - interview notes, survey responses, support tickets, sales-call feedback - into one structured read: themes grouped and counted, findings ranked by frequency and impact, and recommendations tied back to the evidence. It is about COMPRESSING many scattered sources into a single coherent synthesis, not running an experiment, sizing a market, or filing and circulating the raw materials.

Situations it fits

After your customer-champion squad gathers two months of angry Yelp screenshots, Zoom transcripts, and handwritten thank-you cards, you need a single plain-language memo that tells your product lead which gripes happen most often and which ones would actually drive away renewals.
Synthesize Research. This skill takes a messy pile of raw qualitative input - screens, transcripts, and notes - and turns it into one structured read that ranks grievances by frequency and flags the ones that actually threaten renewals. It wins because the scenario is all about compressing scattered customer feedback into a ranked summary, which this skill is specifically built to do.
A product manager has fifty customer interviews, two months of support tickets, and dozens of sales call summaries and needs one short report that groups the recurring complaints and feature requests, shows how often each occurred, and recommends the top three actions tied back to specific quotes and threads of evidence.
Synthesize Research. This is a classic synthesis job: compress many raw qualitative sources into a coherent, prioritized set of findings with counts and actionable recommendations linked to evidence. The closest mistaken choice might be the market analysis option (market-research-affaan-m), because that also makes recommendations, but market research focuses on sizing, competitive positioning, and investment decisions rather than extracting themes and quotes from interview and support data. The reproduction workflow (ai-research-reproduction-lllllllama) is wrong because there is no code or numeric experiment to re-execute. The literature‑survey workflow (research-papers-firecrawl) is incorrect because it targets published scholarly work and citation tracing, not internal customer conversations and ticket logs.

3 scenarios in the bank answer to Synthesize Research. The rest are in the game.

Skills it gets confused with

These share a family with Synthesize Research, which is another way of saying they are the ones you might reach for by mistake.

Knowing what Synthesize Research does is the easy half. Telling it apart from the others under time pressure is the game.

Today's session

Synthesize Research is part of anthropics/knowledge-work-plugins. Licence: Apache-2.0. The description above was written for this game, not taken from the skill.