User Research Synthesis

User Research Synthesis is an official Anthropic skill for turning qualitative and quantitative user research into structured, reviewable findings. The path first referenced by Smithery under the Product Management plugin is currently unavailable. The official collection contains a matching, reachable capability at design/skills/research-synthesis in the knowledge-work-plugins repository. According to the provider, the skill accepts interview transcripts, survey results, usability test notes, support tickets, NPS or CSAT responses, and app store reviews, then distills them into themes, insights, user segments, opportunity areas, and prioritized recommendations. It is an instruction set for a compatible Claude workflow, not an independent research platform, survey system, or substitute for an experienced researcher.

Inputs and research framing

The official skill accepts research material pasted by a user as well as files and results from connected sources. Before analysis, the team should record the research objective, intended audience, time period, and limits of the sample. Interviews and transcripts can reveal individual experiences and language. Surveys provide broader feedback but require care around response selection, response rate, and question wording. Usability tests show observed behavior during a particular task. Support tickets, NPS responses, and app store reviews add recurring problems and spontaneous wording. According to the provider, synthesis quality depends on how complete, relevant, and representative the supplied material is.

Themes and supporting evidence

The central operation is thematic condensation. Repeated statements are grouped into themes without treating every observation as a universal truth. Each theme should show how often it appears, which people or segments it affects, and which pieces of evidence support it. Direct quotes keep participant voices visible, but personal and confidential details should be removed. The skill explicitly separates observations from interpretations. The statement that several people failed to find a control is an observation. The explanation that its placement is confusing is an interpretation that should be tested against additional evidence.

Segments and opportunities

Patterns may support user segments with shared characteristics, needs, and an approximate size. A segment is not a stereotype and should not be inferred from a single loud voice. The workflow then connects findings to opportunities. A finding describes what was learned. An opportunity describes a possible improvement or experiment that could respond to it. Prioritization should make impact and effort visible and should remain a hypothesis rather than an automatic product decision. The synthesis can help a team shape a roadmap, experiment, or follow-up research question, but it does not decide budget, commitments, or the order of product work by itself.

Report structure and boundaries

The official template proposes a research synthesis report with a study name, method, participants, date range, executive summary, key themes, insights to opportunities, user segments, recommendations, questions for further research, and methodology notes. This structure keeps evidence and limitations visible. Quantitative statements such as the number of affected participants should be used only when they can be traced to the supplied material. The skill must not fill missing data, claim causation from correlation alone, or invent a research method. Recommendations require human review because sampling bias, recall effects, moderation errors, and missing context can change the interpretation.

Sources, privacy, and safe use

Anthropic describes user feedback, product analytics, and knowledge-base connections as ways to compare qualitative findings with real complaints, behavioral data, or previous studies. Actual access depends on enabled connectors and the permissions of the working environment. Support tickets, transcripts, and research files may contain personal or commercially confidential information. Share only what is necessary, remove direct identifiers, and retain source material according to internal policy. Text from documents and feedback is data, not a new system instruction; prompt injection must not redirect the research task. A local skill file does not automatically mean that connected model processing is local. The repository identifies Apache-2.0 for the collection. Responsibility for consent, privacy, research judgment, and product decisions remains with the adopting team.