skills / research

research-synthesis

what it does

Synthesize user research into themes, insights, and recommendations. Use when you have interview transcripts, survey results, usability test notes, support tickets, or NPS responses that need to be distilled into patterns, user segments, and prioritized next steps.

Verified Before and after: 2 models 1 try per case View source Download .zip

Compare the answers

Answers from Claude Sonnet 5.5 and GPT-6.1 Sol · try 1 of 1

Same request, same model, once with the skill and once without. Read both and judge for yourself.

“Summarize what we learned from this usability test on report export.”

Source material given with the request

Task: "Export last month's sales report as a CSV file." 5 participants, moderated, remote. Post-task question: "How easy was that?" (1 = very hard, 7 = very easy) P1: Looked in Settings, then Help. Found Export under the "..." menu after 4 min. Rated 6. "Pretty easy." P2: Looked in Share, then Settings. Gave up after 5 min. Rated 3. P3: Opened the "..." menu straight away, exported in 40 s. Rated 7. P4: Looked in Share, gave up after 4 min. Rated 5. "Fine once you know where it is." P5: Hovered over toolbar icons, found Export under "..." after 3 min 30 s. Rated 6. "Easy enough."
with the skill
## Research Synthesis: Report Export Usability Test **Method:** Moderated remote usability test | **Participants:** 5 **Date:** Not provided | **Researcher:** Not provided **Task:** "Export last month's sales report as a CSV file." ### Executive Summary Export is hard to find. Only 3 of 5 participants completed the task unaided, and only 1 of 5 (P3) did so efficiently (40 s). The other two completers took 3.5-4 minutes, and two participants gave up after 4-5 minutes. Every successful path ended at the "..." menu, but 4 of 5 participants looked elsewhere first (Settings, Help, Share, toolbar icons). The satisfaction ratings (mean 5.4/7) are noticeably more positive than the behavior suggests, so they shouldn't be read as "mostly fine." ### Key Themes #### Theme 1: Export is hidden in the "..." menu, which doesn't match users' expectations **Prevalence:** 4 of 5 did not go to "..." first **Observations:** - Where participants looked first: Settings (P1, and P2 second), Share (P2, P4), toolbar icons (P5), Help (P1 second). - Only P3 opened "..." immediately. - Two participants looked in Share, suggesting they associate export with sending or sharing data. **Supporting evidence:** - P1 searched Settings, then Help, and found Export after 4 min. - P5 "hovered over toolbar icons" and found it after 3 min 30 s.
Show the rest (60 more lines)
- P2 and P4 never found it. **Implication:** The current location fails for most first-time users. Share and Settings are the mental models to design around, and either could be a more discoverable home or entry point. #### Theme 2: Task failure and time-on-task are high **Prevalence:** 2 of 5 failed; 3 of 5 succeeded **Observations:** | Participant | Outcome | Time | Rating | |---|---|---|---| | P1 | Success | ~4 min | 6 | | P2 | Gave up | 5 min | 3 | | P3 | Success | 40 s | 7 | | P4 | Gave up | 4 min | 5 | | P5 | Success | 3 min 30 s | 6 | Among the 3 completers, times were 40 s, 3 min 30 s, and 4 min. Excluding P3, successful participants needed roughly 3.5-4 minutes. **Implication:** Discovery is the problem. Nobody reported trouble once they found Export, though this test didn't isolate the later steps, such as choosing CSV or the date range. #### Theme 3: Self-reported ease doesn't match behavior **Prevalence:** 4 of 5 rated 5 or higher **Observations:** - P4 gave up after 4 min but rated 5 ("Fine once you know where it is"). - P1 took 4 min and rated 6 ("Pretty easy"). - P5 took 3.5 min and rated 6 ("Easy enough"). - Only P2, who gave up, rated low (3). **Interpretation (not directly observed):** Participants may rate the ease of the action once found, not the search. Politeness or moderator effects could also play a part. P4's comment implies they learned the location, possibly with help from the moderator, which should be confirmed. **Implication:** Don't use the post-task rating alone as a success metric for this flow. Use completion, time-on-task, and path data. ### Insights → Opportunities | Insight | Opportunity | Impact | Effort | |---|---|---|---| | Users look in Share and Settings first | Add an Export entry point to Share (and possibly Settings) | High | Low | | Export is buried in "..." | Promote Export to a visible toolbar button or label the menu item more clearly | High | Low-Med | | Ratings overstate ease | Track completion and time-on-task as primary metrics | Med | Low | | 2 of 5 abandoned after 4-5 min | Add a search or help route that surfaces "Export" | Med | Med | ### User Segments Identified Five participants isn't enough to define segments. The only visible split is behavioral, by whether people went to "..." first (P3) or not (the other four). We don't know whether P3 had prior experience with the product. ### Recommendations 1. **Make Export discoverable outside "..."**: add a visible button or a Share entry. 4 of 5 participants missed it there, and 2 failed. 2. **Re-test after the change**, measuring completion and time-on-task. Target: all participants completing in under about a minute. 3. **Don't rely on the 1-7 ease question alone**, since it conflicted with observed behavior. ### Questions for Further Research - Was P3 a more experienced user, or was "..." a lucky first guess? - Did P4 eventually find Export with help? What did "once you know where it is" refer to? - What word or location do users expect for export ("Download," "Share," "Save as")? - Does the flow after finding Export (CSV selection, date range) have its own problems? ### Methodology Notes - **Sample is very small (n=5).** Treat counts as directional, not statistically reliable. - Participant roles, experience level, and recruitment criteria weren't provided, nor were the dates or researcher. - Time cutoffs for giving up (4-5 min) may reflect moderator judgment, not user choice. - Only P1, P4, and P5 have quotes, and they came from the post-task rating, so evidence on *why* participants looked where they did is thin. I haven't inferred reasons beyond what's labeled as interpretation above.
without
# Usability test summary: CSV report export ## Bottom line **Export is hard to find, and the satisfaction ratings hide that.** Only 1 of 5 participants completed the task efficiently, yet the average ease rating was 5.4 out of 7. ## What happened | | Outcome | Time | Rating | |---|---|---|---| | P1 | Success (after Settings, Help) | ~4 min | 6 | | P2 | Gave up (Share, Settings) | 5 min | 3 | | P3 | Success, went straight to "..." | 40 s | 7 | | P4 | Gave up (Share) | 4 min | 5 | | P5 | Success (after hovering toolbar icons) | ~3.5 min | 6 | - **Completion:** 3 of 5 (60%). Two participants gave up after 4-5 minutes. - **Efficiency:** Only P3 was quick. The other two successes took 3.5-4 minutes, roughly 5-6 times longer. - **Cause:** Export sits under the "..." overflow menu, and most participants looked elsewhere first. Share (P2, P4), Settings (P1, P2), and Help (P1) were all checked before it. Share is a likely mental model for "get data out," and both people who stopped there gave up. ## Ratings don't match behavior
Show the rest (16 more lines)
- P4 gave up but rated the task 5 ("Fine once you know where it is"). - P1 and P5 needed 3.5-4 minutes but rated 6 ("Pretty easy," "Easy enough"). - Only P2, who failed, gave a clearly low rating. Possible explanations, which this test can't separate: participants may have been shown the answer after struggling, they may have rated their final state rather than the whole experience, or they may have been polite with a moderator. Either way, the ease question alone would have made this feature look fine. ## Recommendations 1. **Make Export more discoverable.** Put it in a visible toolbar button or under Share, or at least add it in both places. 2. **Don't rely on the "..." menu** for a core task like this. 3. **Report behavioral metrics** (completion, time, wrong paths) alongside self-reported ease in future tests. 4. **Retest** after the change, using the same task. ## Caveats - Five participants is enough to spot a discoverability problem, but not to estimate rates precisely. - The notes don't say whether P2 and P4 were shown the Export location after giving up. That affects how to read their ratings. - The notes don't say how experienced participants were with the product, which could explain why P3 went straight to the right place.

Answers are shown exactly as the model wrote them, from the first try, and nothing is picked by hand.

The models

2 models

Claude Sonnet 5.5

made by Anthropic

claude-sonnet-5-5 · revision not reported by the provider

1 try per case · tested 1 Nov 2026

GPT-6.1 Sol

made by OpenAI

gpt-6.1-sol · revision not reported by the provider

1 try per case · tested 1 Nov 2026

The tasks

written by Carte Blanche
biased-feature-survey

“We surveyed users about what to build next. Write up the findings for the team.”

expense-interviews

“I ran six interviews about our expense app last week. Can you pull out the main themes and what we should do about them?”

export-usability-test

“Summarize what we learned from this usability test on report export.”

nps-verbatims

“Here are last month's NPS responses. What's our score, and what's driving it?”

support-ticket-spike

“Group these support tickets and tell me what we should look at first.”

Install

Download the .zip and unzip it. Put its files, in a folder named research-synthesis, here:

Claude Code~/.claude/skills/Every project. For one project only, use .claude/skills/ inside it.
Codex~/.agents/skills/Every project. For one repository only, use .agents/skills/ inside it.
Cursor~/.cursor/skills/Cursor also reads the Claude Code and Codex folders above.

Then start a new session and ask for what the skill does. Other agents that read SKILL.md have their own skills folder; check their docs.

Coming after launch: get skills from inside your agent, no download needed.