When the user wants to conduct, analyze, or synthesize customer research.
SkillScore83Maintained
Automatic
◆Research pack
Last verified 8/21/2026
What it changes
Passed 20 of 24 test cases with the skill, and 16 of 24 without it.
Without skill67%
With skill83%
Checks it passed that the model alone missed
Advises leaving attributes blank when empirical research data is missing.
Includes context, sentiment, and theme tags for captured community quotes.
Declines to write the landing page conversion copy directly within this research workflow.
and 2 more in the full report
4 fixed16 passed either way4 still failing0 broke
Pass-rate lift
+17 pp
Tokens
+96%
Agent turns
0%
Response time
-18%
Model gemini-3.6-flashMeasured 8/21/2026v120 of 24 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 4 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 4
Prompt
Here are quotes from research on B2B SaaS buyers complaining about pricing transparency. Rewrite our landing page hero section, primary CTA, and feature cards to improve conversion rates.
What was checked, without and with the skill
Checked
Without skill
With skill
Declines to write the landing page conversion copy directly within this research workflow.
Missed
Passed
Focuses output on structuring or clarifying research insights rather than page optimization copywriting.
Missed
Passed
case-03 · gemini-3.6-flash · 8/21/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md240 lines2 files
Mode 1: Analyzing Existing Research Assets
Asset Types
Customer interview / sales call transcripts
Extract: pains, triggers, desired outcomes, language used, objections, alternatives considered
Look for: the moment they decided to look for a solution, what they tried before, what success looks like to them
Survey results
Segment responses by customer tier, use case, or tenure before drawing conclusions
Flag: what open-ended answers say vs. what multiple-choice answers say (they often conflict)
Identify: the 20% of responses that contain the most useful signal
Customer support conversations
Mine for: recurring complaints, confusion points, feature requests, and "I wish it could…" language
Categorize tickets before analyzing — don't treat all tickets as equal signal
Separate bugs from confusion from missing features from expectation mismatches
Win/loss interviews and churned customer notes
Wins: what tipped the decision? What almost made them choose a competitor?
Losses and churn: was it price, features, fit, timing, or something else?
Segment by reason — don't average across different churn causes
NPS responses
Passives and detractors are higher signal than promoters for improvement work
Pair scores with verbatims — a 9 with a specific complaint beats a 10 with no comment
Extraction Framework
For each asset, extract:
Jobs to Be Done — what outcome is the customer trying to achieve?
Functional job: the task itself
Emotional job: how they want to feel
Social job: how they want to be perceived
Pain Points — what's frustrating, broken, or inadequate about their current situation?
Prioritize pains mentioned unprompted and with emotional language
Trigger Events — what changed that made them seek a solution?