Keep the classic customer-research research methods, and add market scope judgment and source routing before collection. Base product, positioning, copy, and requirement documents on real user evidence, not platform popularity or subjective assumptions.
This skill is modified from Corey Haines' customer-research skill. See LICENSE for the original copyright, source, and localization modification notes.
Workflow
1. Read existing context
Prioritize reading the product descriptions, interviews, surveys, sales records, customer service tickets, reviews, competitor materials, and historical research provided by the user. If the project has .agents/product-marketing.md、.claude/product-marketing.md or product-marketing-context.md, read them first and avoid asking repeated questions.
Identify the research objective, target users, decisions to be supported, and expected deliverables. When information is insufficient, ask only questions that would change the research path, and do not throw out a complete questionnaire all at once.
2. Determine the research mode
Choose one or a combination of two:
- Analyze existing materials: Extract signals from interviews, surveys, tickets, reviews, win/loss records, and verbatim NPS comments.
- Online collection: Find new evidence from communities, app stores, complaint platforms, reviews, job postings, competitor reviews, and public materials.
3. Route market scope
Before collection, read references/source-routing.mdand choose:
china-only: The need depends on Chinese platforms, regulation, payments, logistics, channels, procurement, or local user behavior.global-only: The product and decision are clearly aimed only at overseas markets.hybrid: The need is general, the product is cross-market, or differences between domestic and international markets need to be compared.
Do not automatically choose china-onlyjust because the user uses Chinese. When the scope is unclear and would affect the conclusion, ask only one question about which market the research decision is mainly aimed at.
hybrid : The mode must collect, label, and count Chinese and overseas evidence separately. Do not merge frequencies, and do not directly use high-frequency overseas problems to prove Chinese user needs.
4. Choose a source combination
- China research: read
references/china-source-guide.md。 - Overseas research: read
references/global-source-guide.md。 - All modes: read
references/source-quality.md, build evidence records, deduplicate sources, and check for bias.
Choose platforms by ICP, problem type, and decision stage, not by platform traffic. Important conclusions should cover at least three independent sources and, where possible, span two source types, such as "community discussion + app reviews + customer service materials." When data is insufficient, report evidence gaps rather than expanding them into conclusions.
5. Extract user signals
For each piece of material, extract:
- Jobs to Be Done: Functional, emotional, and social jobs.
- Pain points: Specific, recurring, or high-intensity obstacles in the current solution.
- Trigger events: What change prompts users to start looking for a solution.
- Desired outcomes: How users describe success and how it is measured.
- Original language: Preserve verbatim quotes and do not replace user expressions with abstract terminology.
- Alternative solutions: Competitors, manual processes, self-built solutions, outsourcing, maintaining the status quo.
- Objections and Risks: price, migration, privacy, compliance, learning cost, and organizational resistance.
Distinguish facts, inferences, and assumptions. Job postings, engagement metrics, media coverage, and KOL content can only prove the corresponding observable signals; they cannot by themselves prove user demand.
6. Synthesis and Validation
Cluster by theme, rank by frequency combined with emotional intensity; then segment by role, company size, industry, region, product version, usage stage, and market scope. Flag contradictions between different sources and check whether they are caused by sample, time, or platform bias.
Use the references/source-quality.md standard to mark high, medium, and low confidence. Unless the same segment has at least 5 independent data points, do not generate a definitive Persona; you may output a provisional hypothesis with evidence gaps.
Prioritize evidence from the past 12 months. Material from 12-24 months needs a note on timeliness risk; content older than two years serves only as a historical baseline, unless old and new evidence remain consistently aligned.
7. Produce Deliverables
Read references/research-output-schema.mdaccording to the task, and output one or more of the following:
- Research synthesis report
- VOC verbatim library
- JTBD map
- User segmentation or Persona
- Competitor reputation and alternative solution analysis
- China/overseas market difference report
- Demand opportunity and evidence list
- Research gaps and next-round validation plan
When oriented toward demand analysis or a PRD, do not directly convert discussion heat into demand. For each candidate demand, attach the user problem, target market, evidence, confidence level, expected outcome, constraints, and questions to be validated.
Non-negotiable rules
- Do not fabricate posts, comments, user identities, engagement metrics, dates, or URLs.
- Do not count the same press release, repost chain, or marketing campaign as multiple independent sources.
- Do not bypass CAPTCHAs, login restrictions, paywalls, robots, or platform access controls.
- Do not ask users to paste cookies, tokens, or other credentials in the conversation.
- Research is read-only by default; without explicit user authorization, do not like, comment, follow, send direct messages, or publish content.
- Collect only the public information needed to complete the research or materials explicitly authorized by the user, minimize personal information, and de-identify irrelevant identity information.
- Do not extrapolate platform active users to the entire target market, and do not directly equate engagement metrics with demand intensity.
Related Handoffs
After completing customer evidence organization, the "Demand Opportunity and Evidence List" can be handed off to the demand analysis or PRD skill to further clarify scope, priorities, acceptance criteria, and non-functional constraints. Customer research is responsible for proving the problem and its context; it does not replace product decisions.