YK yukiyuki.ai
Competitive research architecture

竞品调研方法架构

这不是一篇调研总结,而是一套做竞品研究时可以反复使用的结构。每次调研都从决策问题开始,经过市场地图、PMF、产品机制、GTM、早期用户获取和证据分级,最后沉淀成可复用的页面、表格和模板。

Architecture

Competitive Research 的八层结构。

每一层都有自己的输入、关键问题和输出。顺序可以按项目调整,但不要跳过证据分级和最终判断。

Layer 00 调研任务定义

先明确这次研究要支持什么决定。

竞品调研不是为了“了解一下”,而是为了帮助一个明确场景做判断。比如选择工具、设计产品定位、拆增长路径、判断一个赛道是否值得进入。

  • 目标读者是谁
  • 研究对象有哪些
  • 要输出深度页、横向表,还是决策建议
Output
  • Research brief
  • Company list
  • Decision question
Layer 01 市场地图

先判断它们是不是同一种产品。

不要一上来就做功能表。先拆类别、用户、替代方案和生态位置。很多竞品表面相似,实际分别承担入口、网关、观测、治理、工作流等不同角色。

  • 产品类别和替代关系
  • 目标用户和使用场景
  • 上游、下游和生态伙伴
Output
  • Category map
  • Substitute map
  • Positioning summary
Layer 02 事实底座

把基本事实整理成可以引用的底层数据。

包括产品名称、成立时间、创始人、融资、定价、客户、功能、公开规模信号和数据截止日期。事实层要和判断层分开。

  • 官方资料优先
  • 动态数据标注日期
  • 不确定信息标为待补
Output
  • Company profile
  • Source list
  • Data snapshot
Layer 03 PMF 阶段

按时间拆出它为什么能成立。

每家公司都要拆阶段:最早解决谁的什么痛点,为什么有人愿意切换,后来如何从单点价值变成更大的平台价值。

  • 阶段标签和时间范围
  • 目标用户、核心场景、替代方案
  • 为什么选它、为什么不用它
Output
  • PMF timeline
  • Stage thesis
  • Expansion logic
Layer 04 产品机制

拆 Aha Moment、激活、留存和切换成本。

功能清单只回答“它有什么”,产品机制回答“为什么用户会用、会留下、会扩张”。这一层是连接 PMF 和 GTM 的中间层。

  • Aha Moment 是什么
  • 首次激活需要几步
  • 留存来自数据、流程、协作,还是基础设施依赖
Output
  • Mechanism map
  • Activation path
  • Retention drivers
Layer 05 GTM 增长系统

把增长拆成渠道、动作和循环。

用 AARRR、PLG、开源、内容、社区、销售、生态集成和创始人 IP 来拆。重点不是列渠道,而是说明用户如何被获取、激活、留存、付费和转介绍。

  • Acquisition / Activation / Retention / Revenue / Referral
  • 核心分发机制和增长飞轮
  • 个人到团队、团队到企业的扩张路径
Output
  • GTM map
  • Growth loop
  • Channel thesis
Layer 06 前 1000 用户

早期用户获取要拆到具体动作。

把 `0->100`、`100->500`、`500->1000` 分开看。每个阶段都记录渠道、动作、证据类型、转化条件和可复制性。

  • 直接证据:发布记录、访谈、官方复盘
  • 辅助信号:Product Hunt、HN、GitHub、Discord、内容传播
  • 高可信推断:基于时间线和产品机制的判断
Output
  • 0 to 1000 map
  • Cold start actions
  • Copyability notes
Layer 07 证据分级

事实、估算和推断必须分开。

这是整套方法的底线。公开资料不完整时可以推断,但必须清楚标注证据等级,避免把研究者判断写成事实。

  • Level 1:官方和一手材料
  • Level 2:第三方估算和平台信号
  • Level 3:基于公开痕迹的高可信推断
Output
  • Evidence tags
  • Source notes
  • Risk caveats
Layer 08 结论和复用

最后沉淀判断,也沉淀方法。

每次调研结束都要回答:PMF 本质是什么,GTM 本质是什么,最不可替代的机制是什么,哪些动作可复制,哪些条件不可复制。

  • 最终结论和风险张力
  • 可复制 vs 不可复制因素
  • 新增维度写回调研框架
Output
  • Decision memo
  • Reusable framework
  • Next research checklist
Question bank

每次调研默认要问的问题。

这些问题可以直接变成 Notion / Lark Base 里的字段,也可以作为单家公司深度页的章节。

Module Core questions Field examples
调研任务 这次研究服务什么决策?读者是谁?最后要输出给谁看? Research goal, audience, decision type, output format
市场地图 这些公司到底在同一个类目吗?谁是替代品,谁是互补品? Category, target user, substitute, ecosystem role
PMF 最早的痛点是什么?用户为什么从替代方案切过来?哪个阶段开始扩张? Stage, user, pain, alternative, PMF signal
产品机制 Aha Moment 是什么?激活路径短不短?留存靠什么? Aha Moment, activation, retention, switching cost
GTM 用户从哪里来?什么内容或渠道带来转化?增长飞轮是什么? Acquisition, activation, retention, revenue, referral
前 1000 用户 0 到 100 是谁带来的?100 到 500 如何扩散?500 到 1000 是否进入团队工作流? 0-100, 100-500, 500-1000, evidence type
证据 哪些是事实?哪些是估算?哪些是推断?证据够不够支撑结论? Source, evidence level, date, confidence
最终判断 它为什么成立?我们能学什么?什么不能照搬?下一步要验证什么? Thesis, risk, copyable, not copyable, next step
Artifacts

这套架构最后产出四类资产。

以后每次调研都尽量留下结构化资产,而不是只留一段临时回答。

Artifact 01

单家公司深度页

  • 按八层结构写完整
  • 适合放 Notion / 文档库
  • 保留来源和待补数据
Artifact 02

横向对比表

  • 定位、PMF、GTM 对比
  • 早期用户路径对比
  • 可复制动作对比
Artifact 03

决策摘要

  • 结论先行
  • 适用场景和风险
  • 下一步验证动作
Artifact 04

方法迭代记录

  • 新增研究维度
  • 字段和模板调整
  • 下一次默认复用
Codex skill

把这套方法装进 Codex。

这是公开版 skill,不包含本地项目路径、客户信息或私有归档规则。复制后可以放进自己的 Codex skills 目录。

Competitive Research Skill

这个 skill 会在用户提出竞品调研、市场地图、产品对比、定价对比、GTM research 或 user-signal synthesis 时触发。它的目标不是替你堆材料,而是把信息压成一个可以支持决策的 brief。

最小安装方式:在本地 Codex skills 目录中新建 `competitive-research/SKILL.md`,复制右侧内容。详细方法可以继续拆到 `references/method.md`,也可以先只用这一份单文件版本。

Recommended use
  • 先写清楚这次研究要服务什么决策
  • 先画市场地图,再拆单个公司
  • 所有关键结论都标注证据等级和信心
  • 最后必须输出 strategic implications 和 next validation steps
SKILL.md copy ready
---
name: competitive-research
description: Structured competitor and market research for product, GTM, pricing, positioning, user demand, and alternative-solution analysis. Use when the user asks for 竞品调研, competitive research, competitor analysis, market mapping, product comparison, pricing comparison, GTM research, user-signal synthesis, or decision-ready research briefs.
---

# Competitive Research

## Core Workflow

Use this skill to turn competitor information into a decision-ready brief, not a pile of notes.

1. State the decision question before collecting facts.
2. Map the market before deep-diving one company.
3. Analyze each competitor with the same dimensions.
4. Label evidence quality for every important conclusion.
5. Prioritize user signals over website copy.
6. End with implications, open questions, and recommended next validation steps.

## Required Output Shape

Prefer this structure unless the user asks for another format:

```markdown
# Competitive Research: <topic>

## Decision Question
<What decision this research should support>

## Market Map
<direct competitors, indirect competitors, substitutes, platforms, open-source options>

## Competitor Breakdown
<repeat the same dimensions per competitor>

## Comparison Table
<positioning, ICP, core use case, capabilities, pricing, distribution, strengths, weaknesses>

## User Signals
<what users praise, complain about, switch from, and pay for>

## Strategic Implications
<what to copy, avoid, differentiate, test next>

## Evidence Notes
<source, date, evidence level, confidence>
```

## Evidence Levels

Use evidence labels consistently:

- `L1 Official`: company website, docs, pricing page, changelog, official demo, public job posts.
- `L2 Third-party`: user reviews, Reddit/HN/X/YouTube, G2, Product Hunt, media, traffic tools, funding/news databases.
- `L3 Inference`: reasoned interpretation from observed product, pricing, messaging, traffic, or user behavior. Mark as inference.

For important claims, include source, date, evidence level, and confidence.

## Research Rules

- Do not start with "what does this company do?" Start with "what decision are we trying to make?"
- Do not compare competitors using different dimensions unless the asymmetry is the insight.
- Do not treat homepage positioning as truth. Cross-check with docs, pricing, user complaints, and observed workflows.
- Separate feature demand from willingness to pay.
- Separate loud discussion from commercial value.
- Make uncertainty explicit. A useful `unknown` is better than a fake conclusion.

## Detailed Method

For a deeper version, add `references/method.md` with your full teardown framework: market map, fixed competitor dimensions, evidence levels, user-signal analysis, decision outputs, comparison table, and public sharing hygiene.

Claude Project Instructions

Claude 没有 Codex 的本地 skill 自动触发机制,但可以把同一套方法放进 Claude Project Instructions。这样这个 Project 下的所有竞品调研都会默认按这套结构输出。

适合放在 Claude Project 的 `Instructions` / `Project knowledge` 里。相比 Codex 版,它更短、更偏行为约束,不包含 YAML frontmatter。

Best for Claude
  • Project Instructions:长期默认生效
  • Project Knowledge:作为方法文档被引用
  • 一次性对话:直接复制到 prompt 前半段
CLAUDE.md project ready
# Competitive Research Project Instructions

Use this project for structured competitor and market research. The goal is to turn messy market information into decision-ready briefs, not generic summaries.

## Default Workflow

1. Start with the decision question.
2. Build a market map before deep-diving into individual companies.
3. Analyze competitors with consistent dimensions.
4. Separate facts, third-party signals, and inference.
5. Prioritize user behavior and user complaints over homepage copy.
6. End with strategic implications and next validation steps.

## Required Research Dimensions

For each competitor, capture:

- One-line positioning
- ICP and primary user
- Core use case and adoption trigger
- Core capability and product mechanism
- Activation path, retention driver, and switching cost
- Business model and pricing
- Distribution and GTM channels
- Moat and weaknesses
- User sentiment and repeated complaints
- What we should learn, avoid, or test

## Evidence Rules

Use evidence labels:

- L1 Official: website, docs, pricing page, changelog, official demo, public job posts.
- L2 Third-party: user reviews, Reddit, Hacker News, X, YouTube, G2, Product Hunt, media, traffic tools.
- L3 Inference: reasoned interpretation based on observed product, pricing, messaging, traffic, or user behavior.

Important claims should include source, date, evidence level, and confidence. If evidence is weak, say so directly.

## Output Format

Prefer this structure:

1. Decision Question
2. Market Map
3. Competitor Breakdown
4. Comparison Table
5. User Signals
6. Strategic Implications
7. Open Questions
8. Next Validation Steps

Do not produce a polished-sounding report if the evidence does not support it. Make uncertainty visible.