大多数产品翻车都能追溯到一段模糊的问题陈述。团队直接跳进功能，然后在 scope 上吵几个月。下面 15 个 Prompt 强制你在画第一张图之前，把用户、痛点、触发时刻和”什么都不做的代价”讲清楚。覆盖创始人段（一段话）、PRD 段（一页）、高管简报段（三条 bullet）三个版本。
适合哪些场景
正在打磨新产品的创始人、写 PRD 开头的 PM、做 discovery 的产品设计师、想阻止自己写错方向的独立开发者。
什么时候不建议这样写 Prompt
产品已经上线、只需要发布文案时不要用。如果它本质上是一个研究问题而不是产品问题——先去访谈再写陈述。
Prompt 结构公式
一份强的问题陈述 Prompt 总带这六个要素：
- 角色：让 AI 扮演谁（资深 PM / 独立创始人 / 产品设计师 / 独立开发者 / 增长负责人）。
- 上下文：阶段（想法 / MVP / 增长 / 规模化）、团队规模、流量或 ARR、平台（web / iOS / Android）、受众、限制。
- 目标：一个具体交付物——一段 PRD、一组用户故事、一个实验设计、一篇上线公告。
- 限制：时间线（本 sprint / 本季度）、要砍的范围、不能动的东西（现有流程、计费、合规）。
- 输出格式：表格、清单、可贴 ticket 的 JSON、或带标签的段落，能直接粘到 Linear / Notion / Jira。
- 示例 / 信号：1-2 份你欣赏的参考或竞品、加 1 个想避开的反例。
这套 Prompt 适合用在哪
- PRD 开头段
- Pitch 第二页（问题）
- Discovery 访谈启动
- 季度 roadmap 复盘
- 独立开发者在动手前自检
15 个可直接复制的 Prompt 模板
1. 经典”在为 Y 解决 X，因为 Z”
最有用的一条。任何 PRD 之前都跑一次。
You are a senior PM. Help me write a problem statement in the form: "We are solving {problem} for {specific user segment} because {consequence today}." Constraints: less than 40 words, no buzzwords, the user segment must be narrow enough that I could name 3 real people in it.
Draft inputs: {raw idea, target user, pain you noticed}可替换变量： raw idea, target user, pain you noticed
优化建议： 输出泛泛时追加：“Rewrite so the consequence today is a specific moment in the user week, not an abstraction.”
2. PRD 级一页问题段
Write a one-page problem statement for a PRD. Sections: (1) User segment in one sentence, (2) Current workaround they use, (3) What breaks about the workaround, (4) Cost of inaction in money or time, (5) Why now (what changed in market, tech or behavior), (6) Why us (asymmetric advantage). Each section 2-3 sentences.
Context: {paste discovery notes or 3 user quotes}3. 三条 bullet 的高管版
Compress this problem into 3 bullets for an exec briefing. Bullet 1: the user and pain in 15 words. Bullet 2: the cost of doing nothing this quarter. Bullet 3: the wedge — what we can ship in 6 weeks to learn.
Source: {paste long problem doc}4. 五个为什么深挖
Take this surface-level problem and run a five-whys interrogation. Start from the symptom and dig down 5 layers until you land on either a behavior, a system constraint or a belief. Output as 5 numbered Q-A pairs, then a one-sentence root statement.
Surface problem: {paste}5. 问题 vs 症状诊断
Below is what my team calls "the problem". For each line, classify: is this a problem, a symptom, or a solution-in-disguise? Suggest a rewrite that puts every item in the right column.
{paste raw problem list}6. JTBD 改写
Reframe this product problem as a job-to-be-done statement: "When {situation}, I want to {motivation}, so I can {expected outcome}." Then list 2-3 functional, 2-3 emotional and 1-2 social dimensions of the job.
Product idea: {paste}7. 反向问题陈述
Write the opposite version of this problem statement — what would make the problem worse on purpose. Use it to surface hidden assumptions about what "better" looks like. End with 3 assumptions to test.
{paste current problem statement}8. 痛点量化器
For the problem below, estimate: (a) how often the pain occurs per user per week, (b) time lost per occurrence, (c) approximate $ value of that time, (d) total annual pain per user. Show your assumptions. Flag any number where confidence is less than 50%.
Problem: {paste}9. “最小有意义受众”过滤
Take this broad problem and narrow the user segment 3 times. Each pass: cut the segment in half by adding a constraint (role, stage, tool used, frequency, geography). End with a segment you could name 5 real people in.
Broad statement: {paste}10. 问题陈述批评轮
Critique this problem statement on 7 dimensions: (1) specificity of user, (2) is it actually a problem or a feature wish, (3) cost of inaction quantified, (4) why now, (5) why us, (6) testable, (7) free of solution language. Score each 1-5 and suggest the smallest rewrite.
{paste}11. 创始人一句话
Turn this into a founder one-liner I can say at a dinner party without sounding like a deck. Voice: warm, specific, slightly understated. Less than 25 words. Avoid "platform", "solution", "ecosystem".
{paste problem statement}12. B2B vs B2C 双版本
Write two parallel versions of the same problem statement: one for a B2B audience (buyer = a manager) and one for a B2C audience (buyer = the end user). Show how the consequence, urgency and decision criteria shift between the two.
Underlying problem: {paste}13. 问题的 pre-mortem
Imagine 12 months from now we built the product and it failed. List the 5 most likely reasons the problem statement itself was wrong — wrong user, wrong frequency, wrong cost, wrong "why now", wrong scope. For each, propose a cheap test to run before building.14. 竞品差异陈述
For this problem, list 3-5 existing solutions users already pay for or hack together. For each: what it does well, where it fails, what segment it leaves underserved. End with one sentence: the underserved gap we will own.
Problem: {paste}15. Discovery 访谈启动 brief
Convert this problem statement into a 1-page discovery interview brief: 3 hypotheses to test, 8 open questions (no leading), 3 things to observe in the user environment, 2 things that would falsify the hypothesis. Format as a checklist.
{paste problem statement}容易踩的坑
- 把方案伪装成问题（“用户需要更好的 dashboard”）。
- 用户定义太宽（“小企业”——先点出 5 个真人）。
- 没有”不做的代价”——如果今天什么都不坏，这个问题就还不该现在做。
- 同一份文档里既写问题又列功能愿望。
- 把问题陈述当一次性产物——每轮 discovery 都应该改写。
- 跳过”为什么是现在”——没有触发因素，这个问题已经被忽视很多年了。
- 让 AI 把单薄问题写得更花，而不是让 AI 去戳它的弱点。
优化技巧
- 让 AI 出陈述之前先喂 3 句真实用户原话——原话胜过你转述。
- 让 AI 标出它做了哪些假设，先去测试那些假设。
- 念给一个不做产品的人听；他复述不出来就重写。
- 强制 40 字以内的标题版——约束暴露含糊。
- 每稿都跑一次模板 10（批评轮），然后改一遍。
- 在 Notion 建一个 /problems 文件夹，每次改写都存一版——长期漂移就是信号。
- 问题陈述要配一个可以推动的数字；如果没有数字，问题太软。
FAQ
- 问题陈述应该多长？: 一句话（25 字内）做 pitch、一段话给团队、一页放 PRD。三者要一致。
- 在用户访谈之前还是之后写？: 先写 v0 暴露假设，访谈 5-8 个之后再改写一遍。前后差异是最有价值的产物。
- 怎么判断一份陈述足够好？: 陌生人读完能说出谁会在乎、什么时刻会痛、哪个指标会动。
- AI 能替我发明问题吗？: 不能。AI 只能锐化、结构化、批评。问题必须来自真实用户，否则你只是在做没人要的功能。
- 多久应该重写一次？: 每季度，或任何一个你以为会动却没动的指标出现时。陈旧的问题陈述是慢性杀手。