In recent exchanges regarding AI usage, I’ve noticed a common phenomenon: many people (myself included, previously) are still stuck using AI as a "Search Engine Plus."
We are used to throwing it a question and expecting a quick answer. While this works fast for writing outlines or polishing emails, once we touch upon complex business problems requiring deep thinking, the answers often turn into "correct nonsense"—logically sound, but superficial and empty.
The problem isn't that AI isn't smart enough; it's that our interaction mode is too one-dimensional.
If we treat it as a cold, Q&A machine, we will get generic templates. However, if we borrow methodologies from "Qualitative Research" and treat AI as a "Junior Researcher" that needs guidance, the results are vastly different.
Today, I’m sharing two core methods I’ve summarized from practice. The core logic is simple: Don't rush for the answer. Spend time building high-quality "Question Banks" and "Context" first.
Method 1: The Reverse Interview — Extracting Your "Tacit Knowledge"
Often, AI output is unsatisfactory because we assume it knows the background information (it doesn't), or because we haven't fully clarified our own requirements.
Instead of struggling to write thousands of words of background context, use role reversal. Like mentoring an intern, force the AI to become the interviewer, using questions to thoroughly understand your intent.
Universal Prompt Template:
"I want to solve/achieve [XXX Result] (e.g., design a marketing campaign for a specific demographic). Before we start the execution task, please act as a 'Senior User Researcher.' Interview me to fully understand my ideas, background constraints, and deeper intent.
Please strictly adhere to the following interview rules:
Structure then Detail: Ask about the macro direction first, then progress to specific details.
Single-Point Inquiry: Ask only one question at a time. Do not throw a list of questions at me at once.
Thorough Understanding: If my answer is not clear enough, you must follow up until you completely understand my goal.
Quantity Control: Please keep the overall interview within 20 questions."
Through a dozen rounds of Q&A, AI not only helps structure your vague ideas but also gains incredibly rich context. At this point, the precision of the solution it produces will rise exponentially.
Method 2: Brainstorm Coding — Building a "Question Bank" with Programmatic Thinking
When facing a completely new or unfamiliar field, the hardest part isn't finding answers—it's that we don't even know what core questions to ask.
If we rely on guesswork, we easily fall into blind spots. Borrowing the concept of "Grounded Theory" from qualitative research, I’ve summarized a workflow to use AI to build a structured "Question Bank":
Diverge: Give AI a core topic (e.g., "Shortening the reservation chain") and ask it to brainstorm 30 related questions.
Code: A critical step involving human intervention. Review these 30 questions, categorize them into major themes (e.g., Experience, Service, Pricing), and refine a "Main Question" for each category.
Iterate: Feed these highly abstract "Main Questions" back to the AI and ask it to extend them into deeper sub-questions based on these themes.
Cross-Verify: Repeat the above steps across different models (ChatGPT, Gemini, Grok, DeepSeek, etc.) to catch the blind spots of individual models.
Synthesize: Finally, combine the question lists produced by multiple AI models. Remove duplicates, simplify, and refine them into a final, high-quality core question bank for that topic.
Through this process, you are no longer relying on luck with a single question, but possess a systematic mind map covering the entire issue.
AI is not just an efficiency tool; it is a cognitive amplifier.
When we start training it and building context with the mindset of a "Researcher," we are actually upgrading our own way of thinking. I hope these two methods help you pull yourself out of busy work and use AI for deeper business contemplation.
Feel free to share the Prompts you use to solve complex problems—let’s tune a more powerful work partner together!
最近在和大家交流 AI 使用心得时，我发现一个普遍现象：很多人（包括之前的我）对 AI 的使用依然停留在“搜索引擎 Plus”阶段。
我们习惯扔给它一个问题，期待一个快速答案。用来写大纲、润色邮件确实很快，但一旦涉及深度思考的复杂业务问题，它的回答往往变成了“正确的废话”——逻辑通顺，但浮于表面。
问题不在于 AI 不够聪明，而在于我们与它的交互模式太单一。
如果我们只把它当成无感情的问答机器，得到的就是通用模板。但如果借鉴“质性研究”的方法论，把它当做一个需要引导的“初级研究员”，效果将天差地别。
今天分享两个我实践总结的核心方法，核心逻辑很简单：不要急着要答案，先花时间构建高质量的“问题库”和“上下文”。
方法一：反向采访 —— 榨干你的“隐性知识”
很多时候 AI 产出不理想，是因为我们默认它知道背景信息（其实它不知道），或者我们自己都没想清楚需求。
与其费劲写几千字的背景提示词，不如使用角色转换的方法。像带实习生一样，强制 让AI 变为采访者，通过提问来彻底理解你的意图。
通用 Prompt 模版：
“我想要解决/达成「XXX结果」（例如：设计一个针对特定群体的营销活动方案），请在正式开始执行任务之前，先担任「资深用户研究员」的角色，通过采访我的方式来充分了解我的想法、背景约束和深层意图。
采访请严格遵守以下规则：
先结构后细节： 先问宏观方向，再递进到具体细节。
单点追问： 每次只问一个问题，不要一次性抛出一堆。
彻底理解： 当我的回答不够清晰时，请务必追问，直到你完全理解我的目标。
数量控制： 整体采访请控制在 20 个问题以内。”
通过十几轮的你问我答，AI 不仅帮你结构化了模糊的想法，更获得了极其丰富的上下文。此时它产出的方案，精准度会呈指数级上升。
方法二：脑暴编码 —— 用程序化思维构建“问题库”
面对全新的陌生领域，最难的不是找答案，而是我们根本不知道该问什么核心问题。
此时如果只靠拍脑袋，容易陷入盲区。我借鉴质性研究中“扎根理论”的思路，总结了一套流程，利用 AI 帮我建立结构化的“问题库”：
发散： 给 AI 一个核心议题（如“缩短预定链路”），和它一起头脑风暴出 30 个相关问题。
编码： 关键步骤，由人类介入。审视这 30 个问题，将其归纳为几大类（如体验、服务、定价），并为每类提炼一个“母问题”。
迭代： 把高度抽象的“母问题”喂回给 AI，要求它基于此延展出更有深度的子问题。
交叉验证： 在不同模型（ChatGPT, Gemini, Grok, DeepSeek 等）重复上述步骤，获取不同模型的结果。
综合提炼： 最后，由你来综合多个 AI 模型产出的问题列表，去重、精简，提炼出最终针对该议题的高质量核心问题库。
通过这套流程，你将不是再拿着单一问题碰运气，而是拥有一张覆盖该议题的系统性思维导图。
AI 不仅仅是效率工具，更是一个认知扩充器。
当我们开始用“研究员”的思维去训练它、构建上下文时，我们实际上是在升级自己的思考方式。希望这两个方法能帮你从繁杂工作中抽离，利用 AI 进行更深层次的业务思考。
欢迎分享你解决复杂问题的 Prompt，我们一起调教出更强大的工作伙伴！