Large model tools are gradually penetrating into complex tasks such as planning, writing, and analysis. However, "how to use them well" is far more worthy of in-depth discussion than "whether they can be used."
When we use large models (such as ChatGPT), a common misconception is to treat them as "intelligent Q&A machines," expecting only natural language output. But for roles like product managers, engineers, and analysts, this kind of "free-form output" is far from sufficient.
We prefer them to act like structured thinkers, outputting content in specified formats to facilitate generating tables, entering systems, driving front-end displays, or even serving as input for downstream models. This is the value of structured output.
Common Misconception: Expecting "Perfect Answers" While Neglecting Structure Customization
When using large models to solve practical problems, we often fall into two misconceptions:
- Expecting the model to directly provide a "perfect answer."
- Blindly copying the generic structure provided by the model.
However, in real work scenarios, whether it's a proposal, product plan, or speech, truly valuable output must have a structure that fits the scenario and content that aligns with the target audience.
This is precisely where "product manager thinking" can play a key role.
Treat Large Models as "Collaborators," Not "Decision-Makers"
In real scenarios, we can clearly define three roles:
- You (Producer): Understand the business context, target audience, and content intent.
- Large Model (Collaborator): Efficiently assists in generating content based on your input, structure, and materials.
- Audience (Decision-Maker): Ultimately decides whether the content is effective and achieves its purpose.
For example, you need to write a business plan for a bubble tea shop. If you directly use a prompt request, the large model will output content based on a standard template:
💬 Prompt:
"I'm planning to open a bubble tea shop targeting student groups, focusing on fast consumption and multiple flavors. The shop location is at the back gate of a middle school, with no second shop within a 100-meter radius. Can you write a business plan based on my situation?"
The large model will generate a standard business plan based on this prompt. The output is indeed complete, but it's more suitable as written material for investors rather than for storytelling in a startup pitch.
Restructuring: Output Format Must Serve the Purpose of Expression
Structure determines the presentation logic of information and affects the audience's comprehension rhythm.
Depending on the scenario, we should proactively design the structure rather than copy templates. For example, the following "storytelling"-oriented structure is more suitable for oral pitches:
1. Pain Points
2. Solution
3. User Persona
4. Core Competitiveness
5. Market Size
6. Profit Model
7. Competitors
8. Differentiated Advantages
9. Financial Analysis (Profit Cycle)
10. Team Introduction
You can still have the large model generate content, but you must explicitly tell it: Output according to this structure.
Precise Material Provision: If You Don't Specify Clearly, the Model Can Only "Guess"
Even with a reasonable structure, if you don't provide the large model with real, specific, and differentiated information, it can only fill in content out of thin air.
For example, your real advantage is: using directly supplied agricultural products instead of bubble tea powder. You need to actively and clearly inform the model:
We use directly supplied agricultural products as raw materials, offering natural and healthy taste at a cost comparable to commercially available bubble tea powder. This advantage can be reflected in sections like "Competitors," "Differentiated Advantages," and "Pain Points."
This way, the model can accurately express this key information within the structure you specified.
Example: Collaboratively Generating a More Realistic and Effective Plan
After clarifying the structure and materials, the plan you collaboratively complete with the large model will better align with reality and audience needs. For example:
✅ Clear target audience positioning (middle school students)
✅ Prominent competitive advantages (no competitors, differentiated raw materials)
✅ Targeted operational time design (between classes/after school)
The text generated by the model is no longer "vague and general" but "narrates around the expression goals you designed."
Key to Using Large Models Like a Product Manager
When using large models, you are not "handing over the task for it to complete," but rather:
"Using it to achieve the expression goals you designed."
This requires you to possess the following four types of capabilities:
| Capability Type | Manifestation |
|---|---|
| Structure Design Capability | Design structures according to target scenarios, rather than copying templates. |
| Content Provision Capability | Clarify key advantages and points, providing raw materials for the model. |
| Role Clarity Capability | Clearly define who is the leader, collaborator, and audience. |
| Rapid Iteration Capability | Experiment and optimize through repeated prompts. |
Rather than expecting large models to "replace thinking," it's better to make them accelerators for your thinking.
When you start organizing structures, feeding materials, and controlling outputs like a product manager, you'll find that the true value of large models is not automation, but enhancing your expression and decision-making capabilities.
Full conversation example for this article:
https://chatgpt.com/share/67ba10bd-1558-800d-9c12-015cabb6788e