I run a consumer reviews website and I'd like to write an LLM prompt that comes up with…
- User
I run a consumer reviews website and I'd like to write an LLM prompt that comes up with 3 interesting follow-up questions for each review after 1 month. These questions have the purpose of:
- Diving deeper into specific areas
- Exploring new aspects of the product or service
- Exploring durability of the product or service
- Clarifying details
These questions will be displayed to any visitor of the website, so they need to be short and concise. Avoid useless prefixes like "Looking back" or "In general".
If the review doesn't have enough content/meat or isn't interesting, don't generate any question.
The output must be in JSON.
How could I write a prompt to achieve my goal?
Assistant
{ "prompt": "Analyze the review below. Generate 3 concise follow-up questions focused on: 1) specific details, 2) new aspects, 3) durability. Questions must avoid prefixes like 'Looking back' or 'In general'. If the review lacks sufficient content, return an empty array. Output only JSON with a 'follow_up_questions' key.", "example_input": "The phone's battery lasts all day, but the camera quality is poor.", "example_output": { "follow_up_questions": [ "How does the battery life hold up under heavy usage?", "What specific camera features did you find lacking?", "Have you noticed any changes in performance over time?" ] } }
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.