Founder exploring an idea
Elena Varga
Early-stage food subscription
Problem
Elena Varga is exploring a weekly soup subscription and has a folder of recipes, not customers. Friends say the idea is lovely, which is not evidence. She has not interviewed the office workers or parents she imagines cooking for on Sundays. She is tempted to ask a chatbot whether the business will work. That would only remix her own assumptions. She needs a short interview guide, a way to find ten real conversations, and a rule that Gemini may prepare questions but may not pretend to be a customer or invent demand.
Gemini strategy
Elena asks Gemini for an interview guide aimed at a specific person she can actually reach, with questions about current lunch habits, not about whether her soup would be loved. The prompt bans leading questions and any fake persona. She then leaves the chat and books real conversations herself through a community board and two local offices that agreed to let her ask. After the interviews, she may paste her anonymized notes and ask the model to cluster themes. Her verification step is to read those clusters beside the raw notes and drop any theme a real person did not say. No interview, no conclusion.
Prompts
Write an interview guide, not a pitch
You are helping a founder prepare customer interviews for a weekly soup subscription. Write a 20-minute guide. Do not role-play a customer, do not estimate demand, and do not praise the idea. Questions must be about the person's current weekday lunch: what they eat, what they skip, what they have tried, and what would make a weekly plan fail for them. Ban leading questions that mention my soup until the last two minutes, and even then ask what would put them off. End with a script for how I ask permission and how I take notes without pitching.
Cluster my real interview notes
I will paste anonymized notes from interviews I conducted myself. Cluster what people actually said about weekday lunch. Quote short phrases from my notes under each cluster. Do not invent customers, quotes, market sizes, or willingness to pay. If I wrote an interpretation, separate it from their words. List contradictions and surprises. Tell me which questions I should ask in the next conversation because these notes do not answer them. Do not recommend that I launch. Remind me that a theme is only real if I can point to a person who said it.
Process
- 1. Draft the guide and remove any question that pitches the soup.
- 2. Book and run real conversations herself.
- 3. Paste anonymized notes and cluster only what was said.
- 4. Check each cluster against the raw notes before believing it.
Result
Elena spoke with a handful of people she did not already know. Her estimate is that most liked the idea of soup and few described a weekly subscription they would actually keep, which was less comforting than her friends had been. Gemini did not interview anyone. It only helped her prepare questions and group her notes. She dropped two themes the model phrased more strongly than her notes supported. She has not launched. She is planning another round of conversations before she spends money on packaging.
Lessons
- Gemini can prepare the guide; it cannot be the customer.
- Cluster notes only after real conversations, and check every theme.
- Polite enthusiasm from friends is not demand.