Ask AI for “some good ideas” and you may get a familiar list that could fit almost anyone. A more useful brainstorm starts with a real situation: who the idea is for, what you want to achieve and which limits matter. Those details give the assistant something to work with.
Google lists brainstorming and planning among the uses of Gemini Apps. The process below is an original practice exercise: it helps you request different kinds of ideas, compare them and choose one small thing to try. It does not establish that an idea is original or guaranteed to work.
Source: Google’s overview of Gemini Apps
What makes a useful brainstorming brief?
Imagine you are organising a welcoming activity for adults attending their first AI workshop. “Suggest icebreakers” gives little direction. “Suggest ten-minute activities for eight adults who have never used AI, with no phones or accounts required” gives a much clearer starting point.
- Audience: who will take part, and what do they already know?
- Outcome: what should feel easier afterwards?
- Limits: how much time, money, space or equipment is available?
- Preferences: what would suit your group, and what should be avoided?
Ask for different directions, not just a longer list
A list grouped by approach is easier to explore than twenty variations of the same activity. If every suggestion still involves discussing a question, ask for a different mechanism: sorting cards, drawing a simple example or making a choice between two fictional scenarios.
What if all the ideas still sound ordinary?
Give specific feedback. “The paired discussion works, but it assumes people know each other. Adapt it for strangers.” Or ask for a deliberate contrast: “Give me one version for a very quiet group and one for people who already enjoy talking.” You are providing judgement that the original brief did not contain.
You can also bring your own imperfect idea. Describe what you like and where it breaks down. Request three changes that preserve its strongest feature. This often makes the conversation more useful than asking the assistant to invent everything from scratch.
How do I choose an idea?
Choose criteria before asking for a favourite. For the workshop, those might be ease of explanation, comfort for nervous participants and preparation time. Ask the AI to describe trade-offs, then make your own choice. A confident ranking is not evidence that real participants will enjoy the activity.
Treat factual assumptions as separate questions. If an idea depends on a venue, price, permission or accessibility feature, check that detail directly. Generated names, slogans and concepts may resemble existing work, so do not describe them as unique simply because AI suggested them.
Test one small version
Pick one idea and rehearse the explanation with a friend. Can you explain it in thirty seconds? Are the materials easy to prepare? Does it meet the original limits? Use what you learn to adjust the activity. A small test gives you more useful evidence than another page of enthusiastic suggestions.

