Yes-and improv prompt builder
Brainstorming fails when generation and judgement happen at once — ideas die before they grow. Improv solves this with yes-and: accept every offer and build on it, never block. This builder generates a prompt that applies yes-and to an LLM, forcing it to accept and extend each previous idea round after round, so concepts compound into something you would not have reached by listing ideas in isolation.
How it works
You provide a starting idea, the domain it lives in, a round count, and a build depth. The generated prompt instructs the model to run numbered rounds, each beginning with “Yes, and…”, each building directly on the cumulative idea from prior rounds, with no criticism or feasibility caveats allowed during the chain. Build depth controls how far each round leaps — adding one angle, combining ideas, or pushing to an extreme. After the rounds, the model writes a “where this landed” summary and only then notes practical considerations. The prompt is built in your browser.
Why the no-criticism rule matters
Most LLM brainstorming sessions degrade into hedged idea lists: the model generates something reasonable, qualifies it, suggests an alternative, and the chain never builds. The yes-and constraint breaks that pattern by making acceptance mandatory. The model cannot say “that might be challenging because…” — it must say “Yes, and to make that work we also add…” The result is a compounding thread rather than a flat list, and the final combined concept often contains elements that would never have appeared had each idea been evaluated on its own.
The “where this landed” summary serves as a decompression valve: after enforcing divergence for the whole chain, it collapses the session into one coherent concept the model can then assess practically.
Worked example
Seed idea: a subscription box for people learning a new language, domain: consumer products.
Round 1 (shallow): “Yes, and each box includes a physical phrase card set matched to the learner’s current level.” Round 2: “Yes, and the phrase cards connect to a companion app via NFC so scanning one plays the audio pronunciation.” Round 3: “Yes, and the app tracks which cards have been scanned and adapts next month’s box to prioritise words the learner struggles with.” Round 5 (deep): “Yes, and the whole system eventually generates a personalised language-learning memoir — a printed story using only the vocabulary the learner has mastered, delivered as a bonus at the one-year mark.”
At round 5 there is an idea — a personalised physical memoir — that no brainstorm starting from “what should go in a language-learning box?” would have reached without the compounding chain.
Tips and notes
- Separate generate from judge. The no-criticism rule is the point — let ideas compound through all the rounds, then prune in a second pass with a critical prompt.
- Tune build depth to your appetite. Deep settings produce wilder concepts that need more filtering; shallow stays close enough to the seed to stay practical.
- Five to eight rounds is the useful range. Enough to compound, not so many it drifts into concepts with no connection to the original goal.
- Follow with a critical prompt. The yes-and output is divergent raw material. After reading the “where this landed” summary, switch to a feasibility or prioritisation prompt before acting on anything.