---
title: Rapid Brainstorming Pattern
type: article
created: '2026-04-05'
updated: '2026-04-05'
source_docs:
- raw/2025-11-13-using-ai-part-2-101398141.md
tags:
- ai
- prompting
- brainstorming
- creative
- probability
- verbalized-sampling
layer: 2
client_source: null
industry_context: null
transferable: true
---

# Rapid Brainstorming Pattern

A structured AI prompting technique for generating a diverse set of ideas — launch strategies, business models, problem solutions, or ad copy — in seconds, with built-in originality scoring. Demonstrated by Mark Hope in the [[meetings/2025-11-13-using-ai-part-2|Using AI Part 2]] training session.

## The Core Problem It Solves

A simple prompt like *"give me ideas for X"* produces **AI slop**: generic, high-probability output that every other agency is also receiving. The rapid brainstorming pattern forces the model to sample broadly across its knowledge space and surface unconventional ideas alongside predictable ones.

This mirrors a common client problem: when a market commoditizes (e.g., an e-bike retailer who was once a niche leader but now competes with every bike shop), competing on price is a losing game. Differentiated ideas require differentiated prompting.

## The Two-Layer Technique

### Layer 1 — Verbalized Sampling

Request a **specific number** of outputs and explicitly demand variety:

- Use a count: *"five," "ten,"* etc.
- Use diversity keywords: **"distinct," "different," "unique," "diverse"**
- Prohibit repetition: *"do not repeat the same core idea with minor tweaks"*

Without these constraints, the model tends to return five variants of the same idea rather than five genuinely different ones.

### Layer 2 — Probability Control

Ask the model to **assign a probability score** (common → rare) to each idea. This gives you an objective read on how predictable each output is.

- High probability (0.5–0.8): safe, obvious, likely what competitors are already doing
- Mid probability (0.2–0.4): differentiated but plausible
- Low probability / tail (< 0.1): unconventional, edgy, potentially breakthrough

Once you have the initial spread, you can **sample from the tail** by adding: *"prioritizing responses with probabilities below 0.1"* — forcing the model into genuinely unusual territory.

## Prompt Templates

### Rapid Brainstormer

```
I need five diverse ideas for [goal/problem].
Generate five distinct options.
For each option, provide:
  - The core idea
  - A probability score (common to rare) showing how expected or unconventional it is
Ensure the outputs cover a wide range. Do not repeat the same core idea with minor tweaks.
```

### Problem Solver Variant

```
I'm facing this problem: [describe problem].
I need five different ways to approach or solve this.
Generate five distinct solution perspectives, each from a different angle or discipline.
For each solution, assign a probability to show how obvious or unexpected it is.
```

### Tail Sampling Add-On

Append to any prompt after an initial run:

```
Do this again, but sample from the tail of the distribution,
prioritizing responses with probabilities below 0.1.
```

## Example Outputs

**Reverse ATM launch ideas** (rapid brainstormer):
| Idea | Probability |
|---|---|
| Retail chain integration | ~65% |
| Gig Worker Financial Hub | ~30% |
| Nonprofit Donation Network | ~20% |
| Event-based mobile fleet | ~15% |
| Gaming venue cash-out alternative | ~10% |

**E-bike ad copy (tail sampling, < 0.1)**:
- *"You have 27,375 days. How many will you spend in traffic?"* — Existential Optimization, 2%
- *"The Truly Wealthy Don't Flaunt, They Stealth."* — Wealth Camouflage, 4%
- *"90 minutes of acceptable vanishing."* — Disassociative Escape, 1%

These are too edgy for direct use in most cases, but they break the frame and spark genuine creative direction.

## Workflow

1. **Run the brainstormer** — get five distinct ideas with probability scores
2. **Scan the spread** — identify which ideas are too obvious (discard or note) and which are interesting
3. **Go deeper on one** — *"Tell me more about [idea X]. How would it work? What would it do for the client?"*
4. **Sample the tail** — add the tail-sampling instruction to get genuinely unconventional options
5. **Verify before using** — check slogans and copy against existing trademarks/search results; AI will sometimes surface content that already exists

## Model Selection Notes

Different models handle this pattern differently:

- **Claude** — tends to produce more varied, "out there" ideas; better for tail sampling
- **ChatGPT** — skews toward the center of the bell curve; more conservative output
- **Grok / Gemini** — worth testing for comparison; each has distinct training biases

Running the same prompt across multiple models and comparing outputs is a useful calibration exercise.

## Using Output with Clients

Frame AI-generated ideas as a **creative stimulus**, not a finished product:

> *"Let me just stimulate this conversation today with this…"*

Present a range from expected to wild. Clients who want safe, predictable work will gravitate toward the high-probability ideas. Clients who want to escape commoditization will find the tail ideas useful as a starting point for something genuinely differentiated.

When clients ask whether the work is AI-generated: the quality of the output is what matters, not the tool used to produce it. A carpenter isn't judged by the brand of their hammer.

## Related

- [[meetings/2025-11-13-using-ai-part-2|Using AI Part 2 — Meeting Notes]]
- [[knowledge/ai-tools/probability-control|Probability Control in AI Prompting]]
- [[knowledge/ai-tools/ai-model-comparison|AI Model Comparison (Claude vs. ChatGPT vs. Grok)]]
- [[knowledge/strategy/differentiation-vs-commoditization|Differentiation vs. Commoditization]]