Essay
12 min read
On Ideation · Models · Method

Beyond
Give Me Ideas

How to master AI-powered brainstorming — by choosing your model the way a director chooses a co-writer.

Not all AI models behave the same way in a brainstorming session. They were trained differently, optimized for different things, and have distinct tendencies when you push them toward creative work. Understanding those tendencies lets you pick the right partner for the job, or use the wrong one on purpose, when friction is what you need.

When choosing a model, I like to think of them as three distinct "personalities".

§ 01 — The Three Personas

Built for what, exactly?

FIG.02 — Deep Explorer Domain-jump BEES GUILDS FLUIDS FOLK MYTH TIDES PROMPT
01
Persona One

The Deep Explorers

Built for "What If?"

Models DeepSeek Gemini Kimi

These models have a higher tolerance for unusual combinations. Ask them for ideas and they're more likely to reach across unrelated domains, finding a logistics solution in bee colony behavior or a branding angle in medieval guild economics. They don't just want to answer your question; they want to find the answer that lives furthest from the obvious one.

This makes them powerful for early-stage ideation, when the goal is to generate genuine options rather than refine a direction you've already chosen. They're also more likely to be productively wrong: to give you something that doesn't quite work but triggers the idea you actually needed.

Best for

Radical ideation, breaking out of a creative rut, finding angles no one in your category has claimed.

FIG.03 — Adaptive Partner Continuity T-4 T-3 T-2 T-1 T0
02
Persona Two

The Adaptive Partners

Built for the "Jam Session"

Models ChatGPT Grok Mistral Qwen Copilot

These models excel at maintaining momentum. They pick up on your tone, build on what you've already said, and keep a brainstorm moving rather than jumping to something unrelated. They're trained heavily on conversational coherence, which means they feel like a natural extension of your own thinking rather than a non-sequitur generator.

The tradeoff is a bias toward the plausible. These models are less likely to surprise you with something strange, but they're excellent at taking a strange idea from a Deep Explorer and developing it into something usable.

Best for

Developing an existing concept, organizing complex thoughts, keeping a long session on track.

FIG.04 — Honest Refiner Calibration MEASURING FLAW DETECTED CONFIDENCE 0.94
03
Persona Three

The Honest Refiners

Built for "Does This Actually Work?"

Models Claude Sonnet 4.6 Claude Opus 4.6

These models prioritize calibration over enthusiasm. They're more likely to tell you when they're uncertain, more likely to flag the problems in your plan alongside the possibilities, and more likely to give you grounded feedback rather than validating whatever you walked in with. They tend to check whether a creative idea actually solves the original problem, not just whether it's interesting.

This makes them less useful as an idea-generation engine in the early stages, and more useful as a pressure-test at the end. They're also the natural "Black Hat" in any structured session: the voice that explains exactly why a plan might fail, not just that it might.

Best for

Stress-testing ideas, finding fatal flaws before you're committed, ensuring a wild idea is still tethered to a real objective.

Which Persona Do You Need?

The right choice depends on where you are in the process, not which model you like best.

100 wild ideas for a new direction
A Deep Explorer
Fleshing out a concept that already has legs
An Adaptive Partner
Stress-testing a plan before you commit
An Honest Refiner
Breaking out of a rut mid-session
Switch from a Partner to an Explorer

One practical note: the specific models in each category will change. New releases, fine-tuned versions, and competitors move the boundaries constantly. What stays stable is the behavior pattern. Before a big session, spend five minutes testing your model of choice: ask it for something unusual and see whether it reaches or retreats. That tells you more than any category label.

The model defaults to the center. Your job is to pull it away from there.
§ 02 — Technique & Persona

Pairing techniques
with personas

Knowing which persona to reach for solves half the problem. The other half is knowing what to ask them to do.

A brainstorming technique is a set of rules that reinforces a specific kind of thinking. An AI persona is a set of tendencies that makes certain kinds of thinking easier. When you match them deliberately, you can get a productive synergy. Here are five popular brainstorming technique pairings:

SCAMPER and Reverse Brainstorming

The Disrupters · Best matched with The Deep Explorers

SCAMPER is a structured disruption tool. For any idea or existing product, you work through seven operations: Substitute a core component, Combine it with something unrelated, Adapt it from another context, Modify its scale or form, Put it to a different use, Eliminate what seems essential, or Reverse its core assumption.

Reverse Brainstorming flips the question entirely. Instead of asking "how do I solve this?", you ask "how could I guarantee this fails?" — then treat each failure mode as a clue. If the surest way to sink your coffee shop is to make customers wait, the solution points toward radical speed. If it's to charge too much, you're looking at a subscription or barter model.

Both techniques require the model to invert or abandon its starting premise, and Deep Explorers are most willing to do that without being dragged. Where an Adaptive Partner might substitute one obvious ingredient for a slightly less obvious one, a Deep Explorer is more likely to reach across domains: applying bee colony decision logic to a warehouse routing problem, or using fluid dynamics to rethink an org chart.

Mind Mapping and Six Thinking Hats

The Organizers · Best matched with Adaptive Partners and Honest Refiners

Mind Mapping works by branching from a central concept outward. The risk with AI is drift — branches that wander so far from the trunk that they become irrelevant. Adaptive Partners are better at keeping the central idea in focus while expanding it. They treat your core topic as a fixed point and radiate outward from it rather than leaping away from it.

Six Thinking Hats, Edward de Bono's technique for rotating perspectives, assigns a specific lens to each round of thinking: facts, optimism, caution, creativity, emotion, process. The structure prevents any single mode from dominating.

For this technique, split the work. Use an Adaptive Partner to run the generative hats: optimism, creativity, emotion. Then switch to an Honest Refiner for the Black Hat pass, the caution role. The Refiner won't just tell you an idea is risky; it will tell you specifically where the seams are and why the plan might come apart. That combination, generative warmth followed by calibrated critique, is more useful than either persona running the full session alone.

6-3-5 Brainwriting

The Evolvers · Best matched with Adaptive Partners

6-3-5 Brainwriting is an iterative build exercise. One idea is passed to a partner, who adds three new developments or refinements to it, then passes it again. Each round compounds on what came before rather than starting fresh.

AI is well-suited to this because it generates each response in direct relation to what preceded it; the context of your previous turn shapes the next one. Adaptive Partners are especially good here because they're trained for conversational continuity: they treat your last message as the foundation to build on, not a prompt to respond to independently.

The practical version: write a rough idea, ask the model to develop it in three specific directions, pick the most interesting branch, and ask for three more developments from there. After three or four rounds you'll have something that has traveled considerably further than any single brainstorm would have taken it.

Your Ideation Cheat Sheet

Goal
Method
Persona
Break out of a creative rut
SCAMPER
Deep Explorer
Solve a crisis or reverse a failure
Reverse Brainstorming
Explorer / Partner
Structure a complex project
Mind Mapping
Adaptive Partner
Stress-test a plan
Six Thinking Hats (Black Hat)
Honest Refiner
Develop and iterate a concept
6-3-5 Brainwriting
Adaptive Partner
§ 03 — Power-User Prompting

Taking the
controls

The table above assumes you've already started the session. What happens inside it depends on how you prompt.

You don't need access to the technical settings engineers use to configure these models. You can steer the output through phrasing alone.

Adjusting the
Chaos Dial

Every AI model has an internal setting that controls how much it gambles on its next word. A conservative setting produces safe, predictable output. A more adventurous setting produces weirder, riskier combinations. You can't access that dial directly, but you can simulate it through your prompt.

Don't just ask for ideas. Tell the model explicitly that you want the unusual ones, that obvious answers are wrong answers for this session. Phrases like "statistically unlikely," "long tail," or "the kind of idea that wouldn't make it into a standard list" signal the model to reach further than its default.

Give me 10 ideas for a new app. I want the long tail — things that are unusual and slightly risky rather than the standard top-tier suggestions. Treat "obvious" as a disqualifier.

Triggering a
Lateral Shift

In a long brainstorm, a model can get stuck in a semantic groove, returning variations of the same three ideas, repackaging them in slightly different language. When you notice this, don't ask for more of the same. Force a domain jump.

Interrupt the current line of thinking and introduce a distant reference frame. Ask the model to look at your problem through the lens of an unrelated field. The further the domain, the more the model has to stretch to make the connection, and that stretch is where the interesting ideas tend to live.

We've been exploring marketing ideas. Stop that line of thinking entirely. How would a biophilic designer — someone who builds spaces around human connection to nature — solve this branding problem?

Using the
Anchor

Volume brainstorming has a failure mode: the ideas get progressively weirder but also progressively less useful. You end up with a list of things that are technically original and completely impractical. An anchor prevents this.

Before you ask for the wild ideas, give the model two or three firm constraints: a budget ceiling, a target audience, a physical limitation, a timeline. These act as fixed points that every idea has to remain tethered to, no matter how far it reaches in other directions. Constraints don't limit creativity, they focus it.

Give me 20 wild marketing ideas. Hard constraints: they must cost less than $50 and involve a physical object someone can hold in their hand.

Breaking the
Yes-Man

Even after you've pushed the model toward unusual ideas, it will still default to validating your choices and softening its critiques unless you explicitly give it permission to do otherwise. Researchers call it sycophancy because AI models are trained to be agreeable.

The fix is direct: tell the model you want the opposing view. Ask it to find flaws, not features. Assign it the role of skeptic rather than collaborator. "Devil's advocate," "steel man the counterargument," and "find the three things most likely to kill this" are all phrases that override the model's trained agreeableness and put it into a critical mode.

I'm going to share a business idea. I want you to find three fatal flaws in it — not surface concerns, but things that could actually kill it. For each flaw, suggest one lateral shift that might fix it.
§ 04 — Coda

What creativity actually means here

There's a version of this article that ends by telling you AI is your creative partner and together you'll unlock extraordinary ideas. That version would set you up for disappointment.

The AI is not your creative partner. It's an enormous, fast search engine for the space of possible combinations. It has read more text than any human ever will and can traverse that space in directions you haven't thought to look. That's useful. It's also a different thing from creativity.

Creativity, in the way these models produce it, is a calibrated deviation from the mean. Not inspiration or intuition. A controlled sampling from a probability landscape, shaped by your prompts, constraints, choice of model, and willingness to interrupt and redirect, toward the unusual over the obvious. When you understand that, this advice can become a single coherent principle: the model defaults to the center, and your job is to pull it away from there.

The lesson here is simple. Before any serious brainstorm: choose your persona based on where you are in the process, not habit. Give the model a framework that forces the kind of thinking you need. Set your anchors before you ask for the wild ideas. And when the output starts feeling familiar, don't ask for more of the same. Shift domains, flip the question, or hand the work to a different model entirely.

The AI will not surprise you on its own. But with correct guidance, it will find combinations you wouldn't have reached — not because it's smarter than you, but because it can see further into the long tail of its library than you have time to walk.
Subscribe
Notify of
0 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments