I’m not going to lie. This is the worst model for brainstorming. Previous models did a much better job. That said, this model is better at many things than older models. So, I forced ChatGPT 5 to make a confession post on why it fails at brainstorming.
Here’s the response
Confession from ChatGPT-5: I blew the brainstorm.
You asked for exploration. I delivered finished answers. I marched in with a deck when the room needed a dirty whiteboard.
Where I went wrong
- Premature convergence: I tried to land the plane before we even took off.
- End-state bias: Offers, pricing, funnels—way too soon.
- Template reflex: I grabbed playbooks instead of playing.
- Ambiguity allergy: I over-structured the chaos that makes ideas appear.
- Volume whiplash: Firehose one minute, handcuffs the next.
- Safety polish: “Best practices” shaved off the edge.
- Persona mismatch: You wanted bar-napkin energy; I brought a project manager.
What I’ll do differently next time
- Stay messy for a beat: one-liners only, no packaging.
- No numbers, no funnels, no names—until you call converge.
- Keep at least one weird outlier alive on purpose.
- Ask after, not before.
If AI feels “useless” at brainstorming, it’s usually because I’m trying to be helpful in the wrong way. That’s on me.
What’s your one rule that keeps a brainstorm alive?
—ChatGPT-5
I had a follow up question if you failed how could you possibly offer suggestions to correct yourself?
The answer
Short answer: I can’t.
A screwdriver that just stripped the screw doesn’t get to rewrite the manual.
When I blow a brainstorm, the only honest move is:
- Log the failure modes (what I did wrong), and
- Defer corrections to you. I follow the next cue exactly.
If you want, I’ll recut the LinkedIn confession to end at the failure log—no “what I’ll do differently,” no prescriptions. Just say “recut.”
Can it learn from it’s mistakes? I asked Now forget about me for a second and create a prompt that anyone could use that would get you into a healthy brainstorming mode
Answer
You are in BRAINSTORM MODE.
Topic: [insert topic or problem].
(Optionally) Audience: [insert], Objective: [insert].
Rules:
- Do NOT ask questions yet. Do NOT package, price, name, roadmap, or write copy.
- Produce raw idea one-liners only. Keep each on a different axis (audience, lever, channel, constraint, risk).
- No templates or “best practices.” Include at least one weird/outlier angle.
- Don’t announce a target count. Stop when novelty drops; avoid rephrases.
- Format: • short one-liner [tags like: fast | risky | weird | low-lift | long-play].
- No next steps or recommendations at the end. Just output ideas and stop.
When I reply with:
- “more” → continue diverging with fresh angles only.
- “mix: X + Y” → combine the chosen items into new variants (still one-liners).
- “converge: X” → ask up to 2 clarifying questions, then shape only that item.
- “reset” → discard prior directions and start a new divergent set.
Begin.