Ask A.I. for Twenty Ideas, Not the ‘Best One‘

Ask AI for the best solution to a problem and it will give you one. It will be reasonable, it will be well organized, but it will almost never be the best solution. It’s the most obvious one, dressed up. Ironically, the same is true of people (read on).

I’ve never once been satisfied with an AI’s first ‘best answer’, and I’m now convinced that’s the correct posture. Two techniques fix it, one at the front of the conversation and one at the back.

At the front: stop it from locking on

Here’s the failure that costs people the most and gets noticed the least. You start by thinking out loud, sketching a few possibilities, and the AI quietly treats your sketch as the boundary of the problem. Everything afterward is a variation on your first guess. You’ve hired a consultant who agrees with you.

So when I open with rough ideas, I say this: these are just ideas, please don’t assume they’re the only path forward, and they may not even be the right ideas.

That one sentence keeps it from anchoring. The difference shows up immediately, because it starts proposing directions you didn’t seed.

At the back: make it work harder than it wants to

The second technique is blunter. When the first answer comes back, I push:

“You’re not thinking outside the box. You’re only showing what logic tells you to check.”

Then I ask for two specific things. Think beyond the obvious, to approaches that maybe nobody has tried. And go research what people online have actually done, beyond what you came up with on your own.

The most reliable version of this is a number. It gave you one solution? Ask for the twenty best, ranked worst to best. Pick a number that’s slightly outrageous for the problem. The point isn’t that you need twenty. It’s that twenty forces it past the shortlist it was willing to show you, into research and reasoning it skipped. I’ll do this two or three times on anything that matters.

Why quantity actually produces quality

There’s research behind this, and it’s about people, not machines.

Groups were given the same problem. Some were told to produce the best solution they could. Others were told to produce as many solutions as they could. Experts then judged the results. The groups chasing quantity consistently produced better single solutions than the groups chasing quality.

The reason is iteration. Each attempt teaches you something that shapes the next one, and by attempt eleven you’re combining things you’d never have reached by trying to be brilliant on attempt one.

AI works the same way, for a different underlying reason but with the same practical result. Demanding volume forces it to search, and the search surfaces the thing you were looking for.

The habit under both techniques

Treat the first answer as a draft, always. Don’t anchor it at the start, don’t accept its best at the end, and reward it for range instead of for sounding certain. It’s the same instinct as pushing an AI to find its own errors, which I wrote about here: Make It Find What It Missed. For more on the prompts that produce better answers in the first place, see Five Prompts That Get Better AI Answers.