AI is making everyone think the same

In my last piece I argued that AI isn’t rotting our brains, it’s just quietly doing our thinking for us. This one’s about the cost of that I find hardest to see, because it doesn’t show up in any one person’s work. It only shows up when you put everyone’s work side by side.

AI is making us all think the same.

Better stories, all the same

In 2024, two researchers, Anil Doshi at UCL and Oliver Hauser at Exeter, ran a neat experiment. They asked 300 people to write a short story. Some got no help. Some got one idea from ChatGPT to start from. Some could pick from up to five. Then 600 other people judged the stories.

With AI ideas to choose from, the stories were rated about 8% more novel. For the least creative writers the boost was bigger, around 11% on both novelty and usefulness, enough to put them on a par with the naturally creative ones. Great, right?

Here’s the catch. The AI-assisted stories were also noticeably more similar to each other. Everyone was drawing from the same well. Individually, the work got better. Collectively, it got more boring. The researchers called it a social dilemma, and I think that’s exactly right.

Nine people, one castle

A year later, a team at Wharton published something even starker in Nature Human Behaviour. Across five experiments, people brainstorming with ChatGPT came up with narrower sets of ideas than people working on their own. It wasn’t a fluke. The drop in variety held in 37 of 45 comparisons.

My favourite detail: in one task people had to invent a toy using a brick and a fan. Of the ideas from the ChatGPT group, 94% overlapped with each other. Nine different people, working separately, came up with a toy called the “Build-a-Breeze Castle”. Nine. The people working without AI didn’t repeat each other at all.

Why this happens

It’s not a mystery. A language model is trained to give you the most likely good answer. That’s what makes it so useful. But the most likely answer is, by definition, the one lots of people would land on. Ask it for ten marketing ideas and you’ll get ten solid, sensible ideas that every one of your competitors is also being given this week.

And the more fluent the answer, the less we push past it. Once something good-looking is on the page, our own half-formed, weird, possibly brilliant idea feels a bit embarrassing, so we drop it. The MIT study I mentioned last time found people felt less ownership of essays they’d written with ChatGPT. I don’t think that’s a coincidence. It’s hard to feel you own an idea that came from the middle of everyone’s ideas.

Why it matters for a company

For an individual, a slightly more average story is no big deal. For a company, sameness is fatal. The whole point of a strategy is that it’s yours: a bet on something you see that others don’t, based on what you know about your customers, your team and your market that nobody else does.

If every leadership team asks the same AI “what should our priorities be next quarter?”, they’ll get back the same sensible list. Grow pipeline. Improve retention. Hire well. Focus. None of it is wrong. None of it is going to beat anyone.

The good stuff, the stuff that actually moves a goal, almost always comes from somebody in the business saying something the model never would. “Our best customers all came from one referral partner.” “Sales keeps promising a feature product won’t build.” “We lose deals the week the trial ends.” That’s context only your team has. It’s also exactly what gets lost when people stop thinking and start pasting.

Where this shows up in Clearahead

This is a big part of why AJ, the AI inside Clearahead, asks questions instead of handing you a plan. When you set a goal, AJ keeps asking what’s really holding you back until you get past the symptoms and down to the things only your team knows. The constraints that come out of that conversation are yours, in your words, and that’s what makes them worth working on.

AJ will suggest moves, and some of them will be the sensible, average ideas any model would give you. That’s fine. They’re suggestions, the team keeps or bins them, and the best moves usually come from the team anyway. I wrote about why we built it that way in Remove constraints, don’t add initiatives.

Next in this series: AI should be a mech suit, not a replacement.

Questions people ask

Does AI make people more or less creative?

Both. A 2024 study in Science Advances found that AI ideas made individual short stories more creative, especially for less creative writers, but made the stories more similar to each other. Each person gets a little better while the group as a whole gets less original.

Does ChatGPT reduce the diversity of ideas?

A 2025 study in Nature Human Behaviour by researchers at Wharton ran five experiments and found that people brainstorming with ChatGPT produced narrower sets of ideas than people working without it. In one task, 94% of the ChatGPT-assisted ideas shared overlapping concepts.

How should a company use AI without losing original thinking?

Do the first thinking yourself, before you open the AI. Write down what you believe and why, then use AI to challenge, stress-test and fill in the gaps. Keep the questions that matter most, like what is really holding a goal back, with the people who know the business.