- In a 2025 Nature Human Behaviour trial, about 820 people debated either a human or GPT-4. When the AI was given a few basic facts about the person, it won far more often: 81.7% higher odds of shifting their view than a human opponent. Strip the personal details and that edge disappeared. Personalisation, not eloquence, is what moved people.
- This is a discernment question before it is a technology one. The capability that protects you is your own independent judgement.
- The effect ran on personal data. The more a system knows about you, the more precisely it can tune a case to you. What a tool knows about you now matters as much as what it can write.
- Executives are the target, not the exception. Board papers, vendor pitches and staff messages are increasingly AI-shaped and tuned to the reader.
- Discernment is a capability to build: knowing why you believe what you believe, so a fluent case cannot move you off it without good reason.
You read the proposal twice. It was persuasive, unusually so. Every objection you might have raised was already met, in the order you would have raised it. The tone was pitched exactly to you. And a beat after you finished, a small thing nagged. You were convinced, but you were not sure you agreed. Those are not the same feeling, and the gap between them is important.
Something has shifted in the last year. The most persuasive thing crossing your desk is no longer written by the best writer in the room. It is increasingly shaped by a system that knows a little about you, and uses it. That is not a worry about robots. It is a question about your own judgement, and how you keep it clear when the case in front of you was built to move you.
Can an AI really change your mind?
The honest answer, now backed by evidence, is yes, under the right conditions. Researchers ran a clean test and published it in Nature Human Behaviour, a peer-reviewed journal, in 2025. About 820 people held short, structured debates. Their opponent was either another person or GPT-4, a leading AI model. In some debates, one side was given a few basic facts about the other: age, gender, rough political leaning, that kind of thing.
The result is the part to sit with. When the AI had those personal details, people came away with 81.7% higher odds of moving toward its position than when they debated a human. Read the scope carefully. This was an online debate format with roughly 820 participants, not a survey of the whole population. Even so, the signal is clear and it held up statistically. And here is the twist that matters most: the machine did not win by out-writing the human. Without the personal details, its edge over people was not significant. Personalisation did the work, not eloquence.
What does that mean for the person being persuaded?
It means the ground has tilted, quietly. A generic argument is easy to weigh. A case tuned to you is harder, because it meets your specific doubts and speaks in your register. Now imagine that at scale. The vendor deck, the internal business case, the change-management email, the analyst note: more and more of them are AI-shaped, and the good ones are tuned to their reader. You are not reading arguments any more. You are reading arguments about you.
This is why it lands on leaders first. You decide under a firehose of persuasion, most of it now machine-assisted. The risk is not that you get fooled by something crude. It is that you get moved, smoothly and often, by cases that fit you a little too well, and mistake being convinced for having judged. It is one more reason trust in AI is such a live question right now.
The most persuasive thing in your inbox is no longer the best writer. It is the system that knows the most about you. Judgement is the new literacy.
| What the persuasion trial found | Figure |
|---|---|
| People in the trial (Salvi et al., Nature Human Behaviour, 2025) | ~820 |
| GPT-4 with basic personal details vs a human opponent | 81.7% higher odds of shifting the view |
| GPT-4 without personal details vs a human | no significant edge |
| What drove the effect | personalisation, not eloquence |
So how do you keep your own judgement?
You stop asking the small question and start asking the big one. The small questions are the ones a reader arrives with: is this tool trustworthy, which vendor is safest, is the writing any good. The better question is quieter and it points inward. Do I know why I believe what I believe, well enough that a fluent case cannot move me off it without a real reason? That is discernment, and it is trainable. Build it in this order:
- Know your own view before you defend it. Write down what you think and why, in a line, before you read the persuasive case. You cannot be moved off a position you never located.
- Separate how well it argues from whether it is true. Fluency is not evidence. Ask what would have to be real for the claim to hold, then check that, not the polish.
- Mind what a system knows about you. The more personal the data, the more tuned the pitch. Treat what you reveal as part of the negotiation.
- Slow the high-stakes yes. A day between a convincing case and a committing decision is the cheapest insurance you can buy.
- Build the team's discernment, not just its access. A team that can weigh a tuned argument is worth more than one with the newest tools and no defence against them.
Want judgement that holds under pressure?
The Strategy Session works on the human side of AI: the discernment your decisions run on when every case in front of you is fluent, fast and tuned to you. We build the judgement first, then point it at the technology.
Book your Strategy SessionWhere does this leave you?
In a stronger place than the headlines suggest, if you choose it. Persuasion technology will keep improving, and there is no putting it back. But the counter to a case built to move you is not a better filter or a new tool. It is a clear inner read: knowing what you value, why you hold your positions, and what it would actually take to change them. That is one of the questions a board should be asking about how it decides at all.
Picture the next big call. The proposal is flawless and tuned to you, as they all will be soon. You take it in fully, and you still decide from your own clear ground, moved only by what is sound. That is not resistance to persuasion. It is the judgement that sits at the centre of leading well, and it becomes the rarest and most valuable thing you bring to the room.
Frequently asked questions
Can AI change your mind?
Is AI more persuasive than humans?
How do you resist AI persuasion?
About the author
British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.