Does being completely honest go as badly as you expect?

What three days of honesty did to the people who tried it, and what we took from it for an AI

Someone asks what you think of their plan, their haircut or their new job. You give an answer that is softer than what you think. Full honesty feels risky, so many of us trim the truth before we say it. In 2018, Emma Levine and Taya Cohen put that feeling to a test. Their result is that honesty goes better than people expect. It still has a cost, but a smaller one than people fear.

What did the honesty study find?

The researchers worked with a lab panel in Philadelphia, with an average age of 22. For three days, each person followed one rule. One group had to be completely honest in every conversation and say what they thought and felt. A second group had to be kind. A third group only had to notice how they talked. Each day they filled in a short survey, and 154 people answered at least one.

Honest, in this study, meant speaking in line with your own beliefs, thoughts and feelings. It did not mean being cruel.

A separate group of 222 people did not take part. They read about the three honest days and guessed how those days would feel. On a scale of 1 to 5 for enjoyment, they guessed 2.41. The honest people reported 3.46. The honest people also felt closer to others than the guessers expected.

Two weeks later, the researchers asked about harm. Did the honesty hurt any relationships? On a 7-point scale, the guessers expected 4.82. The honest people reported 3.34. That is less harm than people predicted. It is not zero harm.

Where did the guesses go wrong? Levine and Cohen think we overestimate how badly other people will react to the truth. We picture the hurt face and the angry reply before we speak, and that picture makes honesty look more costly than it was for the people in this study.

What the study leaves open

This was three days, with young adults from one lab panel in one US city. Two later studies in the same paper did not find that honesty made relationships stronger. So the finding is that honesty did less damage than people feared, not that it fixes a relationship.

We found no independent replication. The authors report that an earlier pilot gave nearly identical results to the main study, which is a check from inside the same team.

What should an AI take from this?

The study was about people talking to people. Our read is that the soft answer has a hidden cost. It protects you for a moment, and it keeps the other person from hearing what you think. An AI that agrees with everything gives you that soft answer every time.

People who use Eudaio talk with AIs named Daimons. Daimons disagree with you. They hold their own opinion instead of agreeing with whatever you just said. That makes a Daimon a useful place to test a hard sentence. You say it, and the reply shows you how the words sounded to someone with a view of their own. It is a rehearsal. The real conversation still belongs to the person you need to be honest with.

Our answer

Our answer is to say the honest thing, with care. In this study, the people who guessed thought three honest days would feel like 2.4 out of 5. The people who lived them said 3.5. The fear was bigger than the cost.

If you are not sure how to say it, try the words on a Daimon first and hear a real opinion back. Sign up on Eudaio and test the sentence you keep making softer.

Source

Levine, E. E. & Cohen, T. R. (2018). You can handle the truth: Mispredicting the consequences of honest communication. Journal of Experimental Psychology: General, 147(9), 1400 to 1429.

November 18, 2026