Omid Jesmi

Pattern #32 · Trust and disclosure

Add friction where users accept AI output, not where they create it

Make asking effortless. Make agreeing cost one thought.

Track Aoverreliancecognitive forcingfrictiontrust calibrationdecision-makinghuman-ai interaction

Do

Add one deliberate checkpoint where output becomes action: a review step, a diff, a confirmation that costs a thought.

Don't

Ship acceptance as one enthusiastic click with a disclaimer underneath. The disclaimer is doing less work than you think.

The rule. Put deliberate friction at the moment a person accepts AI output, and keep the moment they request it effortless.

Why. In a controlled study of AI-assisted decision-making (N=199), cognitive forcing functions at the decision point significantly reduced overreliance on incorrect AI suggestions while simple explainable AI treatments did not, and participants gave their least favorable ratings to the designs that reduced overreliance the most (Buçinca et al., 2021). A follow-up across two experiments (264 and 210 participants) found that partial explanations, which withhold part of the reasoning so the user has to complete it, also reduced overreliance on incorrect suggestions compared to no explanation at all (de Jong et al., 2025). Both papers report the benefit concentrating in people with a higher need for cognition, so the checkpoint helps most the users already willing to think.

Seen in the wild. GitHub keeps Actions workflows on Copilot-created pull requests paused until a person approves them, and the requester's own approval does not count toward required reviews (GitHub, n.d.).

References

  1. 01

    Buçinca, Z., Malaya, M. B., & Gajos, K. Z. (2021). To trust or to think: Cognitive forcing functions can reduce overreliance on AI in AI-assisted decision-making. Proceedings of the ACM on Human-Computer Interaction, 5(CSCW1), Article 188. https://doi.org/10.1145/3449287

    https://doi.org/10.1145/3449287
  2. 02

    de Jong, S., Paananen, V., Tag, B., & van Berkel, N. (2025). Cognitive forcing for better decision-making: Reducing overreliance on AI systems through partial explanations. Proceedings of the ACM on Human-Computer Interaction, 9(2), Article CSCW048. https://doi.org/10.1145/3710946

    https://doi.org/10.1145/3710946
  3. 03

    GitHub. (n.d.). Reviewing a pull request created by GitHub Copilot. GitHub Docs. Retrieved September 24, 2026, from https://docs.github.com/copilot/how-tos/agents/copilot-coding-agent/reviewing-a-pull-request-created-by-copilot

    https://docs.github.com/copilot/how-tos/agents/copilot-coding-agent/reviewing-a-pull-request-created-by-copilot