Omid Jesmi

Pattern #3 · Expectation setting

Set capability expectations before the first prompt

Expectations form before the first output. Unmanaged, they default to magic.

Track Aexpectation settingonboardingcalibrated trust

Do

Put the capability statement at the point of first use: what the system is for, where it fails, and how often. An introductory blurb and a plainly stated error rate beat a feature tour.

Don't

Let an empty chat box do the onboarding. A blinking cursor promises everything and commits to nothing.

The rule. State what the system can do, and how well, before the first prompt: the scope, the strengths, and the mistakes it is likely to make.

Why. Of the 18 guidelines for human-AI interaction, the pair applied at initial interaction is "make clear what the system can do" and "make clear how well the system can do what it can do," because unclear expectations about scope and performance lead to disappointment, abandonment, and in higher-stakes settings real harm (Amershi et al., 2019; Microsoft, n.d.-b, n.d.-c). The effect is measurable: Kocielnik et al. (2019) held an AI scheduling assistant at 50 percent accuracy and found that adjusting expectations before use prepared people for the system's imperfections and significantly increased acceptance. Same accuracy, different framing, different verdict.

Seen in the wild. Gemini's welcome screen carries "Gemini can make mistakes, so double-check it" before the user types anything; Microsoft's HAX design library files it as a live example of guideline 2 (Microsoft, n.d.-a).

References

  1. 01

    Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., Nushi, B., Collisson, P., Suh, J., Iqbal, S., Bennett, P. N., Inkpen, K., Teevan, J., Kikin-Gil, R., & Horvitz, E. (2019). Guidelines for human-AI interaction. Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, 1-13. https://doi.org/10.1145/3290605.3300233

    https://doi.org/10.1145/3290605.3300233
  2. 02

    Kocielnik, R., Amershi, S., & Bennett, P. N. (2019). Will you accept an imperfect AI? Exploring designs for adjusting end-user expectations of AI systems. Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, 1-14. https://doi.org/10.1145/3290605.3300641

    https://doi.org/10.1145/3290605.3300641
  3. 03

    Microsoft. (n.d.-a). Google Gemini | G2: Make clear how well the system can do what it can do. Microsoft HAX Toolkit. Retrieved July 16, 2026, from https://www.microsoft.com/en-us/haxtoolkit/example/google-gemini-g2-make-clear-how-well-the-system-can-do-what-it-can-do/

    https://www.microsoft.com/en-us/haxtoolkit/example/google-gemini-g2-make-clear-how-well-the-system-can-do-what-it-can-do/
  4. 04

    Microsoft. (n.d.-b). Make clear how well the system can do what it can do (Guideline 2). Microsoft HAX Toolkit. Retrieved July 16, 2026, from https://www.microsoft.com/en-us/haxtoolkit/guideline/make-clear-how-well-the-system-can-do-what-it-can-do/

    https://www.microsoft.com/en-us/haxtoolkit/guideline/make-clear-how-well-the-system-can-do-what-it-can-do/
  5. 05

    Microsoft. (n.d.-c). Make clear what the system can do (Guideline 1). Microsoft HAX Toolkit. Retrieved July 16, 2026, from https://www.microsoft.com/en-us/haxtoolkit/guideline/make-clear-what-the-system-can-do/

    https://www.microsoft.com/en-us/haxtoolkit/guideline/make-clear-what-the-system-can-do/