Set appropriate expectations

Align messaging with real capabilities to prevent overtrust or misuse.

Messaging that overstates what the AI can do leads to overtrust and misuse; messaging that is vague or absent leaves users guessing. Setting appropriate expectations aligns what users believe with what the system actually delivers.

When I design AI-powered experiences, I set appropriate expectations by clarifying scope and capabilities, communicating accuracy and reliability, explaining the role of the AI vs the user, reinforcing expectations over time, and avoiding overpromising in language and branding.

Setting appropriate expectations means:

  • Clarify scope and capabilities: What the AI can and can't do, and for whom or in what context
  • Communicate accuracy and reliability: How often it's right, when to double-check, and what might go wrong
  • Explain the role of the AI vs the user: Who does what—suggestion vs decision, draft vs final
  • Set expectations over time: Reinforce at key moments, not only at first use
  • Avoid overpromising: No "always accurate" or "fully automated" when the system has limits

By setting appropriate expectations, I help users:

  • use the AI in the right situations and avoid misuse,
  • know when to trust and when to verify,
  • and maintain realistic trust instead of disappointment or overreliance.

Avoid Overpromising in Language and Branding

Use accurate, restrained language in product copy and branding—no 'always accurate', 'fully automated', or 'replaces human judgment' when the system has limits or requires human oversight.

off on
Use language that reflects real capabilities: assist, suggest, draft, help—and that the user reviews or decides. Claim always-accurate, fully automated, or no-review-needed when the system has limits or requires human oversight. Align marketing and in-product copy with what the AI actually does and doesn't do. Let branding promise more than the product delivers; avoid "replaces human judgment" or similar overclaims.

Clarify Scope and Capabilities

State clearly what the AI can and can't do, and for whom or in what context—so users don't assume it handles everything or works in every situation.

off on
Describe what the AI is designed to do, in what context (e.g. languages, domains, data), and what it does not do. Use generic labels like "AI-powered" or "Smart" without stating scope or limitations. Call out high-stakes exclusions (e.g. not legal/medical/financial advice) in user-visible copy. Leave scope and boundaries implicit so users infer capabilities incorrectly.

Communicate Accuracy and Reliability

Set expectations for how often the AI is right, when to double-check, and what might go wrong—so users know when to trust and when to verify.

off on
Tell users to review important outputs and that the AI can make mistakes; avoid implying it's always correct. Use language like "precise", "accurate", or "trust our AI" without guidance on when to verify. When the system is uncertain or conditions are edge-case, say so and suggest human verification. Present all outputs with the same implied certainty; hide or downplay uncertainty.

Explain the Role of the AI vs the User

Make clear who does what—suggestion vs decision, draft vs final—so users understand they are in control and where their responsibility lies.

off on
State that the AI suggests or drafts and that the user decides, approves, or is responsible for the final outcome. Let the AI take actions (e.g. send, publish) without making the user's approval and responsibility clear. Use consistent language (e.g. "suggestion", "draft") so the division of roles is obvious. Use wording that implies the AI has made the decision or that the user is not accountable.

Set Expectations Over Time, Not Just Once

Reinforce what the AI can do and what the user's role is at key moments—not only on first use. Remind when context changes or when the stakes are high.

off on
Repeat or reinforce scope, reliability, and user role at important moments (e.g. before send, before high-stakes use). Rely only on first-use or one-time messaging; assume users will remember. When the context becomes higher-stakes (e.g. legal, external audience), restate limitations and user responsibility. Treat every interaction the same; avoid reminding when the cost of misunderstanding is high.

Why this principle matters

Overtrust leads to misuse and harm when users assume the AI is more capable than it is. Undertrust or confusion leads to underuse or frustration.

When we set appropriate expectations:

  • users use the AI where it adds value and avoid relying on it where it doesn't,
  • they verify when it matters and feel less surprised when limits show up,
  • and trust stays aligned with reality instead of marketing.

Without it, users may:

  • rely on the AI for tasks it wasn't designed for or isn't reliable at,
  • be shocked when it fails or discover too late that they misunderstood its role,
  • or dismiss it entirely because early messaging oversold and broke trust.