Safer AI habits

Trust‑but‑verify in 90 seconds

AI can be fast—but “fast and wrong” is still wrong. This lightweight checklist helps you catch the common failure modes (bad numbers, missing assumptions, invented facts) before you hit send.

Apply to AI at Work Read: the 5-line prompt brief
Feb 2026 · ~4 minute read · Verification without overthinking

The 90-second checklist

Run these checks any time the output includes facts, numbers, recommendations, or policy-like statements.

Trust‑but‑verify (copy/paste)
  1. 1Claims: Highlight the top 3–5 claims. Which ones must be true?
  2. 2Numbers: Recalculate quickly. Do units match? Any unrealistic totals?
  3. 3Assumptions: What must be true for this to work? Are we missing constraints?
  4. 4Sources: Ask for citations or links. Verify at least 1 critical claim.
  5. 5Context check: Did it follow the goal, audience, tone, and guardrails?
  6. 6Risk check: Any privacy/policy issues, or advice that needs a specialist?

Two fast prompts that help

Use these after the first draft. They surface weak spots without asking for a full rewrite.

Prompt A (assumptions):
List the assumptions you made. For each one, say how to validate it quickly.

Prompt B (risk):
Point out anything that could be misleading, incorrect, or risky. Suggest safer wording.

When to slow down (on purpose)

The checklist is great for everyday work. If the output affects money, health, legal obligations, or public claims—verify more than one source.

Related reads

Improve inputs, then keep them safe.

Mar 2026 · 6 min read

The 5‑line prompt brief

Goal, audience, context, guardrails, definition of done—plus examples you can copy.