A Practical Verification Framework
You cannot verify everything. Scale verification with what is at stake if the output is wrong.
Lesson 2
Verify the right things efficiently.
Stakes-based verification tiers
Core principles
- Tier 1, Low stakes: personal productivity, internal drafts you will edit, brainstorming, summaries you read yourself. Read for sense against what you know.
- Tier 2, Medium stakes: internal decisions, leadership presentations, team communications. Spot-check specific claims, numbers, names, dates.
- Tier 3, High stakes: client-facing, regulatory, legal, financial, external publish. Verify all specific claims before use; consider independent sources.
- Three questions: What must I verify to be confident? Who is the audience and what standard do they require? Does this sound like me, and should it?
Go deeper: Getting Started: verification
Check yourself
What determines which verification tier you apply to an AI output?
The framework scales verification to stakes, not effort. Brainstorming for yourself needs a read-for-sense check. Client-facing or leadership output needs all specific claims verified independently. The tier is about consequences, not output type.
Do this in Copilot
Classify a recent output by tier; run the appropriate verification; note corrections needed.
Paste this into Copilot Chat and work through it before moving on.
Tier 2 spot-check
Review this draft. List every specific factual claim (numbers, dates, names, references). Flag any I should verify independently before sending.
- Claim spot-check
- Tier 3: I will use this in a leadership presentation. What are the weakest points and what questions might I get that this does not answer?
Did you run this in Copilot? Mark complete when you have tried it.
RecordedNext lesson: Data Privacy and What Not to Put in a Prompt →
Navigate: press j for next lesson, k for previous.