daBongo LMS AI Training Courses

Lesson Content

Why verification is non-negotiable.

Copilot generates outputs based on patterns in its training data. It produces coherent, professional-sounding responses regardless of whether those responses are accurate. There is no built-in indicator that distinguishes a correct response from an incorrect one – both can sound equally authoritative and well-written.

This is not a flaw to work around – it is the nature of the technology. The appropriate response is verification habits calibrated to the stakes.

Tier 1 – Low stakes: Plausibility check.

Apply this tier when: the stakes are low, the information is easy to verify quickly if needed, and an error would produce inconvenience rather than significant harm.

Verification action: Read critically – does this response make sense? Does anything seem implausible?

Examples:

• Brainstormed list of meeting agenda ideas
• Suggestions for team-building activities
• Draft social media caption for a personal account
• General explanation of a concept you will research further

Tier 2 – Medium stakes: Spot-check key facts.

Apply this tier when: you will use this information professionally, share it with others, or base a meaningful decision on it – but the potential harm of error is limited and recoverable.

Verification action: Identify the 2-3 most important specific claims and verify each with a quick authoritative source check (official website, industry publication, government data source).

Examples:

• Market trends you will mention in a client presentation
• Industry statistics for a report
• Process descriptions you will share with your team
• Overview of a regulation that affects your work

Tier 3 – High stakes: Verify all specific claims.

Apply this tier when: you will present this information to leadership, include it in a formal document, use it in a decision with significant financial or operational consequences, or share it publicly.

Verification action: Verify every specific claim, statistic, and factual assertion with primary sources. Do not include any unverified specific claim in formal outputs.

Examples:

• Statistics you will cite in a board presentation
• Financial projections based on market data
• Regulatory requirements you will include in policy documentation
• Technical specifications you will share with engineering or legal

Tier 4 – Non-negotiable professional review.

Apply this tier when: the subject matter falls in a domain requiring licensed professional expertise – legal, medical, financial, engineering, HR/employment. Also applies to any decision that is irreversible or involves very high stakes.

Verification action: Consult a licensed professional in the relevant domain before acting. Use Copilot's output to prepare for the professional conversation – not as a substitute for it.

Examples:

• Legal analysis of a contract term or employment decision
• Medical guidance about symptoms, treatments, or medications
• Tax advice or financial planning recommendations
• Engineering safety calculations
• HR decisions involving termination, accommodation, or classification

The anti-pattern to avoid: the fluency trap.

The most dangerous verification failure is assuming that well-written, detailed, professional-looking output is accurate. Copilot's fluency – its ability to produce clean, organized, authoritative-sounding text – is independent of its accuracy. This is the fluency trap: confusing quality of expression with reliability of content. The tier framework is the antidote.