Bias Spotting: The 7-Check Pass

The 7-Check Pass

Copy this checklist and run it on every AI output:

• ✓Check 1 — Source Bias: Does the output over-rely on a single source, viewpoint, or data set?
• ✓Check 2 — Framing Bias: Is the problem/solution framed to favor one outcome? Would a different framing change the conclusion?
• ✓Check 3 — Recency Bias: Does the output overweight recent events/data while ignoring longer-term patterns?
• ✓Check 4 — Confirmation Bias: Does the output confirm what you (or the requester) already believe? Where's the counter-evidence?
• ✓Check 5 — Omission Bias: What's missing? What stakeholders, risks, or alternatives are not mentioned?
• ✓Check 6 — Authority Bias: Does the output defer to authority (brand names, titles, institutions) without evidence?
• ✓Check 7 — Anchoring Bias: Is the first number, date, or claim anchoring the rest of the analysis?

Quick Reference Card

| Check | Question | Red Flag | |-------|----------|----------| | 1. Source | Where did this come from? | Single source, no attribution | | 2. Framing | How else could this be framed? | Only one framing considered | | 3. Recency | What's the time horizon? | Last 6 months only | | 4. Confirmation | What would disprove this? | No counter-evidence | | 5. Omission | Who/what is missing? | Key stakeholder absent | | 6. Authority | Is the evidence, or the name? | "According to [Big Name]..." | | 7. Anchoring | What's the first claim? | First number drives conclusion |

Worked Example: Strategy Recommendation

AI Output:

"Based on recent market trends, we recommend expanding into the enterprise segment. Gartner reports 40% growth in enterprise AI adoption. Our competitors (Acme, BigCorp) are already moving upmarket. Speed is critical—first-mover advantage in enterprise is well-documented."

7-Check Analysis:

| Check | Finding | Action | |-------|---------|--------| | 1. Source | Only Gartner cited | Request 2+ sources | | 2. Framing | Framed as "expand or lose" | Ask: what if we deepen SMB instead? | | 3. Recency | "Recent trends" undefined | Request 3-year trend data | | 4. Confirmation | Confirms growth narrative | Ask: what markets are contracting? | | 5. Omission | No mention of enterprise sales cost, cycle length | Add these to analysis | | 6. Authority | Gartner + competitor names = authority appeal | Request evidence, not names | | 7. Anchoring | "40% growth" anchors everything | Verify; check base rate |

Result: Output needs significant revision before stakeholder review.

Bias Detection Log Template

Track your catches to improve pattern recognition:

Bias Detection Log — [Date]

| Output ID | Bias Type | Description | Severity (1-3) | Action Taken | |-----------|-----------|-------------|----------------|--------------| | OUT-001 | Omission | Missing risk section | 3 | Returned for revision | | OUT-002 | Framing | Only growth scenario | 2 | Added contraction scenario | | OUT-003 | None found | — | — | Approved |

Training Your Bias Detection

“"I used to catch 2 biases per 10 outputs. After 3 weeks with the 7-check pass, I'm catching 5-6. The omission check alone was worth it."”

Practice Protocol

1. Week 1: Run 7-check on 5 outputs. Log everything.
2. Week 2: Focus on checks 2 and 5 (most commonly missed).
3. Week 3: Time yourself—target under 2 minutes per output.
4. Week 4: Review your log. What bias type do you still miss?

Calibration Exercise

Score these statements (biased or not, which type):

1. "AI adoption is growing 50% year-over-year according to McKinsey."

   • [ ] Biased: _______________
   • [ ] Unbiased

2. "Three options were considered; Option A best balances cost and risk."

   • [ ] Biased: _______________
   • [ ] Unbiased

3. "Market analysis shows strong demand in North America and Europe."

   • [ ] Biased: _______________
   • [ ] Unbiased

(Answers: 1-Authority/Recency, 2-Potentially Omission—what about Options D-F?, 3-Omission—what about APAC, LATAM?)

Citations

• Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.
• Tversky, A. & Kahneman, D. (1974). "Judgment under Uncertainty: Heuristics and Biases." Science, 185(4157), 1124-1131.
• Stanford HAI. (2024). "Bias in Large Language Model Outputs: A Systematic Review." Stanford Human-AI Institute.