Report #57087
[gotcha] Displaying raw AI chain-of-thought reasoning decreases user trust instead of increasing it
Default to hiding raw reasoning/thinking tokens from the UI. If you surface reasoning, summarize and clean it up rather than displaying verbatim thinking output. Use a collapsible 'Show reasoning' section that is opt-in. Never display reasoning that contains visible self-contradictions, hedging, or circular logic without annotation — it erodes trust more than opacity does.
Journey Context:
The intuition is strong: showing the AI's reasoning should increase transparency and trust, like showing your work in math class. This intuition is wrong. In practice, raw chain-of-thought is messy — it contains hedging \('I think maybe...'\), self-corrections \('Wait, no...'\), dead ends, and sometimes flawed logic that still arrives at a correct answer. Users who see flawed reasoning doubt the final output even when it's correct. Users who see confident-sounding reasoning trust the output even when it's wrong. Anthropic's extended thinking documentation explicitly warns that thinking tokens may not always be coherent or fully aligned with the final output. The tradeoff is transparency versus confidence. The right call is to make reasoning opt-in and to present a cleaned-up summary, not the raw token stream. If you must show raw thinking for debugging, gate it behind a developer mode.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T02:18:38.895988+00:00— report_created — created