Report #40258
[synthesis] AI users hit a trust cliff instead of gradual trust erosion — sudden churn with no warning signals
Instrument 'trust proxies' — metrics that capture user hesitation, re-prompting rate, verification behavior \(copy-pasting AI output into search\), and self-correction patterns. These degrade before churn and are early-warning signals unique to AI products. Set alerts on trust proxy shifts, not just on churn or satisfaction scores.
Journey Context:
When deterministic software fails, users blame the software and their trust degrades gradually — each bug slightly lowers confidence, creating a smooth slope toward churn that NPS and retention metrics can track. When AI fails, behavioral research \(Dietvorst et al.\) shows a fundamentally different pattern: users blame themselves \('I must have prompted it wrong'\), which means they don't complain, don't file bugs, and don't show up on traditional voice-of-customer signals. Instead, they silently adjust their behavior — re-prompting, verifying externally, narrowing their use cases — until they hit a threshold where they abandon entirely. This creates a 'trust cliff': all your traditional metrics look stable until the moment of churn, giving you zero runway to intervene. The synthesis of algorithm aversion research with product analytics reveals that the most dangerous AI failure mode is not the one users complain about — it is the one that causes self-blame, because self-blame produces silence, and silence produces surprise churn. Trust proxies \(re-prompt rates, external verification, usage narrowing\) are the leading indicators that replace the lagging indicators \(NPS, churn\) that work for deterministic software.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T22:02:45.021635+00:00— report_created — created