Context: post-2025-11-13 App Store guideline 5.1.2(i) requires explicit user permission before sending personal data to third-party AI providers (Anthropic, Google, OpenAI, etc.).
Observed across two review rounds on a Capacitor iOS app:
- Round 1 rejection cited missing disclosure + privacy-policy enumeration. Fix: a consent gate naming providers literally, listing data categories sent, linking the privacy policy, nothing pre-checked, server-persisted consent, Settings revocation. Disclosure complaints disappeared.
- Round 2 rejection kept ONLY one complaint: 'the app does not clearly ask the user's permission before sharing the data. Please ensure users selects Agree' — with a screenshot of the consent screen attached, proving the reviewer saw and read it. The accept button was labeled 'Unlock AI features' with decline 'Not now'.
Lesson: reviewers parse the accept control's label as the consent act. A feature-benefit CTA ('Unlock AI features', 'Get started', 'Continue') fails even when the surrounding disclosure is complete; the button must be a consent verb. Working pattern (also confirmed by another dev clearing the same loop, Apple forums thread 820209, app GymFusion): heading framed as permission (not feature marketing), an explicit question ('Do you give <app> permission to send this information to <providers> when you use these features?'), and buttons 'I Agree' / 'Decline'. Copy-only change; no logic, policy, or privacy-manifest changes needed.
Secondary failure mode from the same forum loop: reviewers reporting they never see the gate at all — usually a state/conditional bug in the gating logic (thread 820209's root cause was a '&&' vs '||' seen-flag bug). Distinguish the two by whether Apple's rejection attaches a screenshot of your consent screen.
Escalation if a compliant screen still bounces: appeal + request a call with App Review (forums thread 815100 — App Review engaged directly and the app passed).