How Daylogue Keeps AI Accountable

Daylogue uses AI to summarize check-ins, notice patterns, and turn past entries into narrative reflections. The trust layer adds three controls around that experience: users can see where an insight came from, tell Daylogue when a read is wrong, and remove a remembered detail from future interpretation.

When Daylogue shows an AI insight, eligible views include a source receipt such as "from your past entries." Opening it shows the check-ins that helped shape the insight, using dates, time of day, and short previews from entries the signed-in user can already access.

Users can respond to AI output with "yes, that's it," "not quite," "almost," "forget this," or "remember it differently." These actions are stored in an append-only correction log so Daylogue can learn which surfaces are accurate without silently rewriting history.

SMS users can text NOT QUITE, FORGET, orREMEMBER: followed by replacement wording. Voice check-ins can also record explicit correction intent during the session.

• The correction action, surface, channel, and timestamp.
• Linked check-in IDs when the correction relates to specific entries.
• Short sanitized user wording only when the user supplies replacement text.
• A bounded snapshot of the AI output being corrected.

We do not store raw prompt logs for this feature, and correction records are not used to train foundation models.

Correction UI is suppressed during active crisis handling. When someone expresses self-harm, suicidal thoughts, or acute danger, Daylogue prioritizes crisis resources and exits the normal feedback flow.

Daylogue remains a pattern journal, not therapy or medical care. The trust layer helps users correct product output. It does not create a diagnosis, treatment recommendation, or clinical assessment.

Trust-layer correction records are tied to the user account and are removed on account deletion. Source receipts only return entries owned by the signed-in user under the same row-level access controls used across Daylogue.

For broader commitments, see Trust & Security, AI Training Transparency, and Consumer Health Data Privacy.