Correcting The AI

Every correction is a learning signal. This page is the reference for the specific correction surfaces. For the bigger picture of how each signal feeds back into drafting, see How the AI learns from your team.

Draft feedback signals

When operators work with AI drafts, Ticket0 records one of three outcomes:

• accepted: sent as-is
• edited: changed before send
• rejected: dismissed

These signals are tied to each AI eval run and power quality metrics and nightly analysis.

In-chat corrections

In Live Chat, the per-message Flag button on any AI reply lets you mark the response as incorrect and optionally write what it should have said. This feeds the same quality loop used by ticket drafts.

Coaching notes (explicit teaching)

Beyond per-message flags, you can leave a coaching note on any ticket — widget chat or email. From the conversation, click Coach AI, describe what the AI should have done differently, and save. Notes are distilled overnight into one-line lessons that the AI silently retrieves and applies on similar future questions, across both channels. Manage existing notes and lessons in AI → Lessons.

See How the AI learns from your team for the full loop, including the Review queue that surfaces eval-flagged conversations awaiting a coaching note.

Customer thumbs feedback

Widget customers can leave thumbs up/down on AI messages.

• Positive feedback indicates good answer quality.
• Negative feedback helps prioritize sessions for quality review.

How corrections are used

Corrections feed:

• draft acceptance/edit/rejection analytics
• edit-pattern categorization insights
• team voice and resolution-pattern learning
• coaching-lesson distillation (from saved coaching notes)
• quality filtering in live chat views and the Review queue

Next: Monitoring AI quality.