Every time you correct a classification (in the Inbox or the Simulator), GeniReply learns from the mistake and improves future accuracy. The learning system works in two ways:
1. Exception keywords
When you move a comment from the wrong rule to the correct one, GeniReply analyses the comment text and identifies keywords from the wrong rule that were present. These words are added as exceptions to the wrong rule, so next time the AI encounters them in a similar context, it knows NOT to classify the comment under that rule.
2. Temporal weighting of corrections
Corrections made within the last 48 hours carry much more weight in classification than older corrections. This means:
- When you correct an error, the improvement is immediate — just 2-3 corrections are enough to stabilise a pattern
- Recent corrections are presented to the AI as "MAXIMUM WEIGHT", while older ones serve as supporting context
- The system uses up to 50 corrections as examples for the AI (most recent ones have priority)
💡 You don't need to correct dozens of comments: even just 2-3 targeted corrections on the same error pattern are enough to realign the classification. The AI learns quickly.