Why is my AI response wrong, empty or slow?
Updated
Open the request in Request Logs. It shows exactly what was sent, what came back, which model answered and how long each step took, which almost always points to the cause.
Where do I start?#
- Go to Request Logs and filter by workflow, source (Production, Playground, Test) or revision (Live, Draft).
- Open the request and check the Timeline: each model tried, each failover, webhook deliveries and backend callbacks, with timings.
- Compare the request and response text with what you expected.
The answer is wrong or off-topic#
- Check which revision ran. Production always uses Live, so an unpublished Draft change won't apply yet.
- Read the full prompt in the log. Missing context or a conflicting instruction is the usual cause.
- Try the same input in the Playground with the Draft, adjust, then publish.
The response is empty or fails the schema#
The model's output didn't match your structured output. Simplify the schema, describe each field, or use a more capable model. See How do I make sure the AI always returns valid JSON?
The response is slow#
- The timeline shows whether the time went to the provider, to failover attempts, or to your own webhook or callback.
- Slow providers: add a faster backup model, or switch the primary.
- Long answers: use streaming so users see text immediately.
- Repeated identical requests: turn on caching.
I got a test answer instead of a real one#
The workflow is in test mode. Switch it back to Production and publish it.