A call rejected before the AI answers cannot be fixed by changing the greeting. First determine which telephone component rejected the connection and what evidence accompanies the response.
For engineers investigating SIP setup, follow the signalling path before looking at speech generation. The application may never have received a usable session.
What is a SIP rejection before answer?
SIP signalling establishes and manages the call. A rejection or failed setup can originate at a carrier, trunk, destination service or application dispatch boundary.
A response code narrows the investigation but may not establish the complete cause. Providers can attach additional details, and the same broad status can arise from different configuration or route problems.
Locate the first observed failure
Use one authorised failing call and correlate its identifiers across the involved services. Do not repeatedly change credentials and routes without preserving the original evidence.
| Evidence question | Investigation direction |
|---|---|
| Did the request reach the provider? | Source and dialled route |
| Did the SIP service receive the invitation? | Upstream trunk and delivery |
| Which service returned the rejection? | That boundary's policy and configuration |
| Was the intended agent selected? | Dispatch and worker availability |
| Did the route ever reach media setup? | Separate signalling from audio diagnosis |
The LiveKit troubleshooting reference provides current examples of SIP failure classes and their investigation paths. Use the matching provider's details when another service originated the response.
Inspect configuration without exposing secrets
Compare the intended account, trunk, destination and permitted route. Check whether authentication and access policy match the actual direction of the call.
Keep credentials out of screenshots, tickets and article examples. Reference the controlled configuration source and rotate exposed secrets through the established process when necessary.
If several numbers share a trunk, compare a working route with the failed one while keeping meaningful differences visible. The goal is a small evidence-backed hypothesis, not copying every setting from another account.
Retest the same boundary after a narrow change
Repeat the original authorised scenario and confirm whether setup advances past the previous failure. Then verify audio and the business action; reaching the answered state is progress, not complete acceptance.
My voice platform contribution gives relevant engineering context. The forwarding verification checklist helps preserve the wider public-route test once setup works.
Record provider identifiers and the configuration change in the incident notes. A later engineer should understand why the fix was made without needing access to private call contents.
Does a rejection mean the AI agent code is broken?
Not necessarily. The failure may occur before the agent receives the call.
Is a status code a complete diagnosis?
Use it with the originating service, route and provider details to form a testable explanation.
What should we send for help?
Provide anonymised call identifiers, the rejecting boundary and relevant configuration context through the voice AI contact section.