SHVYA AI
Book a call ↗
☰
Features ↗Pricing ↗SHVYA Services ↗SHVYA Product Suite ↗Integrations ↗Documentation ↗ Blog ↗ Login
Skip to documentation content
SHVYA Docs Dashboard ↗
Go to workspace

AI not replying or repeating questions

Diagnose missing, delayed, inaccurate or ungrounded AI responses.

SHVYA product documentation

Step 1 — Confirm an eligible new inbound

Find the latest customer message in the correct account and linked CRM lead. A historic import or a message on another provider account is not proof of a new AI-triggering inbound.

Step 2 — Check AI switches and credits

Verify the organization-wide automation AI setting, current pipeline AI, stage AI and lead AI, plus available SHVYA credits and platform restrictions. Ensure the lead's current pipeline matches the connected WhatsApp business number.

Step 3 — Check processing state

Review any exposed AI reply status and the lead's AI Trace. Determine whether the turn was never queued, is processing, was skipped by policy, failed inference, or produced an outbound message awaiting acknowledgement.

Step 4 — Check the Playbook

For repeated qualification questions, inspect the exact saved questions, existing captured attribute values and most recent customer evidence. Use AI Sandbox with one reproducible conversation. For unsupported factual answers, confirm About/FAQ/Knowledge processing and source relevance.

Step 5 — Check channel delivery

A valid AI generation does not prove the message reached WhatsApp or Instagram. Inspect the send status, selected channel, sender, reply window and error code.

Advanced checks for authorized operators

On an authorized SHVYA deployment, compare worker/Beat health, channel callback receipts, queued jobs and safe diagnostic output. The repository provides a read-only WhatsApp AI diagnostic command. Operators should follow the private deployment runbooks and must not paste secrets or full message contents into public support replies.

Common observations

  • AI generated but not sent: delivery/transport or account eligibility issue.
  • AI never generated: toggles, credits, source routing, provider limits or queue issue.
  • Questions repeat: Playbook evidence extraction or attribute mapping mismatch.
  • Language wrong: check configured Bot Languages and conditional Rules.
  • Brochure not sent: sharing readiness/condition and actual file-delivery receipt.

AI Brain · Guided files · Support.

Still need help with this?

Open a support ticket with the affected feature, what happened and a safe error reference.

Contact support ↗

More in Troubleshooting & Debugging