What happens when a customer asks something the AI doesn't know?
A well-configured AI voice agent says it doesn't have that detail, tells the customer a team member will confirm it, and records the exact question in the call summary. It should never guess. You prevent most gaps by giving the agent your FAQs, prices and policies up front, and you close the rest by reviewing the questions it could not answer every week.
Guessing is the real risk
Language models are built to produce a fluent answer, and without clear instructions they may produce a plausible one that is wrong — a fee, a date, a loan eligibility rule. This is called hallucination. A wrong answer delivered confidently does more damage than an honest ‘let me check’.
The honest fallback
- 01Acknowledge the question in plain words.
- 02Say clearly that the agent does not have that detail right now.
- 03Capture the question exactly as asked.
- 04Give a realistic timeframe for a callback, one your team can actually meet.
- 05Carry on with the rest of the call instead of ending it.
What it sounds like
यह detail अभी मेरे पास confirm नहीं है — मैं note कर लेती हूँ, हमारी team आज शाम तक आपको call करके बता देगी।
Translation: “I don't have that detail confirmed right now — I'll note it down, and our team will call you by this evening to let you know.”
Shrinking the unknowns
- —Start with your ten most-asked questions and their answers; add more each week from transcripts.
- —Put changing facts, such as slot availability, stock or today's prices, in a sheet or system the agent can read live rather than in the script.
- —Write explicit no-go areas: medical advice, legal opinions, guaranteed returns, approval decisions.
- —Check whether the open question was one the caller cared about or a passing curiosity; only the first needs a fast callback.
The open question lands in the call summary and in any custom field you create for it, and goes to your webhook so your CRM can raise a task for a person. For facts that change daily, the agent can read a spreadsheet live or call your own API during the conversation — order status, available slots, inventory — so fewer questions go unanswered in the first place.
Want to hear an AI voice agent handle this on a real call?