What happens if the AI voice agent doesn't know the answer?
If an AI voice agent does not know the answer, a well-configured one says so, promises that a person will follow up and records the question so the team can respond. A poorly configured one may guess, which is the real risk. What happens depends almost entirely on the instructions and facts the business gives the agent, not on the underlying model.
Why this question matters
Language models are built to produce fluent, likely-sounding text. Asked a question with no answer in its instructions, a model may produce a plausible guess, such as a price, a delivery date or an eligibility rule, and say it with confidence. On a sales or service call, a confident wrong answer is worse than an honest 'let me check'.
What a good fallback looks like
- 01The agent recognises the question is outside what it has been told.
- 02It acknowledges the question plainly, for example 'Good question, I don't have the exact figure with me right now'.
- 03It offers a concrete next step: a call from the team, details on WhatsApp or a visit.
- 04It captures the question as a field so it appears in the call record.
- 05It returns to the purpose of the call instead of improvising further.
How to set it up
Give the agent the facts it should use: prices, timings, locations, eligibility rules and policies. Then add an explicit rule: if the answer is not in these facts, do not guess. Name the topics it must never commit to, such as discounts, final loan approval, medical advice or legal opinions, and give it the exact words to use for a follow-up promise.
Some platforms connect the agent to a larger knowledge base that it searches during the call, a technique called retrieval-augmented generation. This widens what it can answer but does not remove the need for a 'do not guess' rule, because no knowledge base covers every question a caller might ask.
How Outpero helps you close the loop
Every Outpero call produces a transcript and summary, and you can define custom fields for the agent to capture, such as 'unanswered question'. These appear in the dashboard and arrive on your webhook, so a team member can follow up the same day. Reviewing those questions weekly is also the fastest way to see which facts the agent is missing, and to add them.
Want to hear an AI voice agent handle this on a real call?
Follow-up questions
Can the agent look things up on my website during a call?
+
Not usually in real time, and it is rarely a good idea. Browsing a website mid-call is slow and unreliable. The better pattern is to give the agent the relevant facts up front, or connect a custom API that returns specific answers, such as order status or available units, quickly enough to fit inside a natural pause.