"I'm not sure, let me check with the team": why a good agent says it, and how to make it say it less

New owners often ask why the agent said "I'm not sure" to a question they think is obvious. The short answer is that the agent did not have the information, and it chose honesty over invention. That choice is the most important thing about it, and it is also fixable in ten seconds.

By Ritchie, ReplyKit · Published 5 September 2026 · 5 min read

Quick answer

A well-designed AI agent answers only from the knowledge you give it. When a question has no matching knowledge, it says it is not sure, offers a handover, and logs the question for you to teach. The alternative, an agent that always answers, invents prices, policies and promises that you are then bound by. To reduce "not sure", add the missing facts; every occurrence appears in ReplyKit's unanswered-questions list.

Why it happensThe four usual causesMaking it say it lessWhen you should be glad it said itFAQ

Why it happens

Language models are fluent; that is the danger. Asked "do you do Sunday appointments?" with no information, a model will happily say yes. A grounded agent instead searches your knowledge for something relevant, finds nothing about Sundays, and declines to guess. In ReplyKit, that decline is a specific behaviour: say you are not sure, offer a person, and write the question to the unanswered list. The customer is never stuck, and you are never on the hook for an invented answer. Our terms make clear that what the agent says is said on your behalf, which is exactly why guessing is not allowed.

The four usual causes

  1. The fact is not written anywhere. Sunday hours live in the owner's head. Write them down.
  2. The fact is written badly. A price in a photo, a policy split across two pages, a table that lost its headers. See writing a price list the agent can read.
  3. The question uses different words. "Do you take walk-ins?" versus a knowledge note that says "appointments only". Add the customer's phrasing to the answer.
  4. The question is genuinely outside scope. "Can you recommend a good dentist nearby?" Handing over is correct.

Making it say it less

Open the unanswered-questions list weekly. Each entry is a customer's exact words. Type the answer once; it becomes knowledge and the next customer gets it. Ten minutes a week drives the rate down fast, and the teach loop guide covers the routine. Then test the seeded questions in the test chat with different phrasings, because a customer will never ask the way you wrote it.

When you should be glad it said it

An agent that never hedges is not smarter; it is unsupervised. The hedge is the supervision.

Frequently asked questions

Why does my AI agent keep saying it is not sure?

Because the answer is not in its knowledge, or is written in a way it cannot use. Add or rewrite the fact; each occurrence is logged for you.

Can I make it answer anyway?

Not in ReplyKit, and you would not want to. An agent that answers without knowledge invents prices and policies you are then bound by.

What happens to the customer when it is not sure?

It offers a handover, so the customer reaches a person, and the question goes to the unanswered list.

How fast can I fix it?

About ten seconds per question from the unanswered list. Ten minutes a week keeps it near zero.

Is 'not sure' ever the right final answer?

Yes, for out-of-scope, regulated or customer-specific questions. Handing over is correct there.

Turn every 'not sure' into an answer

The unanswered list shows exactly what to teach. 7-day free trial.

Start free
Written by Ritchie, ReplyKitPart of the small team in Malaysia that builds and runs ReplyKit. Writes about WhatsApp automation, AI customer service and small business lead response.

How we know this. ReplyKit is deliberately built to answer only from your content and to hand over when it cannot; this article explains that design and how the unanswered-questions list turns each 'not sure' into knowledge. Product figures in this article (limits, prices, per-reply costs) are taken from ReplyKit as it runs today and are re-checked when we update the page. Where we cite outside research, the source is linked below. We sell ReplyKit, so read our product claims with that in mind; we say where a different tool or no tool is the better choice. About ReplyKit.

Sources and further reading

  1. ReplyKit Terms of Service, section 4 · statements by the agent are made on your behalf
  2. ReplyKit Acceptable Use Policy · what we refuse to automate

ReplyKit is an independent product and is not affiliated with or endorsed by Meta or WhatsApp. This article is general guidance for running a small business, not legal, financial or regulatory advice.