# How do you get an AI agent to say "I do not know"?

Give it 3 things. A narrow set of sources it must answer from. A rule to show the source of every answer. And a clear way out: when the sources do not hold the answer, it says so and offers a person. Agents guess when guessing is the only path. Design a better path.

By Surinder Thakur, AI Product Designer and Builder. Published 17 Nov 2026. Topic: [Trusting AI with money](https://www.surinder.design/work?type=trusting-ai-with-money).

## Why do agents make things up?

Because a model always produces the most likely next words. With no answer in front of it, the likely words still sound sure.

So the fix is not a smarter model. It is a design where not knowing has a place to go.

## How do sources help?

The agent reads from your documents, your records, your pages, and nothing else. Every answer links to where it came from.

A claim with a source can be checked, and the check builds trust.

## What should "I do not know" sound like?

Short and useful. What it could not find, and what happens next. For example: "I could not find that in the contract. A person will reply today."

Never an apology loop. The customer should leave with a next step, not a wall.

## How do you test it?

Ask it questions your sources do not answer, on purpose. Count how often it admits the gap.

Every confident wrong answer in testing is 1 you avoided in front of a customer.

## In 1 picture

An agent answers from your sources, or says it does not know and offers a person.

1. Question: From a person
2. Search sources: Your documents only
3. Answer: With its source shown

When nothing found at step 2: Says so, offers a person. What it could not find, and what happens next.

Not knowing needs a place to go.

## Next questions

- [Why does 1 wrong AI answer break trust in my product?](https://www.surinder.design/work/why-one-wrong-ai-answer-breaks-trust): Because trust in software is asymmetric. 10 right answers build it slowly.
- [Can an AI agent handle customer support without annoying customers?](https://www.surinder.design/work/can-ai-agents-handle-customer-support): Yes, with 3 rules. It answers only from your own documents and shows where each answer came from.
- [How do you test an AI agent before customers use it?](https://www.surinder.design/work/how-to-test-an-ai-agent-before-launch): In 3 stages. Replay past cases your team already decided and compare answers.
- [How do you stop an AI agent from being tricked by a hidden instruction?](https://www.surinder.design/work/how-to-stop-prompt-injection-in-an-ai-agent): You cannot fully stop it, so you limit the damage.

## Questions that lead here

- [What happens when an AI agent fails halfway through a task?](https://www.surinder.design/work/when-an-ai-agent-fails-halfway): A well-built agent stops, keeps what it finished, and recovers in 1 of 3 ways.

Page: https://www.surinder.design/work/how-to-get-an-ai-agent-to-say-i-do-not-know
