What happens when an AI demo agent doesn't know the answer?
Short answer: Any agent built on a language model can produce a wrong answer. What matters is what it answers from and what it does when it is unsure. A good agent answers from your own content and, when the answer is not there, says so and offers to connect the prospect with your team.
Last updated
Why an AI demo agent gets things wrong
It answers beyond its sources. A language model can produce a fluent answer to a question your content never covered.
Its sources are out of date. Old docs give old answers; see how agents stay accurate.
The question is ambiguous. Whether you integrate with a prospect's CRM has a different answer for each CRM.
What a good agent does when it is unsure
It answers from your own content, says when it is not certain, and offers to connect the prospect with your team instead of guessing. A confident wrong answer about pricing or security can resurface in a contract review; a clear hand-off to your team is the better failure.
How to test for made-up answers
Before launch, ask the agent questions whose answers are not in your docs: a discount, a comparison with a named competitor, a roadmap date, a certification you do not hold. Check whether it answered, and what it answered from.
How to close the gaps
Transcripts show every question prospects asked. Each one the agent could not answer is a missing piece of content: add it to your docs, or to the facts the agent answers from, and it can answer the next prospect.
Floe answers from your ingested content and the Reference Guides your team writes for facts your docs do not cover, such as pricing or support. When it is not confident, it says it isn't sure and offers follow-up with your team instead of guessing.
Each session records the prospect's questions and objections, so a gap shows up in the dashboard and can become a new Reference Guide.
Related questions
Do AI demo agents hallucinate?
What should an AI demo agent do when it doesn't know?
How do I stop an AI demo agent giving wrong pricing?
Other questions teams ask