Glossary
What is an AI hallucination?
An AI hallucination is a confident statement from a language model that is not true: an invented fact, a feature the product does not have, a price nobody approved or a meeting that never happened. It happens because the model predicts plausible language rather than checking facts. In sales outreach it is a real risk, because the message carries your name.
Published by Sellinger AI. Numbers and settings checked on 24 October 2026 against the fact register. Report an error.
What does a hallucination look like in sales?
| Type | Example |
|---|---|
| Invented capability | Yes, we integrate with that system, when you do not |
| Invented price | A discount nobody approved |
| False familiarity | Great to reconnect after our call last month, when there was no call |
| Wrong fact about the prospect | Congratulations on your funding round, when it was a competitor |
| Invented proof | A customer result that does not exist |
Why does it happen?
A language model is trained to produce text that fits the context. When it lacks a fact, it fills the gap with something plausible instead of saying it does not know. Thin inputs make it worse: a vague product description invites the model to fill in features.
What reduces it?
- Give the model your facts: a knowledge base with features, prices and limits. See retrieval augmented generation.
- Write rules for what it must not say, and what it must hand to a person.
- Exact guardrails on capabilities. In the Lamothermic study, answers on alloys were loose until the president wrote down what the foundry does and does not pour.
- Human approval, especially in the first weeks.
- Reading the conversation log regularly.
How Sellinger handles it
A campaign can run in approval mode, where every message waits for a person to approve it, or send automatically. A person can take over any conversation, and the agent stops on its own when it cannot answer, for example a price it has no answer for.
The capability guardrails on day one playbook shows how one customer wrote down exactly what the business does and does not do, so the agent stopped guessing.
Common questions
Can hallucinations be removed completely?
Who is responsible for a hallucinated message?
How do I spot hallucinations?
Sources and method
General AI practice; examples are illustrative. Lamothermic from its case study. Sellinger behaviour from the fact register, checked 24 October 2026.
Related: AI guardrails, Large language model, Capability guardrails playbook, What an AI SDR cannot do.
