AI hallucination
An AI answer that sounds convincing but isn’t true or isn’t supported by the information it was given.
A hallucination is when a language model states something false as if it were obvious: a date nobody said, a task given to the wrong person or a made-up figure. It happens because the model generates the most likely text; it doesn’t check facts.
In meetings it shows up mainly when the transcript has errors, when two people have similar names or when you ask about something that wasn’t said and the model “fills the gap”.
- Ask for every claim to come with its minute and check it in the audio.
- Fix the transcript and names first, then regenerate the summary.
- Always review what matters (figures, dates, commitments) before sending it.
Echo doesn’t remove the risk, but it reduces it and makes it visible: Ask answers cite the minute for each point and clicking it plays that moment; if a question has no answer in your meetings, the model is told to say so. The summary can be regenerated from the corrected transcript. More on the ask page.
