DEFINITION
When an AI model confidently states something false or made up. To help you spot it, Clivio’s chat answers from one specific document, cites its sources and highlights the passage in the PDF, so you can check it at a glance. Documents that come from outside also wait for your approval before they’re filed.
Why models hallucinate
A language model writes the most likely continuation of a text, based on what it learned and what it is given. It has no built-in sense of true or false, so when information is missing or ambiguous, it can fill the gap with something that sounds right. A typical case: asked for the notice period in a lease, a model may give the usual period rather than the one actually written in your lease.
How Clivio limits the risk
The AI chat answers from one specific document, cites its sources and highlights the passage it relied on in the PDF, so you can check it at a glance. Calculations across your documents, like a yearly total for one supplier, aren’t left to the AI: Clivio does the math on your data. A Factur-X invoice is read from its XML rather than interpreted. And everything that comes in from outside waits in “To sort” until you accept it.
What you can do
Treat an answer as a lead, and read the highlighted passage before you act on it, especially for amounts, dates and deadlines. If a document’s details are wrong, correct the title, type, date or tags, or run the analysis again. Clivio doesn’t publish an accuracy rate, and blurry photos and unusual layouts are where mistakes are most likely.
Frequently asked questions
Can Clivio’s AI make up an answer?
It can get things wrong, like any AI model. That’s why the chat cites its sources and highlights the passage in the PDF, so you can check each answer.
Are totals across my documents calculated by the AI?
No. For a calculation such as a total for one supplier, Clivio does the math on your data, which doesn’t use any AI action.