The questions to ask any AI agent vendor

On the source of the answer

• Where does the answer come from - our documents, the internet, or both?
• Can we see the source behind every answer?
• What happens when there is no verified information? Ask them to demonstrate it, not describe it.
• Can we approve the exact wording for sensitive questions in advance?

On languages

One question separates two entirely different approaches, and the language count in the brochure does not reveal it.

• Is the Arabic answer content that exists in Arabic, or a machine translation of the Hebrew?
• If a phrasing is wrong, can we correct it once so the correction holds?
• Is the admin interface translated, or English only?

On dependence and operation

• Who changes content and wording - us or you? And how quickly does a change go live?
• What exactly is required from us to go live, and on what timeline?
• Is access to our internal systems required?
• What happens if we want to stop - what happens to the content, and what stays with you?

Two questions worth asking last

The first tests whether the vendor knows the limits of their own product. A vendor who says there are none has told you what you needed to know. The second separates a statement from evidence.

• What does your system not do well, and when would you advise against it?
• Which claim from the deck can you show me right now, against a document I hand you here?