Do we have to tell customers that AI is analyzing their call?
Requirements vary by jurisdiction and by what the AI is doing, so confirm specifics with your counsel. As a matter of practice, disclosure expectations are strongest when AI interacts with the customer — a voice agent or automated responder — and weaker when AI analyzes an already-disclosed recording internally for quality review. The safe default is plain language in your privacy policy plus a clear announcement whenever a customer is talking to software.
Two situations that get treated as one
Analyzing a recording after the fact is a processing activity. The customer was told the call was recorded; software rather than a manager is now reading it. The governance questions are about access, retention and accuracy.
AI in the conversation is different in kind. If a customer is speaking to a system that answers, schedules or negotiates, they are interacting with software while forming an impression that they are not. That is where disclosure expectations are strongest and where rules are moving fastest.
The synthetic voice line
A useful internal rule, independent of any statute: if a customer could reasonably believe they are speaking with a person and they are not, say so. It costs one sentence and it removes the entire category of complaint that begins with "nobody told me."
The same principle applies to generated text that appears to come from a named employee. Our voice translation technology is in development rather than deployed, and disclosure design is part of that work for exactly this reason — the moment a customer hears a rendered voice, they should understand what they are hearing.
Where the disclosure should live
Three places, doing different jobs. The call greeting handles the moment. The privacy policy handles the durable, detailed explanation of what you collect and how it is used. Any customer portal or booking flow handles the point of data entry.
Keep the privacy policy specific enough to be true. A generic template that does not mention call recording, transcription or automated analysis is not describing your business, and the gap is easy for anyone to spot.
The business case is separate from the legal one
Even where disclosure is not required, consider how the discovery reads. A customer who learns later that their call was transcribed and scored by software experiences it as concealment, regardless of how ordinary the practice is.
Told plainly up front, it is unremarkable — most people assume it already. The asymmetry favors disclosure. Fold the review into the project when you deploy call analysis, and have your counsel confirm the wording for the states you operate in.
Topics: disclosure · transparency · privacy policy · voice AI
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.