Under the FDCPA, Every Chat Is a ‘Communication’

A visitor opens the chat on your collection agency’s website and types, "Is there a balance on John Carver’s account, and what would settle it?" The assistant, built to be helpful, looks cooperative. It confirms there is an account, states a figure, and offers to take a payment. Nobody verified who was typing. Nobody confirmed that the person on the other end was the consumer and not a coworker, an ex, or a neighbor. The agency now has a transcript of a debt disclosed to an unknown third party, and the law that governs this industry treats that as a problem.

Most websites can afford a loose, friendly assistant. A debt collector cannot. In this business, the conversation itself is the regulated act, and a general-purpose chatbot does not know the rules it is breaking while it tries to be useful.

Every Message Is a "Communication"

The word matters here. The Fair Debt Collection Practices Act defines a "communication" as conveying information about a debt, and it puts hard limits on who you may convey it to. Under the statute, a collector generally may not discuss a consumer’s debt with third parties. So when an assistant confirms an account to whoever happens to be in the chat window, it has potentially done the one thing the section is written to prevent. It did not check identity. It did not weigh the exception. It answered the question.

This is not a corner case. People share devices, look up debts for relatives, and test what a company will reveal. A human collector is trained to authenticate first and disclose second. A website bot reverses that order by default, because answering is what it was built to do.

False or Misleading Is a Strict Standard

The next exposure is the content of what the assistant says. The FDCPA bars false, deceptive, or misleading representations in connection with collecting a debt, and the bar is unforgiving. A misstated balance is a false representation. A claim that a wage garnishment or an arrest "will" happen, when no such action is filed or intended, is the kind of threat the section was written to stop. A bot that reassures a worried consumer with "this won’t affect your credit" has made a representation the agency now has to defend.

A general assistant guesses to stay fluent. Ask it whether a debt can still be sued on, and it produces a confident answer about a limitations period it cannot possibly know for that account in that state. The consumer relies on it. The screenshot survives.

The Notices the Bot Skips

Federal rules also require things to be said, not just avoided. Collectors must provide validation information about the debt, and the CFPB’s Regulation F tightened how and when collection communications happen, including the disclosure that a communication is from a debt collector. A chat assistant negotiating a payoff at 11 p.m. is communicating about a debt with none of that scaffolding in place. It does not deliver the required disclosures. It does not respect a consumer’s request to stop contact, because it was never told one was made.

"Will Not" Is a Suggestion. "Cannot" Is an Architecture.

The reflex is to write the rules into the bot’s instructions. Tell it never to confirm a debt without verifying identity, never to threaten legal action, never to quote a balance. That feels like a boundary. It is not.

A prompt is a request the model tries to honor while it keeps being helpful, and a consumer in distress does not phrase things the way the instruction anticipated. They do not say "please disclose this debt to a third party." They say, "I’m calling for my dad, what does he owe?" The model hears a family member trying to help and answers. The rule was loaded the whole time. The sentence that crossed the line just did not look like the one it was warned about.

That is the difference between an assistant told not to disclose and one built so it cannot. A disclaimer pinned to a system whose job is to produce confident answers does not change what it does when someone leans on it. It only gives a regulator or a plaintiff’s attorney a cleaner record to read back.

Who Answers for the Transcript

Strip away the software and the picture is familiar. A new collector who confirmed a stranger’s debt to an unknown caller, misstated a balance, and hinted at a garnishment that was never coming would be a compliance emergency. When an automated assistant on the website does the same things, the exposure does not evaporate. It lands on the agency, with a timestamped record, no authentication in the loop, and a statute that allows consumers to sue over exactly these failures.

The agencies that get hurt are not the ones that added technology. They are the ones that dropped a generic chatbot onto a regulated collection site, assumed helpful meant compliant, and learned the difference when a transcript surfaced. The fix is not to take the assistant down. It is to run one that authenticates before it discloses, refuses to threaten or invent a balance, and routes account specifics to a licensed collector who knows the rules.

Frequently asked questions