AI calls and robocall rules: where an assistant calling for you sits

Two calls, both made by a machine

Picture the call everybody hates. It's dinner time, the number looks local, and a voice that isn't quite a person starts telling you about your car's extended warranty. You didn't ask for it. Neither did the several thousand other people getting the same call tonight.

Now picture another one. Your assistant rings the garage where you left your car on Monday and asks whether it's ready. One call, to one business, because you asked for it.

In general terms, robocall rules are aimed at the first call and not the second. But an AI voice doesn't get a free pass for being good at conversation, and what you use it for matters a lot. This is general information and not legal advice. Rules vary by place and are being updated as voices improve, so check what applies where you are, especially for business calls.

What the rules were built to stop

Think about what made robocalls hated. A computer works down a list. When someone picks up, it plays a recording, or a synthetic voice, selling something nobody asked about. The caller doesn't know who you are and doesn't much care. You were a row on a sheet.

Rules about this, in the US and elsewhere, tend to turn on the same few ideas. Did the person agree to be called? Is the call selling something? Is it automated dialing, or an artificial or recorded voice? Can the person make it stop? Keep those in mind. They're the same ideas that separate a useful assistant from a nuisance.

Four questions that separate the two

Ask these of any call a machine makes, ours included. An errand gets through all four. A sales blast fails the first.

• Who asked for this call? A person with an errand, or a campaign?
• How many numbers is it going to? One, because that's where the car is. Or every number on the list.
• Is it selling? Booking a table brings the business money. A pitch takes the listener's time.
• Can the other person talk back and be heard? A conversation can be ended with "no thanks". A recording carries on regardless.

An AI voice is still an artificial voice

Here's the part people get wrong. Because a modern assistant listens and answers like a person, it's tempting to assume the old rules about recorded messages don't reach it. Don't count on that. Regulators in the US have made clear that they treat AI-generated voices as artificial voices, which means the consent rules for automated calls can apply to them.

So what you do with the voice matters more than how good the voice is. Use it to ring strangers with an offer and you're in robocall territory, however natural it sounds. Use it for the calls you'd otherwise make yourself, and you look a lot more like a person with an assistant.

Where it gets blurry for a business

Personal errands are the easy case. Business use needs more thought.

Answering calls that come in is the simplest, because the caller chose to ring you. Calling a customer back about their own job, their quote, their appointment on Thursday is ordinary business and what they expect. But the further you drift from "they contacted us about this", the more care you need. Ringing last year's customers with a spring offer is marketing, whoever or whatever dials. Ringing a bought list is exactly what the rules exist for.

Not sure which side a call falls on? Treat that as your answer. Get advice first, or don't make the call.

No bulk dialing here, by design

Our line was built as one assistant's phone, not as a calling system. One line is one number, attached to one assistant. It dials when you say, or when your own assistant says, with a brief you can read. There's no bulk dialing built in, and we left it out on purpose.

That rules us out for some buyers and we're fine with it. If you need outbound sales calling at volume, you need a different kind of product and proper advice on consent. If you need the table booked, the supplier chased and your missed calls answered, that's the job this does.

On every call it introduces itself as your assistant, says it's an AI whenever anyone asks, and reports an honest result, including when the answer was no.

When a person should make the call

There are calls where we'd tell you to dial yourself. Anything where you're not sure the person wants to hear from you. Anything with a pitch in it. A call to someone who has already said they'd rather not deal with an assistant.

A person ringing another person raises fewer questions, and two minutes of your own voice can save a good deal of second-guessing. An assistant is for the calls nobody minds receiving.

Where to read more

We've written a longer guide, can an AI legally make phone calls for you?, which covers disclosure, texting and what changes when a business answers its own line. For the receiving end, an assistant's call is not a spam call looks at how the person who picks up can tell the difference. The short list of rules our line keeps is on the trust page.

Common questions

More on trust and etiquette

• Should an AI say it's an AI on a phone call?
• Is it rude to have an AI call a business for you?
• Should an AI ever give your card number over the phone?
• Call recording consent: the basics, and why we keep text not audio