Can AI Answer Your Business Emails? Yes. Just Don’t Let It Be Steve.

At our Claude Users Meetup in Pasadena on October 28, the AI disclosure question came up. Can AI answer your business emails? Do you have to tell people when it does?

I had a story ready, because I learned part of this the dumb way. I built an outbound AI voice agent, pointed it at accountants and bookkeepers, and annoyed at least a dozen of them. I got called an a**hole so you don’t have to.

So I went digging afterward. Short version: AI can help with your email. The phone is a different story. And the biggest risk isn’t a law at all.

Not legal advice. I build websites, and I’m chronically online. If your setup is anything beyond “AI drafts, a person sends,” talk to a lawyer.

AI voices on the phone need consent first

In February 2024 the FCC issued a declaratory ruling (FCC 24-17) confirming that AI-generated voices count as an “artificial or prerecorded voice” under the Telephone Consumer Protection Act. In plain terms: calls that use an AI voice need the prior express consent of the person you’re calling, unless it’s an emergency or another exemption applies. Same as an old-school robocall.

That’s the side of the line my voice agent experiment lived on. It’s also why Google’s AI calling businesses to ask about prices gets so much attention. That ruling is about calls and robotexts. It doesn’t say anything about the email sitting in your drafts folder.

California’s AI Transparency Act is about images, audio and video

The California AI Transparency Act (SB 942) comes up a lot in these conversations. Read the text and two things jump out. It covers image, video and audio content (or any mix of those). And the duties fall on “covered providers,” meaning companies that make a generative AI system with over 1,000,000 monthly users that’s publicly available in California. They’re the ones who have to offer free AI detection tools and build disclosures into that media.

Text isn’t on that list, and a small business using an AI tool isn’t a covered provider. Your “Thanks so much, circling back!” reply isn’t what that law is about.

The rule to respect: your bot can’t pretend to be a person

California’s bot disclosure law (Business and Professions Code sections 17940 to 17943, from SB 1001) has been in effect since July 1, 2019. It makes it unlawful to use a bot to communicate with someone in California online while misleading them about its artificial identity, in order to push a purchase or sale, or to influence a vote.

The fix is written right into the law: if you disclose that it’s a bot, clearly and conspicuously, you’re not liable under that section.

So if your AI is out there replying to leads as “Steve from Sales,” and there is no Steve, and there has never been a Steve, that’s the part to fix. Whether a reply in someone’s inbox counts as “online” under that law’s definition is a question for a lawyer. I treat it like it does, because the fix costs nothing.

The fix is almost embarrassingly simple

Put “drafted with help from our AI assistant” in the signature. One line. Then move on with your life.

This one is personal for me. Our team inbox has a bot with its own seat, so I think about where the AI ends and the person begins a lot.

The bigger risk is the AI itself

The law isn’t what keeps me up at night. The AI will cheerfully promise a refund you never offered, agree to a deadline you can’t hit, or apologize for something that never happened. It wants to be helpful, and “helpful” sometimes means making stuff up.

Two guardrails handle most of it:

• Keep a human on money, promises and upset customers. Refunds, pricing, deadlines, complaints. A person reads and sends those, every time.
• Check that your tool isn’t training on your clients’ emails. Open the data or privacy settings in the tool itself, not the marketing page, and see what it says about training.

The whole policy fits on a sticky note

AI drafts, you hit send. That’s it. That’s the policy.

If you want a short to-do list for this week:

1. Add an AI-assist line to the signature of any inbox where AI drafts replies.
2. Write down the topics that always get a human: refunds, pricing, complaints, anything that sounds legal.
3. Check your AI tool’s data settings for training on your conversations.
4. If anything in your business makes AI voice calls, sort out consent before the next dial. Learn from my mistakes.

I’m curious how other people word their disclosure. If you’ve got a signature line you like, I’d love to steal it.

Sources

• FCC Declaratory Ruling FCC 24-17, CG Docket No. 23-362, adopted February 2, 2024: docs.fcc.gov
• California SB 942, California AI Transparency Act, bill text: leginfo.legislature.ca.gov
• California Business and Professions Code, Chapter 6, Bots, sections 17940 to 17943 (SB 1001, 2018): leginfo.legislature.ca.gov