Hey. If you have used AI to draft a reply to a Google review, tidy up a testimonial, write an ad headline or answer a customer text, this one is for you. The rules on all four are no longer vague. One is federal and already carries fines. Several are state laws with dates on the calendar. None of them come with a small-business pass.
I am not a lawyer. This is one operator reading the rules so you do not have to start from zero.
Reviews: the federal rule already in force
In August 2024, the FTC finalized its rule on fake reviews and testimonials. It is the piece of AI law every small business already lives under. The FTC's own summary lists what it bans. Reviews cannot misrepresent who wrote them, "such as AI-generated fake reviews." You cannot pay or reward anyone for a review that has to be positive or negative. Owners, managers and employees cannot review the business without clearly disclosing the connection. A website you control cannot pose as an independent review site. You cannot use threats or false accusations to get a negative review taken down, and you cannot buy fake followers or views. The rule lets the FTC seek civil penalties against knowing violators.
What you can still do is most of what matters. You can ask real customers for reviews. You can use AI to draft your replies to reviews, as long as the reply is honest and you stand behind it. You can use AI to clean up a testimonial a real customer gave you, as long as the substance is theirs and they approve the final version.
What you cannot do: have AI write a review and post it as a customer, offer a discount for a five-star rating, or let your office manager leave a review without saying she works there. Each of those is a knowing violation. "The tool wrote it" is not a defense.
Marketing claims: "AI-powered" has to be true
On August 27, the FTC finalized orders against Cox Media Group and two marketing firms. Cox Media Group pays $880,000, and MindSift and 1010 Digital Works pay $25,000 each. They sold advertisers an "AI-powered" service that supposedly listened to smart-device conversations and targeted ads by location. The FTC found the service "wasn't based on voice data, and consumers hadn't opted into this service." The money goes back to Cox Media Group's advertising customers.
Two lessons. First, if a vendor pitches you AI targeting that sounds like magic, ask how it works before you pay. Second, the same standard applies to your own copy. If your website says "AI-powered scheduling" and a person does the scheduling, that is a claim you cannot back up. Say what the tool does.
Customer messages: tell them it is a bot
This area is moving fastest, and it is state law, not federal. Squire Patton Boggs' midyear state AI law update counts twelve states with companion chatbot laws. California, New Hampshire and New York are already in force. Colorado, Connecticut, Georgia, Hawaii, Idaho, Iowa, Nebraska, Oregon and Washington enacted theirs in 2026. Colorado, Connecticut, Oregon and Washington take effect January 1, 2027. Georgia, Idaho, Iowa and Nebraska follow on July 1, 2027. Hawaii's has been enforceable since the governor signed it on July 13, 2026. California signed SB 1119 on September 10. It adds usage limits, self-harm protocols and independent audits, operative July 1, 2027. SB 1000, which changes how disclosures on AI-generated content get verified, went to the governor with a deadline at the end of September 2026.
Colorado is the one I read closely, because it is home. The Attorney General's proposed rules for the Chatbot Safety Act require a chatbot to "disclose to all users that they are interacting with AI and not a human," estimate user age, protect minors, run suicide and self-harm response protocols and file an annual report. The law takes effect January 1, 2027. The rulemaking hearing was set for October 26, 2026, and the draft names no small-business exemption.
In practice, companion chatbot laws are aimed at apps built to be a friend. A booking bot or a voice agent that answers your phone is not that. Still, telling the customer it is AI is cheap. It is the direction every state is heading, and it costs you nothing in bookings. Apex does not sell voice agents, but the rule is simple: any voice agent should open by saying it is an AI assistant for the business. The customer messages we draft in your voice work the same way: a person approves each one. Do the same for any chat widget or text-back bot you run, and give the customer a way to reach a person.
Hiring: you own the tool's bias
If you use AI to screen resumes, Connecticut's first AI employment rules start October 1, 2026. From that date, an employer "cannot shield itself from liability for a discrimination claim by asserting that it relied on an AI tool." Layoff notices under the state WARN Act must disclose whether the layoffs relate to AI. Courts may treat documented anti-bias testing as a mitigating factor. Starting October 1, 2027, applicants must get written notice before an AI-influenced decision. Colorado's proposed rules go further. An adverse decision made with an automated tool in employment, education, financial services, housing or insurance must be disclosed within 30 days, with a path to human reconsideration.
Why this lands on you, not a regulator
You might expect Washington to sort this out. Do not count on it. NPR reported that in mid-September 2026 most lawmakers left town until after the November election, short on time and short on consensus about whether AI regulation is even Congress's job. The Associated Press reported that Trump called the regulatory push a "conspiracy" and Speaker Johnson said "the reflex of legislative bodies is to cover things up with red tape and hyper regulation." Sen. Hawley's investigation into OpenAI asks the question the courts will eventually answer: "Who is held liable when AI goes rogue?"
Until then, the answer for a small business is you. Most of your peers are not ready. GTIA surveyed 520 SMB decision-makers and found 84% report positive business impact from AI. But only 44% have an acceptable-use policy, 40% have training and accountability rules, and 39% have data-security rules.
Related: what the AI CEOs agreed at the White House, and a plain AI operating plan by department for what to hand off first.
The compliance checklist
- No AI-written reviews, ever. Not on Google, not on your site, not "as a placeholder."
- No incentives tied to a rating. Ask for a review, but never pay for a specific rating.
- Insiders disclose. Staff and family who review you say so in the review.
- Testimonials are customer-approved. AI can polish the wording. The customer signs off, and the substance stays theirs.
- Every AI claim in your marketing is provable. If you say AI does it, AI does it.
- Every bot says it is a bot. Voice, chat or text, it says so in the first message.
- A person is one step away. Any customer who asks for a human gets one.
- Hiring tools are tested and documented. Keep the paper trail. It is your defense.
- Write a one-page acceptable-use policy. What staff may paste into AI tools, what they may not, and who checks outputs before they go out.
None of this slows down a well-built automation. The rules separate a review bot that drafts honest replies for you to approve from one that manufactures praise. The first is a marketing asset. The second is a fine waiting to happen. If you want the rules baked into the build instead of bolted on later, that is what we do.