AI-written reviews, texts and calls: the rules contractors must follow

AI has made it cheap to write a review, answer a text and place a phone call. The rules about reviews, texts and calls didn't get cheaper. If anything, regulators have spent the last two years making clear that the old rules apply to AI in full: the Federal Communications Commission ruled in 2024 that AI-generated voices are "artificial" voices under the Telephone Consumer Protection Act, and the Federal Trade Commission's rule on fake reviews, in force since October 2024, covers reviews written by anyone or anything.

This post explains, in plain English, which rules apply when AI writes or sends messages for a contracting business, and gives you a four-question check to run on any AI communication before it goes live. It is not legal advice. For anything you're unsure of, especially state law, talk to a lawyer who handles telemarketing and advertising.

The four-question check for any AI message

Every rule below fits into one of four questions. Run them on each AI feature you use, from the receptionist to the review request.

1. 1

   Who is speaking?

   Does the customer know they're dealing with software and which business is responsible? Disclosure rules and the FCC's identification rule live here.

2. 2

   Who started the conversation?

   Did the customer call or text you first, or is your system reaching out? The TCPA's consent rules apply to calls and texts you initiate.

3. 3

   What kind of message is it?

   Conversational, informational or promotional. The consent you need rises with each step, and carriers filter on it.

4. 4

   Is it true, and is it theirs?

   Is every fact accurate, and if it's a review or testimonial, did a real customer write it about a real experience? This is the FTC's territory.

1. Who is speaking: AI disclosure

There is no single federal rule requiring every AI conversation to announce itself, but several rules point the same way.

For outbound artificial voice calls, FCC rules require the message to state clearly at the beginning the identity of the business responsible for the call, and to give a callback number during or after the message.¹ Since the FCC's 2024 ruling, that includes AI voices.

For online bots in California, state law makes it unlawful to use a bot to communicate with someone online with intent to mislead them about its artificial identity to drive a sale, and a business is not liable if it discloses that it's a bot. The disclosure must be "clear, conspicuous, and reasonably designed" to inform people.² Other states, including Utah, have passed their own AI disclosure rules, and they change often.

The practical rule: have the AI say it is a virtual assistant for your company in the greeting, and answer truthfully if anyone asks. It costs you nothing, it's the law in some places, and customers who discover a hidden bot rarely feel better about the company that hid it.

2. Who started it: inbound versus outbound

This is the question most owners skip, and it decides which consent rules apply.

The TCPA restricts calls made using "an artificial or prerecorded voice" without the called party's prior express consent. In February 2024, the FCC confirmed those restrictions cover "current AI technologies that generate human voices," including voice cloning, so callers need prior express consent before making a call with an AI voice, and prior express written consent if the call includes an advertisement or telemarketing.³

Note the word "making." The TCPA's robocall rules are about calls you initiate. An AI that answers your inbound line is in a different position from an AI that dials web leads. That doesn't make inbound AI a free-for-all (recording consent, disclosure and honesty rules still apply), but it does mean the heaviest TCPA consent rules land on outbound.

Texts follow the same logic. A customer who texts you first and gets a quick reply is having a conversation. A system that texts a phone number from a lead form is initiating contact, and the consent attached to that lead form matters.

3. What kind of message: consent rises with intent

The wireless industry's own guidelines, published by CTIA and used by carriers to filter traffic, sort business texts into three types.⁵

Adding a sales pitch to an informational text can make it promotional. That matters for AI because language models like to be helpful: a reminder text drafted by AI that ends "and ask about our membership!" may have just changed category. Keep promotional language out of transactional templates unless the customer opted in to marketing.

Carrier registration comes first. U.S. carriers require businesses texting from ordinary 10-digit numbers to register their brand and use case under the A2P 10DLC system. Twilio's documentation notes unregistered traffic pays extra carrier fees and registered traffic sees lower filtering.⁶ An AI text agent on an unregistered number is likely to have its messages filtered before they reach anyone.

Opt-outs are where AI can hurt or help. FCC rules say consent can be revoked "using any reasonable method." STOP, QUIT, END, REVOKE, OPT OUT, CANCEL and UNSUBSCRIBE are automatic, but if a customer replies with other words that a reasonable person would understand as a request to stop, you must treat it as one, within ten business days at most.¹ Language models are good at recognizing "please quit messaging me." The rule to build: AI may detect a likely opt-out, but enforcement happens in code, immediately, and a person reviews borderline cases.

Quiet hours. Federal rules bar telephone solicitations to residential subscribers before 8 a.m. or after 9 p.m. local time at the called party's location.¹ Several states set narrower windows or extra limits. An AI that drafts at 11 p.m. should queue its texts until morning.

4. Is it true, and is it theirs: reviews and testimonials

The FTC's Rule on the Use of Consumer Reviews and Testimonials took effect in October 2024. Several parts map directly onto AI tools.⁷

• Fake reviews. A business may not write or create a review that misrepresents that the reviewer exists, used the service, or had the experience described. An AI-drafted review posted under a customer's name is exactly that.
• Buying reviews by sentiment. Offering anything of value in exchange for a review that expresses a particular sentiment, positive or negative, is a violation. A discount for "a five-star review" is out. A neutral ask is fine.
• Insider reviews. Officers and managers writing reviews without disclosing their relationship violates the rule, as does a manager soliciting undisclosed reviews from employees or their relatives.
• Review suppression. Using unfounded legal threats or intimidation to get reviews removed, or hiding negative reviews on your own site while implying you show them all, is a violation.

Google's own policy goes further for its platform: it prohibits offering incentives for reviews, and it prohibits discouraging negative reviews or selectively soliciting positive ones.⁸ That second point is review gating, the practice of asking for a rating privately and only sending happy customers to Google. If you use a private rating step, show every customer the public review links regardless of their score.

Penalties: why this is worth an afternoon

TCPA damages are per message, which is why class actions follow mistakes made at scale. An AI text agent that sends 2,000 non-compliant texts doesn't make one mistake. It makes 2,000.¹⁰

The do and don't table for AI customer messages

Questions to ask your software vendor

• Where is texting consent recorded for each customer and lead, and can I export it?
• What happens when a reply means "stop" but doesn't say STOP?
• Does the AI's greeting disclose it's a virtual assistant, and can I see the default?
• Are texts held during quiet hours automatically, by the customer's local time?
• Can the AI add promotional content to reminder texts on its own?
• Does the review request go to every customer, and does every customer see the public links?

How TradesWork handles these

In TradesWork, review requests go out after every completed job with your public review links shown to every customer, whatever they rate, and the reviews feature doesn't offer rewards for reviews. The lead inbox text agent needs approved business texting and the customer's consent on website, webhook and Zapier leads before it texts, holds texts during quiet hours, never texts a price unless you allow your published from-prices, and hands emergencies, complaints and opt-outs to a person. The AI receptionist (early access) answers inbound calls only and introduces itself as a virtual assistant by default. For the texting rules in more depth, see texting customers legally, and for the review rule from the marketing side, asking for reviews after the FTC rule.