Coaching teams are putting AI to work in a lot of places: check-in messages, chat between sessions, session summaries, reminders. Meanwhile, state legislatures have been writing rules for exactly that kind of tool, and they've moved fast. Nevada and Illinois banned AI therapy in the summer of 2025. By mid-2026, a handful of other states had joined them, and more than a dozen had passed chatbot safety laws.
Most of these laws weren't written with health coaching in mind. Some of them reach it anyway. Here's what they say, read from the statutes, and what we'd check in your own program.
Two kinds of state laws now reach AI in coaching
The first kind bans AI therapy. These laws say only a licensed professional can provide therapy, and that a company can't offer it through AI instead. Nevada's law took effect July 1, 2025, and Illinois' followed on August 1. In 2026, Colorado, Maine, Rhode Island, Tennessee and Vermont passed their own versions, according to Becker's, which reported on a mid-year count by the Transparency Coalition, a nonprofit that tracks AI legislation. That's seven states.
The second kind sets safety rules for chatbots. New York was first, with a law that took effect November 5, 2025 and covers users of any age. California followed, signing its own in October 2025. The Transparency Coalition counts 14 more chatbot safety measures passed in 13 states in the first half of 2026. Most focus on minors. Several also apply to adults, usually through two requirements: tell users they're talking to AI, and have a plan for when someone talks about suicide or self-harm.
| Type of law | What it requires | Where |
|---|---|---|
| AI therapy bans | Therapy delivered only by licensed professionals; limits on how those professionals use AI | Illinois, Nevada, Colorado, Maine, Rhode Island, Tennessee, Vermont |
| Chatbot safety laws | Disclosure that the bot isn't human, a crisis protocol, extra protections for minors | New York, California, plus 14 measures passed in 13 states in early 2026 |
Where health coaching ends and therapy starts
This is the part that matters most for a coaching team, and it lives in the definitions.
Illinois defines therapy as services "provided to diagnose, treat, or improve an individual's mental health or behavioral health." That's broad. It then defines therapeutic communication, the kind of interaction even a licensed professional can't hand to AI, and one item on that list reads like a description of a good coaching check-in.
Offering emotional support, reassurance, or empathy in response to psychological or emotional distress.
Illinois Wellness and Oversight for Psychological Resources Act
Here's our read. A program built around walking more, eating better or losing weight isn't therapy, and nothing in the Illinois law suggests otherwise. The gray zone is a stress, sleep or substance use program where an AI answers a distressed member with comfort and coping strategies. That starts to look like the thing the law describes, whatever the program calls itself.
The Illinois law carves out three things: religious counseling, peer support from people with lived experience, and self-help materials and educational resources that don't claim to offer therapy. It also bans advertising therapy that isn't delivered by a licensed professional, so the words on your sign-up page count too.
If your team includes licensed counselors or social workers, the rules for them are stricter. Under the Illinois law, they can use AI for scheduling, billing, notes and spotting trends in anonymized data. They can't let it make therapeutic decisions, talk with clients therapeutically, or detect emotions or mental states. Using AI on a recorded or transcribed session takes the client's written consent first.
What chatbot safety laws ask of AI that talks to members
California's SB 243 covers what it calls companion chatbots: AI that gives "adaptive, human-like responses," can meet a user's social needs, and can "sustain a relationship across multiple interactions." It excludes bots used only for customer service or a business's operational purposes.
An AI coach that has a name, remembers each member, and asks how they're feeling this week looks a lot more like the first description than the second. If that's your setup, we'd assume the law applies. Under it, an operator has to:
- Tell users clearly that the chatbot isn't human, whenever a reasonable person could be misled.
- Keep a protocol for responding to suicidal ideation or self-harm, including a referral to a crisis line, and publish the details on its website.
- For users it knows are minors, disclose that it's AI and remind them to take a break at least every three hours.
From July 1, 2027, operators also report each year to California's Office of Suicide Prevention, including how many crisis referrals they made. Anyone harmed by a violation can sue for at least $1,000 per violation.
New York's law reads even closer to coaching. It covers AI that remembers past sessions, asks "unprompted or unsolicited emotion-based questions," and keeps up a dialogue about matters personal to the user. Operators need a crisis protocol that refers users to services like the 988 hotline, plus a notice that the user isn't talking to a human at the start of an interaction and every three hours after. It applies to users of any age, and the attorney general can seek up to $15,000 a day per violation.
How to check your program against AI health coaching laws
These laws follow the member, not the company. California's applies to anyone making a chatbot available to a user in the state. Illinois' covers therapy offered to the public in Illinois. So start with a list of where your members live, then check four things.
This isn't legal advice, and the details differ by state. Have counsel check the states where most of your members live. If you're comparing vendors, ask each one how its AI handles the four points above. It's a fair question to put on any platform comparison.
The technology that works doesn't pretend to be a coach
None of this argues against technology in coaching. It argues for a specific kind: technology that carries the coach further instead of standing in for one.
That's what our own data shows. In a 300-participant controlled study, program completion ran:
The technology in that study wasn't a chatbot. It was ten weeks of scheduled emails and texts carrying the coach's encouragement between sessions. Two people in the coaching-only group reached a third session, against seventeen once those messages were added.
For a coaching company or wellness platform, that's the design that holds up on both counts. The coach stays the coach, the technology keeps members connected between sessions, and nobody has to wonder whether a bot just did therapy in Illinois.
