Two different counts, two different questions. FPF is measuring chatbot-specific legislation, the bills written around AI systems that talk to people directly. Manatt is measuring the whole category of health-AI legislation, which includes chatbot bills but also billing automation, prior-auth AI, and clinical-decision-support rules. Neither number is wrong. They're just not counting the same thing, and conflating them is how a stat gets mangled on its way into a LinkedIn post.
How many states are actually doing this?
Thirty-four states have introduced chatbot-specific bills this year, and 43 states have introduced some form of health-AI legislation, according to two separate trackers that measure different things (FPF, 2026; Manatt Health, Q2 2026). That's a majority of the country writing rules for this problem at the same time, which almost never happens with a single narrow issue.
The FPF number, 98 bills, 34 states, 3 federal proposals, tracks legislation aimed specifically at AI systems designed to converse with users, including therapy-style chatbots, companion apps, and mental-health-adjacent tools. The Manatt number, 240+, is the wider net: any bill touching AI in a health context, which sweeps in documentation tools, billing systems, diagnostic-support software, and chatbots together.
Why does the split matter to you? Because if you're only using AI for progress notes or intake summaries, you fall into Manatt's broader category, not FPF's chatbot-specific one. Different bills, different exposure. Knowing which bucket your tool sits in is the first compliance question, not the last.
What Illinois started, and everyone's copying
Illinois passed the first version of this rule in 2025, and its structure, admin-only AI with a hard dollar penalty, has become the template other states are now writing into their own statutes. The Wellness and Oversight for Psychological Resources Act (HB1806) took effect August 4, 2025, capping AI's role at administrative and supplementary support with fines up to $10,000 per violation (Illinois IDFPR).
The Illinois rule is worth knowing even if you don't practice there, because it set the pattern. AI can schedule, transcribe, summarize, and draft. AI cannot make an independent therapeutic decision, and it cannot function as the treating clinician. Every state that's followed has kept that same shape: admin support in, independent clinical judgment out.
That's also the line VibeCheck.luxury was built to stay on. Documentation and admin support, never a stand-in for your clinical judgment, which is the same distinction Illinois wrote into law first.
What changed this year in Maine, Rhode Island, and Texas?
Three states moved from bill to enacted law in the last four months, and each one adds a different piece: written consent, live disclosure, or plain-language notice. None of them ban AI documentation tools. All three require you to be upfront about how you're using one.
Maine's LD 2082. Maine's LD 2082, "An Act to Regulate the Use of Artificial Intelligence in Providing Certain Mental Health Services," passed the legislature April 8, 2026, and was signed April 13, 2026 (Maine Legislature, LD2082). It restricts licensed mental-health professionals to administrative and supplementary AI use only, and it requires written informed consent from the client before any supplementary AI use begins. If you're using AI for anything beyond scheduling and notes, and you're in Maine, get the consent form signed first.
Rhode Island's disclosure and fine structure. Rhode Island signed its therapy-chatbot ban and related AI laws in June 2026, under Gov. McKee (Transparency Coalition, June 2026). A chatbot has to disclose its non-human status at the start of a session and again every three hours. Fail to route a self-harm expression to crisis services, and the fine runs up to $15,000 per day. The substantive requirements take effect January 1, 2027, so this is close, but it isn't live yet.
Texas TRAIGA. Texas's TRAIGA (HB149) took effect January 1, 2026, and it requires healthcare providers, including mental-health clinicians, to disclose AI use in treatment in clear, plain language (TMLT guidance, 2026). No exemption for "it's just a note-taking tool." If AI touches the treatment record, the client gets told, plainly, in writing.
The two laws that are enacted but not yet live
Idaho's S1297 and Nebraska's LB525, both titled the Conversational AI Safety Act, were signed into law in April 2026 but don't take effect until July 1, 2027 (Orrick legal analysis, April 2026). That's a real gap between "enacted" and "in force," and it's worth being precise about which one you're citing. A bill that's law but not yet operative still shapes how vendors are building their products right now, even before enforcement starts.
Watching enacted-but-not-yet-live laws matters because vendors read the writing on the wall before the effective date arrives. A tool built to comply with Idaho and Nebraska's 2027 standard today is a safer bet than one racing to retrofit compliance next June.
The six-state bright line: no claiming to be a therapist
Six states, Colorado, Georgia, Hawaii, Idaho, Iowa, and Nebraska, now explicitly bar AI chatbots from claiming to provide professional mental or behavioral healthcare, according to legal analysis from Orrick and the IAPP published in April 2026 (Orrick/IAPP, April 2026). This is the clearest version of the line every other state law gestures at: a chatbot can help, but it cannot say it's a clinician.
This is a narrower, sharper rule than the admin-only laws. It's not about what the AI is allowed to do behind the scenes. It's about what the AI is allowed to say about itself to the person on the other end of the chat. A documentation assistant that never talks to your client doesn't run into this rule at all. A consumer-facing chatbot that lets a user believe they're talking to a licensed provider does, every time.
What this means if you use AI in your practice
Here's the plain-language version. If your AI tool drafts notes, summarizes sessions, or handles scheduling, and it never claims to be a clinician or deliver independent treatment, you're on the admin-support side of every one of these laws. That's the legal category. The risk sits with tools that blur the line, or with any tool your client might reasonably mistake for a person licensed to treat them.
Three things to actually do this quarter. First, check whether your state has a disclosure requirement (Texas and Maine both do now) and make sure your informed-consent paperwork says, in plain language, that you use AI for documentation support. Second, ask your vendor directly whether their tool ever generates client-facing text that could read as clinical advice, not just internal notes, because that's the seam where a compliant tool turns into a liability, and it's exactly what vendor-vetting and BAA questions were built to catch. Third, don't wait for your state to pass a law before you write the disclosure into your own consent form. Illinois clinicians have been living under this since August 2025. The rest of the country is catching up, not inventing something new.
None of this is about whether AI belongs in a therapy practice. It's about which side of the desk the AI sits on. Documentation, drafting, and scheduling sit on your side. Independent clinical judgment and any claim to be the treating provider do not, and the states writing these laws in 2026 are making that distinction a legal one, not just a good-practice one.
FAQ
Is it illegal for a therapist to use AI for progress notes?
No. Every state law in this wave, from Illinois's 2025 statute to Maine's LD 2082, permits administrative and supplementary AI use by a licensed clinician. The restriction targets AI making independent clinical decisions or claiming to be the treating provider, not AI drafting a note you review and sign.
Do I need written client consent to use an AI documentation tool?
It depends on your state. Maine's LD 2082 requires written informed consent before supplementary AI use begins. Texas's TRAIGA requires plain-language disclosure of AI use in treatment. Check your state's specific statute, since the consent bar varies from a disclosure statement to a signed form.
What's the difference between the FPF and Manatt bill counts?
FPF tracks chatbot-specific legislation only: 98 bills across 34 states plus 3 federal proposals. Manatt tracks the broader category of all health-AI legislation, including chatbot bills, at 240+ bills across 43 states. Both are real counts of different, overlapping datasets, not the same number measured twice.
Are Idaho and Nebraska's chatbot laws in effect right now?
Not yet. Both states' Conversational AI Safety Act laws (Idaho S1297, Nebraska LB525) were signed in April 2026 but don't take effect until July 1, 2027. They're enacted law, meaning vendors are already building toward the standard, but there's nothing to enforce on that specific statute until next year.
Which states explicitly ban AI chatbots from claiming to be a therapist?
Colorado, Georgia, Hawaii, Idaho, Iowa, and Nebraska all have laws explicitly barring chatbots from claiming to provide professional mental or behavioral healthcare, per Orrick and IAPP's April 2026 legal analysis. That count is separate from, and narrower than, the broader admin-only restrictions other states have passed.