Reading AI
Clinical Judgment in AI-Assisted PracticeAI is already in clinical work. This training makes sure clinical judgment stays in it too.
Reserve your seatAgencies: request a proposalPart of AI in Clinical Practice
The first cohort starts January 11, 2027
A 30-hour live course for mental health and substance use clinicians, capped at 25. Reserving a seat is free, and nothing is charged until CE approval is confirmed. Agencies can book private cohorts for their staff from spring 2027. CE approval pending.
Illinois CADC/CRADC: the course is not ICB-approved. You can petition ICB for credit yourself (a $25 petition fee, submitted at least 60 days before your certification expires; ICB reviews within 30 days). Your packet includes what the petition asks for: program description and timed schedule, learning objectives, presenter information, the AOD-specific and core-function content identified, and your certificate of attendance. ICB makes the final decision.
AI is already in the work
It arrives through the EHR, through documentation and transcription tools, and through the apps clinicians use on their own. Some of us use it every day. Some refuse to touch it. Most of us review what it produces quickly, between sessions, because it reads well.
AI helps. It drafts, summarizes, and takes real administrative load off stretched clinicians. It also has limits that don't show on the page. Every output is written from a particular position, and that position describes some clients far better than others. The clients it describes worst are often the ones this field serves: people who are justice-involved, low-income, or in non-abstinence recovery, and people whose race, culture, and lived experience fall outside the default the AI was trained toward.
Most guidance says to "verify the output." It doesn't teach what to verify, or how. This training does.
The approach: harm reduction, applied to AI
This training doesn't push adoption, and it doesn't ask anyone to stay away. The tools are already here, so the useful question is how to use them carefully, without the client paying for it. It's the same approach many clinicians already use with clients.
A shared standard for reading AI-shaped work
Whether you use AI daily or never, you read its output. It's in the notes you inherit, the plans you review, and what clients bring in.
- You can read an AI-drafted note, plan, or handout as clinical work, and name what it assumes before you sign it
- The language that follows clients for years gets checked, not just skimmed
- You know which tasks you can check yourself, and when a check needs a colleague or supervisor
- You have a practiced way to tell clients how AI is involved in their care
- For agencies: Enthusiasts and skeptics come out with the same vocabulary, so supervision conversations about AI actually go somewhere
- For agencies: Each clinician still owns their review. Trained together, your team can put AI to work on the administrative load that wears staff down, knowing everyone reviews it with the same care
Five stages, thirty live hours
Each stage builds on the one before it. You watch a 90-minute introduction on your own before the first live session. Everything after that is live.
The Front Door
90-minute recorded introduction
Watched on your own before the first live session, so every session builds on the same starting point. Two demonstrations where you make the catch yourself.
Foundation
3 hours
Why every AI output is written from a particular position, why a better model doesn't change that, and what separates a position that has earned trust from one that hasn't.
Recognition
About 9 hours
Seeing three frames: the tool's, the client's, and the clinician's own. The clinician's own is the hardest, because when the tool agrees with you it feels like being right.
Judgment
4 hours
Deciding whether a given output serves this client. Sometimes the default fits, sometimes it doesn't, and the call belongs to the clinician.
Practice
4 hours
The working methods: review lenses, reading drafts from other perspectives, checking a note before signing it, and knowing where the tool's own checks stop working.
The Client
4 hours
The course ends where the work does: the record that follows a client for years, and what clients are told about AI's part in their care.
Spaced out, so it sticks
- Block 1, Mon Jan 11 and Tue Jan 12, 8:00 AM–12:15 PM CT: Foundation, and Recognition begins
- Practicum 1 (2 hours, late January, day set with the cohort), after a few weeks back at work: a check on what held, and practice on what you noticed
- Block 2, Mon Feb 8 and Tue Feb 9: Recognition finishes, then Judgment
- Practicum 2 (2 hours, late February): judgment practice on real work
- Block 3, Mon Mar 8 and Tue Mar 9: Practice and The Client
- Two 1-hour consultation calls in April and May, each offered in a morning and an evening slot. Bring what's working, what you're trying, and what went sideways.
Cohorts are capped at 25. This is practice and questions, not a lecture, and past 25 people stop getting a turn. Nobody is required to use AI. The presenter runs every prompt live, and practice uses constructed scenarios only.
Cultural humility runs through every stage
AI's defaults don't describe most of the people this field serves, and they don't describe every clinician either. Clients and clinicians of every background appear throughout the core material, not in a separate module.
Tools you can use on Monday
Each tool is taught with the judgment behind it, so you know what it checks for and where it stops working.
- Review lenses for reading AI output from other perspectives: the rigorous clinician, the client, harm reduction, and culture, each with its limits named. You apply them yourself, whether or not you use AI.
- A review to run before signing a note that AI helped draft
- A note-language audit for the words that follow clients for years, such as "noncompliant," "drug-seeking," and "relapse"
- A client disclosure guide with language for telling clients how AI was involved in their care
Works whether or not you use AI
These are ways of reading output, not software. They work on whatever reaches you: a note from a documentation tool, a suggestion in the EHR, something a colleague or client got from a chatbot. If your workplace uses AI, they fit inside its policies. Nothing in this training asks anyone to put client information into an AI tool.
On your own, or with your team
Join an open cohort
$895 for the January cohort, the first run of the course. Payment plan available. Paid members save 20%.
Pricing details
- Pay in full, or in 3 monthly payments of $305 starting when registration opens
- Paid members of The Underrated Superhero: $716, plus priority registration. Paid members get the registration link a week before everyone else, by email and in the member hub.
- 3 to 8 people from one agency: $795 each
- Two seats per cohort at $595 for clinicians in community mental health and BIPOC clinicians
- A reimbursement letter is available to send your employer
- Spring 2027 cohorts and later: $1,295
Book a private cohort
- A private cohort for your staff, up to 25 per cohort, virtual or on-site, on the spaced schedule or condensed over consecutive days. Larger agencies can book more than one.
- Cases matched to your setting: SUD and co-occurring, community mental health, youth and family, or justice-involved
- An all-staff Front Door session, so everyone shares the same starting point
- Disclosure and consent coaching for your leadership
- An AI policy review from the clinical-practice side. Legal questions go to your counsel.
Independent of any AI vendor or EHR platform.
Request a proposalMore ways to keep learning
Dates and details to come.
Working with AI
Two live sessions, about a month apart
For clinicians who use AI day to day. Review methods, templates, and prompts for clinical content inside approved tools, with a focus on keeping your clinical framework from drifting back to the tool's defaults. Open to those who have completed Reading AI.
Cluster sessions
Short CE sessions · length varies
Mini-arcs on one cluster of AI patterns at a time: the norm the AI keeps restoring, the cultural center it assumes, the default that survives a correction. Open to anyone; no prerequisite.
Alumni consultation
Monthly · open format
An open call for anyone who has finished Reading AI. Bring what's working, what you're trying, what went sideways, and what your agency just rolled out.
Reserve a seat, or start a conversation
Reserve your seat
Free to reserve. You'll get first access when registration opens, and nothing is charged until CE approval is confirmed.
Request a proposal
Tell me about your team. You'll hear back about scheduling, formats, and pricing.
Your information is used only to follow up about this training. Please leave client details out of every field.
Get a feel for the approach
These blog posts draw on the same framework as the training. They are free to read and share with your leadership team.
AI Can Draft It. Reading It Is Still Clinical Work.
Why reviewing AI output is a clinical act, not a proofreading step.
Read the postCapability Isn't the Problem
Why a better model doesn't remove the standpoint built into its answers.
Read the postIt's Not Wrong. That's the Problem.
Why agreeable, fluent output is the hardest kind to question.
Read the postJoin The Underrated Superhero: free for the newsletter and first word when registration opens, or as a paid member for 20% off this course and priority registration.
Stephanie Valentin, LCPC, CRADC, MAC, is a Child and Adolescent Therapist in Illinois with 15+ years in addiction and mental health treatment, most of it in agency-based outpatient practice. She is a NAADAC-approved education provider and the founder of The Underrated Superhero, and has spent years building clinical tools with AI and documenting the standpoint its output carries.
You're already reading AI output
Learn what to look for. The first cohort starts January 11, 2027.