AI training for healthcare practices in Chicago

FitMedical, dental and behavioral health practices · 5–200 people

• Primary care and specialty clinicsreferrals, prior auths, patient letters
• Dental and orthodontic practicestreatment-plan explanations, insurance narratives, recall messages
• Physical, occupational and speech therapyplan-of-care letters, authorization requests, home exercise handouts
• Behavioral health and counseling groupswhere HB 1806 sets hard limits
• Practice managers and compliance leadswho own the policy and the vendor contracts
• Multi-location groups and MSOsone set of rules across sites

The curriculum is the one behind our AI training for Chicago teams, with healthcare examples throughout and labs built on invented patients. The workshop runs on whatever accounts your team uses, and because nothing real is typed in, the training is safe in any of them. What staff may do afterwards with real patients depends on the tool and the contract, and that is the second half of the session.

What does a healthcare prompt look like?

A safe prompt describes the situation without identifiers and asks for a draft that a person will finish in the practice’s own systems.

SOURCE   The denial reason and payer policy excerpt below
         ("Patient X", fictional; no names, dates, member IDs).
TASK     Draft an appeal letter arguing medical necessity for
         the denied MRI, using only the facts listed.
FORMAT   Standard appeal letter. Leave [BRACKETS] for patient
         name, DOB, member ID, date of service and NPI.
RULES    Don't invent clinical facts or cite studies. If the
         policy excerpt doesn't support the argument, say so.

The brackets are the habit. Staff learn to build the letter around placeholders and fill in the identifiers in the EHR or billing system, which keeps PHI out of the tool and still saves the drafting. Once a prompt like this works, it goes into the practice’s shared library with an owner. Our guide to prompt engineering basics for employees covers the structure in more depth.

• Any PHI in a consumer or non-BAA accountnames, dates, addresses, phone numbers, MRNs, member IDs
• Screenshots of the EHR or practice management systemone screenshot can hold a dozen identifiers
• Patient photos, wound images and imagingfull-face and comparable images are identifiers
• Recordings of patients or visitsoutside an approved, consented documentation tool
• Therapy session contentHB 1806 limits AI to administrative and supplementary support
• Requests for a diagnosis or treatment decisionfor a real patient, in any general-purpose tool

The trap most staff fall into is partial de-identification. Under HIPAA’s safe harbor method, information counts as de-identified only when 18 kinds of identifiers are removed (among them every element of a date except the year, any geographic unit smaller than a state, phone and fax numbers, medical record and health plan numbers, and full-face photographs) and the practice has no actual knowledge that the remainder could identify the patient. The other route, expert determination, needs a qualified expert’s documented analysis. A referral note with the name deleted but the date of birth, ZIP code and visit date left in is still PHI.

Write these rules down. Our AI acceptable use policy template is a starting point, and the note on shadow IT in small business explains how to find the AI apps staff already use on their phones.

Behavioral health records carry extra weight in Illinois: HB 1806 ties its confidentiality requirement to the Mental Health and Developmental Disabilities Confidentiality Act. If your practice uses biometrics anywhere, read our guide to BIPA compliance for small business. And since January 1, 2026, Illinois employers must give notice when AI is used in hiring or other employment decisions; the Illinois AI employment notice kit covers that.

This is general information from 312 IT Consulting, not legal or compliance advice. We are not a law firm. Confirm how HIPAA, HB 1806 and BIPA apply to your practice with your healthcare counsel and privacy officer before relying on it.

1. Free first call. Specialty, locations, headcount by role, the AI tools staff already use and who your privacy officer is.
2. Tool and BAA check. Which accounts exist, which (if any) are under a BAA, and whether your HIPAA risk analysis and policies mention AI. If not, we start an interim policy.
3. Build fictional patients. Each group sends two or three routine tasks. We rebuild them around invented patients, claims and letters.
4. Foundations. How AI works, tokens and context, prompting, checking the work and privacy, working from your own material, and AI as a thinking partner on an office problem.
5. Applied labs. Saved instructions, three labs on your processes (for example a denial appeal, a prior-auth letter and front-desk scripts) and a multi-step task such as a weekly schedule-gap summary.
6. Library, certificates, follow-up. Approved templates go into a shared library with an owner; every attendee gets a certificate of completion; a few weeks later we review what stuck.

Formats run from a 2–3 hour workshop to a 1-Day Foundations or 2-Day Hands-On program. The AI training results page explains standard versus fully custom programs and shows how attendees in the rollout our founder led rated their confidence before and after: 83% rated it higher. 312 has delivered AI training for companies from small and mid-size businesses up to Fortune 500 firms. If your team wants a tool-neutral session, see generative AI training; for the short format’s agenda, the AI workshop in Chicago.

Tailoring moves the price most: a dental office and a behavioral health group need different fictional patients, labs and rules. Practices with an Illinois location may be able to recover part of the cost through the state’s Employer Training Investment Program, which requires pre-approval. See Funding AI training in Illinois and What AI training costs in Chicago.