In private practice, every role a hospital distributes across departments lands on you: compliance officer, billing department, records custodian, marketing team. That’s the real frame for LLM use here — not “can the tool do this task” but “which of my roles does this task belong to, and what does that role require?” Nobody is checking your work. That’s the appeal and the exposure.

The Practice Realities

• You are the BAA decision. No institution has vetted a tool for you. Consumer plans (ChatGPT Free/Plus, Claude Free/Pro/Max/Team) cannot be covered by a BAA — if client information touches an AI tool, you either arrange BAA-covered access yourself (API-based or eligible enterprise tools) or de-identify completely, every time. The comparison matrix has the current landscape.
• Your documentation is the payer case. Superbills and claims stand on CPT-coded services (92507, 92523, 92526 and friends) backed by notes that establish medical necessity on their own. LLMs organize this well — and will also happily generate necessity language your data doesn’t support, which becomes your problem at audit, not the model’s.
• Denials and appeals are a writing task with rules. An appeal answers the payer’s stated denial reason with plan-of-care evidence. This is one of the highest-value LLM uses in private practice, and one where fabricated specifics are catastrophic.
• Telepractice crosses state lines; your license may not. Interstate practice runs through individual state licensure or ASLP-IC compact privileges, and the rules are payer-specific too. Never let a model answer “can I see this client across state lines?” — verify with the state boards and the payer.
• Informed consent should mention AI. If tools touch client information or generate client-facing materials, a line in your intake paperwork is cheap protection and honest practice. The policy templates include adaptable language.
• Marketing claims are regulated speech. Generated copy loves outcomes (“proven results!”). The FTC and your ethics code do not.

Worked Example: An Insurance Appeal

“I am an SLP in private practice appealing a denial. The payer’s stated reason: [quote the denial language]. Here is my documentation: [de-identified plan of care, progress data, medical necessity rationale]. Draft an appeal letter that responds point-by-point to the stated denial reason using only the clinical information I provided. Do not invent outcomes, percentages, or clinical details. Professional tone, but keep my clinical voice — do not over-polish.”

What to check: Every clinical claim in the letter must exist in your records — an appeal is a representation to a payer, and an invented “client demonstrates 40% improvement” is indefensible even if the improvement is real but unmeasured. Check that the letter answers their denial reason rather than making the general case for therapy; models drift toward the generic argument.

Red Flags Specific to This Setting

• Pasting client information into consumer AI tools because “it’s just my practice” — solo practices carry full HIPAA obligations
• Marketing copy with outcome claims or testimonials-flavored language you can’t substantiate
• Scope creep: a model will confidently help you into areas outside your competence — it has no scope of practice; you do
• Appeal or necessity language that outruns your actual data

Related

• Prompts: Prior Authorization Letter · Parent-Friendly Summary
• Policy: Private Practice AI Guidelines — adaptable consent and disclosure language
• Foundations: PHI Safety — the de-identification discipline everything above depends on