AI for Psych Providers: Telehealth Consent Forms and Reminders

AI can draft a telehealth consent form for a psychiatric practice, but only as a first draft. The prescriber who runs the service must review the clinical content, and a lawyer who knows your licensing board's rules must approve the final wording before any patient signs it.

The reason is simple. A model writes confident, tidy paragraphs about consent, recording, emergencies and privacy without knowing where your patients sit, which board licenses you, or what your insurer demands. The useful job for AI is the scaffold: a complete list of sections, plain-language patient wording and consistent reminder texts. The accountable decisions stay with named people, and those names are written down before the first prompt is typed.

Follow me on Instagram@sagnikteaches

The review trail: who looks at what, in order

Start with the review, not the draft, because the review decides whether the draft is any use. Write down four roles and put a name against each before you open an AI tool. In a small practice one person may fill two roles, but the lawyer is always a separate step.

Connect on LinkedInSagnik Bhattacharya

The order matters. Do not send a draft to a lawyer before the clinical review, or you pay for their time on text that changes. And do not let a patient see anything before step 4 is complete. If the lawyer's changes alter the clinical meaning, the form goes back to step 2.

Subscribe on YouTube@sagnikteaches

Step 1: write the fact sheet the model works from

A model drafts better from facts than from instructions to "write a telehealth consent form". Prepare one page of facts about the practice. Keep it free of patient information; it describes the service only.

Practice: independent psychiatric practice, 3 prescribers, 1 practice manager
Services by video: initial evaluation (60 min), medication follow-up (20-30 min), therapy sessions (45 min)
Platform: [video platform name]; patient joins by a link; no app install
Hours: Mon-Fri 8-6; after-hours: patients told to call [crisis line] or emergency services; the practice does not monitor messages after hours
Emergency process: at the start of each session the clinician confirms the patient's physical address and a phone number; an emergency contact is held on file
Recording: the practice does not record video sessions; AI note-taking is not used [or: is used, see policy]
Records: kept in [EHR name], retained according to the practice's retention policy
Fees: [fee schedule reference]; cancellation notice 24 hours
Reminders: text and email, 48 hours and 2 hours before sessions, with consent
Age: adults only; minors seen only with a parent or guardian, under a separate form

The bracketed items are decisions you must make, not facts the AI may fill. Every blank you leave is a place where a model will otherwise invent something plausible.

Step 2: prompt for a draft with every judgement flagged

Draft a telehealth informed-consent form for a psychiatric practice, in words a 12-year-old could follow, using ONLY the fact sheet below.
Sections: what telehealth is; benefits and limits; technology requirements; privacy and security; the patient's location at the start of each session; emergencies and the emergency contact; recording; fees and cancellation; reminders and how we contact you; how to withdraw consent; signature block.
Rules: do not invent facts. Wherever a legal, licensing, clinical or insurance decision is needed, write [LAWYER] or [CLINICIAN] in capitals instead of a rule. Do not cite any law or regulation. Do not promise confidentiality beyond what the fact sheet says. Short paragraphs, plain words.
After the form, list every place where you used a placeholder.

[paste the fact sheet]

An illustrative output is a form of about 900 words with a "Where you are during a session" section that begins "At the start of every session, your clinician will ask where you are physically sitting and a phone number you can be reached on" and ends with "[LAWYER: state any rule about patients located outside the area where the clinician is licensed]". The placeholder list at the end may have a dozen items: recording consent rule, minors, capacity, jurisdiction of the clinician's licence, data breach wording, record retention period, fee disclosure, complaints route, cross-border patients, interpreter use, and how consent is withdrawn.

That list is the most valuable part of the output. It is the lawyer's agenda, written for you in a minute. Hand it over with the draft and your fact sheet, so their time goes on judgement and not on reading generic text.

Step 3: the location check, in the form and in the room

The patient's physical location at the start of a session matters for licensing, for emergencies and sometimes for consent rules. Many practices therefore build the check into the form and into the opening of every session, not just the first. Treat the two as separate parts: a sentence in the form that explains why you ask, and a script for the clinician.

Ask AI for the script, not the policy. A prompt such as "Write a 30-second script a clinician reads at the start of every telehealth session to confirm the patient's location, a phone number and that they are in a private place, in plain, calm wording that does not feel like an interrogation" yields something like: "Before we start, can you tell me where you are sitting right now, including the town, and a number I can reach you on if we get cut off? Are you somewhere private?" Keep it. What you must not let the model do is decide what happens if the patient is outside the area your licence covers. That is a lawyer and board question, and the answer may be to reschedule, to decline, or to require a different arrangement. Write the lawyer's answer into the form and the script after review.

Emergency contact and what the clinician does

The form should say who is held as an emergency contact, when the clinician may use it, and what the patient must do in a crisis. A good form is plain on the last point: "This service is not for emergencies. If you are in danger or may harm yourself or others, call [emergency number] or go to the nearest emergency department." The prescriber's step 2 review matters most here. Read the emergency wording aloud and ask whether it matches what your clinicians would actually do. If the form promises that "we will check on you", and nobody monitors messages after hours, the form is promising something the practice cannot deliver.

Step 4: the recording policy, stated without hedging

Recording is the clause patients notice. Decide the practice's position first, then write it plainly. The options are that sessions are never recorded, that they are recorded in defined cases with specific consent, or that AI note-taking is used with its own consent. Each choice has different legal and clinical consequences, and recording rules differ widely between places, so the position is yours and your lawyer's.

A model can write the patient-facing wording once you state the position. For a no-recording policy: "We do not record video sessions. We ask that you do not record them either, unless we agree it first in writing." For a practice that uses an AI scribe, the consent is a separate matter altogether: the form must say what is recorded or transcribed, who processes it, whether it is kept, and that the patient may decline without affecting care. See telling patients you use AI note-taking: consent wording that works and AI scribe consent form: wording and recording laws for those questions, and can therapists use ChatGPT for session notes if you are tempted to put session content into a general chat tool. The short answer to that last one is to take care and keep it out.

The review trail run once: minutes, changes and sign-offs

Say a three-prescriber practice has an old one-page telehealth form from a platform vendor. The practice manager builds the fact sheet in 40 minutes and runs the prompt. The draft arrives in two minutes, along with a list of 12 placeholders. The numbers below are illustrative.

• Clinical review (step 2), 25 minutes. A prescriber reads the draft and changes three things: the emergency paragraph (the model had suggested the practice would "monitor messages during business hours"; the practice does not), a line on who may sit in on a session, and a sentence on what happens if a patient appears intoxicated or in distress. She adds her initials.
• Operations review (step 3), 15 minutes. The practice manager checks platform name, cancellation notice, fees and reminder timing against how the practice really runs, and finds that the fact sheet itself was wrong: the practice applies 48 hours' cancellation notice, not 24. She corrects both the sheet and the draft, which is why the fact sheet carries a date and a version too.
• Legal review (step 4). The lawyer receives the draft, the fact sheet and the placeholder list. Their time goes on the 12 flagged items, plus a read-through. The practice records the approval, version 1.0, and the date.
• Total internal time: about 90 minutes, against the best part of a day that drafting from a blank page can take, and the lawyer sees a better-organised document.

The practice then stores the approved form as the only version, retires the old one, and records the date from which all new patients use it. Existing patients are asked to re-consent at their next appointment, a task the reception team can track on a simple list.

Reminder texts that disclose nothing they should not

Reminders are the most repetitive part of telehealth admin, and the part where the wording matters because a message may appear on a lock screen. Start with the principle: say the minimum. Date, time, a link, a number to call, and nothing about the reason for the visit, the clinician's specialty or the practice's focus if a glance could reveal sensitive information.

Write reminder messages for a psychiatric practice's video appointments.
Three messages: (1) 48 hours before, (2) 2 hours before with the join link, (3) a missed-appointment follow-up.
Constraints: no mention of psychiatry, medication, diagnosis, therapy or mental health; use the practice's short name "[practice short name]"; each under 250 characters; start "Hi {first_name},"; include the number [phone]; message 3 must be kind and offer to rebook, with no mention of fees.
Add "Reply STOP to opt out" to each. Do not give any advice.

An illustrative message 2: "Hi {first_name}, your appointment at [practice short name] starts at {time} today. Join here: {link}. Need to change it? Call [phone]. Reply STOP to opt out." Message 3: "Hi {first_name}, we missed you today at [practice short name]. We'd be glad to find another time: [phone]. Reply STOP to opt out." What you fix: models love to add warmth such as "Looking after your mental health is so important", which is exactly the sort of line you do not want on a lock screen. Delete anything that reveals the nature of the care.

Consent for reminders belongs in the form, with choices: text, email, phone or none, and the patient's preferred number. Record the choice in the patient record as a field the reminder system reads, so that a patient who said "no texts" is never sent one. A missed-appointment message in this field of care has a particular duty of care: a no-show may be a signal. Decide with your clinicians whether a missed appointment by certain patients triggers a clinician call rather than an automatic text, and write that rule in the practice's procedures. For general reminder timing, how a clinic can use AI to cut missed appointments covers the basics, though the sensitivity here calls for more restraint.

Rewriting an absolute privacy promise into an honest clause

Before (illustrative AI first draft): "Telehealth is completely safe and confidential, and your information will never be shared with anyone. You agree that you are fully responsible for the security of your own device and internet connection."

After (after clinical and legal review): "We use [platform name], which is designed to protect video sessions. No system is perfectly secure, and we can't control the security of your device or connection. Please join from a private place and a device you trust. We share your information only as described in our privacy notice and where the law requires it."

The first version promises absolutes ("completely safe", "never shared") that no practice can keep, and it shifts all responsibility to the patient, a clause a lawyer is likely to strike. The second says what the practice does, what it cannot control and what the patient can do. The edit came from a person; the AI draft supplied the slot the sentence fits into.

Capacity, minors and other people in the room

An adult consent form assumes an adult with capacity, speaking for themselves, alone in a private room. Real telehealth practice is messier, and the form should say how the practice handles each departure. Ask AI for a list of situations to decide, not for rules.

List the situations where a standard adult telehealth consent form for a psychiatric practice would not be enough, with a one-line question for the clinician and lawyer to answer for each. Do not answer the questions. Cover: patients under 18, patients whose capacity is in doubt, a family member or carer joining, an interpreter joining, someone else in the room, a patient joining from a car or a public place, a patient who changes location mid-session, and a session that must be ended early for safety.

An illustrative output has eight to ten items, each phrased as a question: "If a patient joins from a car, does the clinician continue, and what do they say?" and "Who may join a session at the patient's request, and is that person's own consent needed?" Take that list to your clinical meeting. Each answer becomes a sentence in the form, a line in the clinician's opening script or a rule in the procedures manual. Some will be one-liners; others, such as capacity and minors, will need their own form and their own legal review, so mark those "separate document" and do not stretch the adult form to cover them.

A practice that does this exercise once usually finds two or three situations nobody had decided: for instance, whether a clinician continues when a patient is driving, and what the clinician does if a second person walks in. Writing those decisions down protects clinicians as much as patients, because they no longer improvise under pressure.

Keeping the form alive: versions, re-consent and audit

A consent form is a controlled document. Give it a version number and an owner, and keep a short change log: date, what changed, who approved it. Store the approved PDF in one place and have the portal and reception scripts point to it, so nobody emails an older copy from a desktop folder.

Decide when a change triggers re-consent. A new platform, a new recording or AI note-taking policy, a change in who can see records or a change in your licensing position usually does. A typo does not. Write the triggers into the change log template so the person making an edit has to answer the question. When re-consent is required, add a tick-box field to each patient's record and let the reminder system prompt for it at the next booking, so it happens through normal flow and not a mass mailing.

Once a quarter, pull ten patient records at random and check three things: the signed form is present, its version is the current one, and the location and emergency contact fields were completed in the last session. Missing items go on a gap list with an owner and a date. This is the same discipline as any documentation audit, and the practice manager can run it in about half an hour. If something in the check surprises you, update the fact sheet and rerun the draft prompt so the next version starts from the truth.

Five ways an AI-drafted consent form goes wrong

A draft goes live unreviewed. The practice manager is busy, the form looks professional, and it is uploaded to the portal. Stop this with one rule: the portal accepts only a file whose filename includes the approved version number, and only the practice manager can upload it.

Borrowed legal language. A model asked for "compliant wording" often produces authoritative phrases that sound like statute. Tell it not to cite any law, as the prompt above does, and treat any citation that slips through as wrong until your lawyer confirms it.

Forms drift from practice. The form says reminders go at 48 hours and 2 hours; the system has been changed to 24 hours only. Review the form every six months against the fact sheet and after any platform change.

One form for everyone. Minors, patients lacking capacity, couples and group sessions all need their own wording and their own legal review. Do not stretch the adult form to cover them.

Patient data in the drafting tool. The drafting task needs no patient information at all. If anyone pastes in a real patient's details, to personalise a letter or to "see how it reads", stop and review. Where AI touches records more directly, AI note-taking tools for therapists: a confidentiality checklist sets out the questions to ask, and should therapists tell clients they use AI covers the disclosure side.

A checklist before the form goes live

1. Fact sheet written and dated, with no blanks left unresolved.
2. Placeholder list from the draft cleared: every item resolved by a clinician or lawyer.
3. Four sign-offs recorded with names and dates, and a version number on the form.
4. Location script agreed and added to the clinician's opening routine.
5. Emergency wording read aloud by a prescriber and matched to real practice.
6. Recording position stated plainly; AI note-taking, if used, has its own consent.
7. Reminder messages checked for lock-screen discretion; patient channel choice stored as a field.
8. Old versions retired; re-consent plan for existing patients; review date set.

Run through it once with the people named, and you will have a form that is clearer than most vendor templates and a record of who approved it. If you would like to map the rest of the practice's consent and reminder steps with someone, that is what my AI implementation consultation covers, through a 1:1 call.

Questions about AI-drafted telehealth consent

Can I use an AI-drafted consent form as it is?

No. Treat it as a first draft that a lawyer who knows your licensing rules and your own board's requirements must review before any patient sees it. The model does not know your jurisdiction, your insurer's conditions or your professional board's current position.

Should a prescriber's own name appear on the review?

Yes. Name the clinician who reviewed the clinical content, the person who checked it against your policies, and the lawyer who approved it, with dates. A review trail shows the form was considered, and it tells staff who to ask when a question comes up.

What should the form say about recording?

State plainly whether sessions are recorded, by whom, why, how long recordings are kept and who can see them, and that patients and clinicians must not record without agreement. Your lawyer decides the exact wording because recording consent rules differ between places.

Can reminder texts include the reason for the appointment?

Keep them minimal: date, time, a link and a contact number. Leave out the clinic's specialty if a message could reveal sensitive information on a lock screen, and offer patients a choice of channel when they consent.

Further reads

• Best HIPAA-Compliant AI for Therapists: Which Vendors Sign a BAA? — Which AI vendors sign a BAA for therapy practices.
• AI Scribe Security: Audio, Retention and BAA Vendor Questions — Security questions to put to any scribe vendor.
• Best AI Note Taker for Therapists: Recording or Dictation Style? — Recording versus dictation for therapy notes.
• Dental Social Media Consent Form: Patient Photo Wording With AI — A consent form built with AI in another health setting.
• Best AI Software for Therapists in a Group Practice: Seat Costs — Software seat costs for group mental-health practices.
• AI Marketing Automation for Doctors: Recall, Reminder and Review Flows — Recall and reminder flows for medical practices.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: general telehealth consent practice and vendor data-controls pages. Your licensing board, professional body, insurer and a qualified lawyer decide what the form must contain; none of this is legal advice.