Build a therapy intake form in two passes: ask ChatGPT to draft the sections from a de-identified prompt with no client details, then check the draft yourself for consent, emergency contact, safety wording and unnecessary questions. Host it in a form tool that will sign a business associate agreement, and have a lawyer review the final version.
The AI is useful for the blank page and for spotting gaps, and nothing more. It does not know your licence conditions, your clients or the rules where you practise, so every line it writes is a suggestion. A bare request such as "write a therapy intake form" tends to come back long, with a diagnosis checklist, a full trauma-history section and no line telling a client in crisis that nobody reads the form in real time. All three are quick to fix, but only if you go looking for them.
Decide what the form is for before opening ChatGPT
An intake form has one job: give you the minimum you need to book a first session safely. It is not a clinical assessment and it is not your whole history-taking. Practices that try to make it both end up with 90-question forms that clients abandon on their phones, and with a store of sensitive detail they now have to protect for years.
Start by writing down, in plain words, four things before you prompt anything. Who completes it (an adult for themselves, a parent for a child, both partners in a couple)? When does it arrive (after an enquiry, or after a first call)? Who reads it, and how soon? And where will the answers sit afterwards? The answers shape the whole template, and they are what you give the AI as context.
A useful rule is to split questions into three bands. Band one is essential before the first session: name, contact details, emergency contact, consent to treatment and to your policies. Band two is helpful but can wait for the first session: presenting concern in a sentence, previous therapy, current medication prescribed by someone else. Band three is better asked in person: trauma history, family detail, substance use specifics. If a question is in band three, leave it off the form.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA prompt that drafts the sections without any client detail
You give the AI a role, a format and your constraints, and you tell it never to ask for client data. Nothing in the prompt is about a real person, so there is nothing identifiable to leak. Here is a prompt I would use in a ChatGPT Project or a Claude Project, so the instructions stay put between sessions.
You are helping a solo private-practice therapist draft a new-client intake form.
Practice type: individual talking therapy for adults, online and in person.
Constraints:
- Ask only what is needed to book and safely start a first session.
- Plain language, reading age about 12, no clinical jargon.
- Maximum 25 questions, grouped under short headings.
- Mark each question Required or Optional.
- Do not include a diagnosis checklist.
- Leave placeholders in [square brackets] for practice name, fees and policy links.
Output: a table with columns Section, Question, Required/Optional, Why we ask.
Then list five questions you decided NOT to include and why.Ask for the "why we ask" column every time. It forces the AI to justify each question, and it gives you a ready-made list of reasons to show a lawyer or a client who asks why you want something. The "five questions left out" instruction is the sneakiest useful part, because it often surfaces a few items you were about to include by habit.
Here is an illustrative sample of what comes back for the first rows. The wording is mine, not a model's guaranteed output, so treat it as a style guide for what to expect and what to edit.
| Section | Question | Required? | Why we ask |
|---|---|---|---|
| About you | Full name and the name you would like me to use | Required | Booking, and respecting how you want to be addressed |
| About you | Best phone number and whether I may leave a voicemail | Required | Reminders, and avoiding accidental disclosure |
| Safety | Emergency contact name, phone and relationship | Required | Used only if I cannot reach you in an emergency |
| Your reasons | In a sentence or two, what would you like help with? | Optional | Helps me prepare; we will talk more in session |
| Health | Name of your doctor, if you want me to know | Optional | Only for coordinating care with your agreement |
What you would fix in this sample: the voicemail question is good, the emergency contact is right, but the doctor question needs a consent line directly beneath it saying you will not contact them without separate permission. The AI will not add that unless you ask. Edit as you read, and keep a note of each change, because the notes become your lawyer pack later.
The sections I would keep, and the reason for each
Different practices end up with different forms, but most sound intake forms share a handful of blocks. The table below is a skeleton you can compare against your AI draft. If the draft has a section missing, ask for it. If it has an extra one you cannot justify, cut it.
| Block | What it collects | Keep it simple by |
|---|---|---|
| Contact | Name, preferred name, phone, email, safe contact method | Asking how it is safe to contact them, not just for details |
| Emergency contact | One person, relationship, phone, permission to call | Making it required, with a stated purpose |
| Consent | Agreement to treatment terms, confidentiality limits, fees, cancellation | Linking to the full documents rather than pasting them in |
| Reason for contact | One or two sentences, optional | A free-text box with a character cap |
| Practicalities | Availability, session format, access needs | Tick boxes, with one "anything else" line |
| Risk screen (optional) | A short set of safety questions | Only if someone reads answers within a stated time |
| Payer details | Who pays, insurer details if relevant | Moving them to a separate billing step |
Notice what is missing: diagnosis history, a symptom checklist, detailed family history. Those belong in the first session, where you can respond to what you hear. A form that collects them just stores sensitive information you may never need before you have met the client.
Consent and emergency contact wording the AI should not write alone
These two blocks carry legal and ethical weight, so treat the AI's text as a first pass only. For each one, give the model your actual policy and ask it to turn that into plain language, rather than asking it to invent the policy.
For the emergency contact, the information the form must communicate is when you would use it, what you would say, and that the client can change it. A prompt like this keeps the AI to your facts:
Rewrite this policy in plain English as a short paragraph for a client form, 60 words maximum, no legal jargon.
Policy: I will contact the emergency contact only if I believe there is a serious risk to the client's safety and I cannot reach them, or if required by law. I will share only what is needed. The client can change the contact at any time by telling me.Check the sample it returns against your actual practice. A common drift is the AI writing "in an emergency" where you meant "if I believe there is a serious risk", which is a wider promise than you intended. Another is dropping "or if required by law". Read every sentence for promises you would not keep.
For consent, resist asking the AI for a confidentiality clause from scratch. Confidentiality limits depend on your profession and the rules where you are licensed, and an invented clause can promise more privacy than you can legally give. Pull the wording from your professional body or your insurer's template, ask the AI only to simplify the language, and then compare the simplified version line by line against the original. If a sentence changed meaning, restore the original wording.
If you already use a separate counselling contract, link it from the form instead of duplicating its terms. The tutorial on a counselling contract template and AI first draft covers that document, and the intake form should simply point to it and ask the client to confirm they have read it.
Run the AI check on your own draft
Once your edited draft exists, use the AI a second time as a reviewer. This is the step most template pages skip, and it is where a model earns its place: it is good at noticing inconsistent wording, missing sections and anything that reads as pushy or confusing. Paste only the blank form, never a completed one.
Act as a careful reviewer of a therapy practice's new-client intake form. Below is the blank form.
Check for, and list separately:
1. Questions that collect more than needed to book a first session.
2. Wording that could be read as a diagnosis, promise or guarantee.
3. Places where consent is assumed rather than asked for.
4. Anything that would be hard to complete on a phone or for someone in distress.
5. Missing items a cautious practice would usually include (emergency contact, how to reach you, what the form is not for).
For each point quote the exact text, then suggest a one-line fix. Do not rewrite the whole form.
[paste blank form]A realistic reply to a draft like the one above might flag three things: the "reasons for contact" box has no line saying that urgent matters should not be written there; the consent checkbox bundles fees, confidentiality and cancellation in one tick; and the doctor question lacks a permission statement. All three are genuine findings. You will also get some noise, such as a suggestion to add a diagnosis list that you deliberately left out. Reject those and move on.
Run the check twice with different assistants if you can, say once in ChatGPT and once in Claude. They tend to disagree at the edges, and the disagreements point to wording that is ambiguous. If both flag the same line, fix it first.
Client-safety steps the template pages leave out
A form that asks about risk creates a duty to look at the answer. Decide before you publish what happens when a client answers yes to a question about harming themselves. If nobody sees the submission until next Tuesday, the form can do more harm than good.
The safest design has four pieces. First, a visible line at the top of the form stating that it is not monitored in real time and that anyone in immediate danger should contact emergency services or a crisis line, with the number for your region added by you. Second, a named person who receives submission alerts and a stated target, such as "I read new forms within one working day". Third, a note of what you will do when a risk answer comes in: a call, a check-in message, or moving the first session forward. Fourth, a decision about whether the form asks at all. If you cannot meet the stated target reliably, leave the risk screen for the first session.
The AI can help write the crisis line but it must not supply the numbers. Insert them yourself from official sources. Models confuse numbers between places, and a wrong crisis number is a serious failure. If an AI receptionist or chatbot also collects first contact for you, set its crisis handling separately; the tutorial on AI receptionists for therapists, with intake, waitlist and crisis rules covers that side.
Where the form lives: a tool that will sign an agreement
Where you host the form matters as much as the questions. A free general form builder may keep answers on servers you have no contract with, and may not offer a business associate agreement (BAA), the document in which a vendor takes on obligations for handling health information on your behalf. If you are subject to health-data rules, ask for that agreement or your region's equivalent before any client answer is collected.
Some options to evaluate, none of them endorsed blindly. Your practice-management software probably has built-in intake forms that sit inside the system where records already live, which is often the simplest route; the tutorial on therapy practice software and what its AI add-ons cost helps you compare. A general form builder is the other route, and there the plan level decides everything. Jotform is a clear example. Its pricing page lists its health-data compliance features as available only on Gold ($129 a month, or $99 a month billed annually) and Enterprise, and not on the free Starter plan, Bronze or Silver. Its FAQ says customers who switch those features on can then sign a BAA, and its healthcare page reminds you that following the rules that apply to you stays your job. So a therapist who built a tidy form on a Silver account would have to upgrade, switch the features on and sign the agreement before the first real answer arrives. Whichever tool you pick, check three things on the vendor's own pages on the day you sign up: whether they will sign an agreement on your plan, where submissions are stored, and how you delete or export them.
Also check what the form tool does with AI features. Several builders now add "AI form generation" or "AI summaries" buttons. Keep those switched off for forms that collect real client answers, unless you have read the vendor's terms and are satisfied. Drafting the blank template in ChatGPT and building it in a vetted tool keeps the model and the real data apart, which is the whole point of the method here.
What to hand a lawyer or professional body
A review is quicker and cheaper if you hand over a tidy pack rather than a wall of text. Send the final form, your list of edits from the AI draft, the practice policies the form links to (fees, cancellation, confidentiality, record keeping), and the answers to four questions: where you are licensed, whom you see (adults, children, couples), how sessions are delivered, and which tool hosts the form.
Ask the reviewer specific questions. Does the consent block create obligations we cannot meet? Is the confidentiality summary accurate for our licence and session types? Is the emergency contact wording enough? Does the form need extra elements for minors or for online sessions? Your insurer or professional body may also publish guidance or a template, and it is worth reading that before you pay for a review.
Keep the reviewed version under a version number and date, for example "Intake form v1.2, reviewed on the date of sign-off", and store it with your policies. When you change a question later, update the version and review the change, however small. This is the habit that keeps an AI-drafted form from drifting.
A sixty-minute build, with numbers
An illustration for a solo therapist seeing about twelve new clients a quarter. She currently emails a PDF, receives photos of filled-in pages and retypes details into her notes. Her time goes roughly like this: ten minutes writing the context and prompt; ten minutes reading and cutting the draft from 31 questions down to 22; fifteen minutes rewriting consent and emergency contact from her own policies; ten minutes running the reviewer prompt and applying the three useful findings; ten minutes building it in a form tool that signs an agreement; five minutes testing on her phone. That is an hour, plus the lawyer's review as a separate, scheduled cost.
The saving shows after the first quarter. If retyping and chasing missing fields currently takes about fifteen minutes per new client, twelve clients cost three hours a quarter, so the form pays back its build time within one intake round. These figures are illustrative, so replace them with your own timings. The larger gain is that every client gets the same safe wording, instead of whatever was in the last PDF you emailed.
Test it like a nervous new client
Before any real client sees the form, submit three test entries yourself. First, a normal one on your phone, noting how long it takes. Second, one that leaves every optional box empty, to confirm required fields are the only blockers. Third, one that deliberately triggers a risk answer, to confirm the alert reaches the person it should, within the time you promised.
Check the confirmation email or page too. It should not repeat what the client wrote, and it should not reveal that they are a therapy client if the email address is shared. Check your own alert email the same way: Jotform's FAQ, for instance, says that with its health-data features on, notification emails carry no submission data and you log in to read answers, which is the behaviour to look for in any tool. Send a test to a second address on a different device. Many small leaks come from a confirmation message addressed to a family inbox.
Finally, offer a non-digital alternative. Some clients cannot use a form on a phone, or do not want sensitive details sent online before they have met you. A line such as "if you would rather give this information by phone or in person, tell me" costs nothing, and it is kinder than turning someone away.
A mistake that is easy to make with a good-looking draft
The most realistic failure is trust in polish. A model produces a clean, confident form with a reassuring consent paragraph, and the owner publishes it after a quick skim. Six months later a client asks what "your information will be kept strictly confidential" means in practice, and the honest answer includes exceptions the form never mentioned. The wording was smooth, and it was wrong for the practice.
The cure is the habit described above: write the policy yourself, let the AI only reword it, compare line by line, and review with a human who knows the rules. A second, smaller mistake is pasting a real client's completed form into a chat window "just to tidy the notes". Keep a standing rule: AI sees blank templates and invented examples only. If you later choose a clinical documentation tool that your rules and vendor agreement allow, that is a different decision, covered in the tutorial on confidentiality checks for AI note tools.
Keeping the form current
Set a reminder to revisit the form every six or twelve months. Fees change, you add telehealth, a new regulation lands, or a client's feedback shows a question is confusing. When that happens, update the policy first, reword with the AI second, then repeat the reviewer prompt and the three test submissions. The whole refresh takes about twenty minutes when the process is already in place.
When a client stops therapy, the same documents keep working: your closing letter can reuse the plain-language approach, and the tutorial on a therapist termination letter and an AI draft shows how to do that without improvising difficult wording.
Questions therapists ask before using AI on intake paperwork
Can I paste a completed intake form into ChatGPT to summarise it?
Not into a consumer account. A completed form is identifiable health information. Summarising real client answers needs a tool your own rules and the vendor's agreement allow, with consent covered in your paperwork. Keep AI to the blank template and de-identified examples.
Does the AI-drafted form need a lawyer if it came from a template?
Yes, for anything that creates obligations: consent, confidentiality limits, emergency contact use and record handling. Rules differ by profession and place, so a local lawyer or your professional body's guidance should sign off the version you actually use.
How long should a therapy intake form be?
Aim for what a new client can finish in 10 to 15 minutes on a phone. Collect contact and safety essentials first, and leave detailed history for the first session, where you can ask follow-up questions in person.
Should the form ask about self-harm?
Some practices include a short risk screen, but only if someone reads answers promptly and the form says it is not monitored in real time. If you cannot commit to a response time, leave the question for the first session and show crisis contacts.
Can I reuse the same AI prompt for couples or child clients?
Reuse the method, not the output. Couples and child clients need different consent, different people signing and different safety questions, so run the prompt again with that context and have each version reviewed.
Further reads
- Automated Booking Systems for Therapists: Which Ones Automate What? — Booking systems that can send the form automatically.
- AI Scribe for Counsellors: A Safe Trial Plan With Consent and Opt-Out — Consent wording for AI note tools, once clients are in.
- Best HIPAA-Compliant AI for Therapists: Which Vendors Sign a BAA? — Which vendors sign a BAA, if you want AI beyond drafting.
- Should Therapists Tell Clients They Use AI? — Whether and how to tell clients you use AI.
- Therapy Cancellation Policy Template: Fees, Wording and AI Reminders — The fee policy your form will link to.
- AI in Psychotherapy Practice: Enquiry Replies and Fee Policies — Enquiry replies that lead into the intake form.
- AI Marketing for Therapists: What to Say and What to Leave Out — Testimonial and outcome-claim rules as the spine: before-and-after examples for a directory profile, website copy and a newsletter drafted with AI.
- Fertility Clinic Intake Form: Digital Version, AI Summary — A fertility clinic intake form built in stages: fields, consent, a clinician-ready AI summary prompt, and the data you should never collect.
- Mental Health Services AI: Intake, Waitlist and Referral Admin Jobs — A checklist of the intake, waitlist and referral admin jobs AI can take on in a mental health service, and the lines it must never cross.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Jotform's pricing page and health-data compliance FAQ (plans with those features, BAA signing, notification emails), OpenAI's Projects help article and Anthropic's Projects help article, all checked 4 November 2026.