A week-by-week plan for therapists who want to switch to AI therapy notes without a risky big-bang change — audit, consent and BAA in week one, parallel drafts in week two, switch primary in week three, measure and tidy retention in week four. Printable checklists included.
Switching from paper notes to AI-assisted documentation works best as a 30-day sequence, not a single switch-over day. Week 1: audit your current notes, update consent, confirm a BAA, and pick one template. Week 2: run AI drafts in parallel with paper for five sessions. Week 3: make AI the primary method with a short review checklist on every note. Week 4: measure time saved and tidy up retention and backups. Each week below ends with a checklist you can print.
Most clinicians who go back to paper after trying AI notes did not fail because of the tool. They tried to change the note format, the consent language, the vendor, and their after-session habit all on the same Monday; something went wrong, it was hard to tell which change caused it, and paper felt safer.
A staged plan separates those decisions. You settle compliance before touching client data, prove the tool on a handful of sessions before you depend on it, and only then measure whether it saved time. If any week fails its checklist, you pause there.
This plan assumes a chat-based drafting tool like PsyFiGPT, where you describe the session in a few sentences and get an editable structured draft. Practices using ambient recording follow the same weeks with a few extra consent steps, which we cover in is ambient AI transcription HIPAA-safe.
The first week involves no client data at all. Its job is to make sure that when you do start, nothing is missing.
Audit what you write now. Pull your last ten paper notes and answer three questions: which format are you actually using (SOAP, DAP, BIRP, or something looser), how long does each note take from sitting down to done, and what do you reliably leave out under time pressure? That last one matters most. If your paper notes routinely miss a documented risk assessment or a plan tied to treatment goals, an AI draft will not fix that unless you tell it to include those sections. Write down the average minutes per note; you will need it in week four.
Update informed consent. Add a short paragraph naming the type of tool, what information it processes, that a clinician reviews every draft before it becomes part of the record, and how a client can opt out. Our guide on how to talk to clients about AI in your practice has sample language. Check your state board rules — several states have specific disclosure requirements, and this post is not legal advice.
Confirm the BAA. Before any real client information goes into any AI tool, you need a signed Business Associate Agreement with the vendor. PsyFi executes its BAA by clickwrap when you create your account, and it covers every plan rather than being an enterprise upsell. If you are evaluating another tool, ask for the BAA in writing and read what it says about de-identification, training use, and sub-processors. "HIPAA-aligned, BAA available" is the standard to look for; there is no such thing as a HIPAA certification, so treat any vendor claiming one with caution.
Pick one template. Resist the urge to redesign your notes. Choose the format closest to what you already write and stick with it for the full 30 days. If you are undecided, SOAP vs. DAP vs. BIRP walks through the tradeoffs. In PsyFiGPT, save your chosen format as a prompt in your library so every session starts from the same request.
Week 1 checklist
Now you use the tool, but paper stays the record of truth. The goal is to learn what a good input looks like and where the drafts fall short, with nothing at stake.
Pick five sessions across the week, ideally a mix of clients and session types. After each one, write your paper note as you normally would. Then open the chat tool and describe the session in three or four sentences: presenting concern, interventions, client response, and plan. Ask for a draft in your chosen format and compare the two.
A concrete example. After a 50-minute CBT session with an adult client working on panic symptoms, a clinician types:
"Adult client, ongoing panic disorder treatment, session 8. Reported two panic episodes this week, both at work, shorter than previous. Reviewed thought records; practiced interoceptive exposure in session, tolerated well. Client rated distress 6 down to 3. No safety concerns. Plan: continue exposure hierarchy at home, review next week. DAP format."
PsyFiGPT de-identifies the text before it reaches the AI model, then returns a structured Data / Assessment / Plan draft in seconds. The clinician compares it with the handwritten note and notices the draft phrased the assessment more clearly but left the distress rating out of the Data section. That is useful information: next time, the input says "include ratings in Data." By session five, most clinicians have a repeatable two-minute input that produces a draft needing only light edits.
The test this week is not whether the draft is perfect, but whether draft plus edits is at least as complete as your paper note in less total time.
Week 2 checklist
If week two's drafts held up, AI-assisted documentation becomes the primary method this week. Paper is no longer written first; the AI draft is. But a draft only becomes a note after you review it, and the review needs to be deliberate rather than a glance.
Keep a short checklist beside you for every note and go through it before you copy the note into your EHR or export it. It should take under a minute:
Two other things to settle this week. First, note the switch date in your practice records so anyone auditing later understands why the format changed, and keep a paper fallback for the rare session where you cannot access the tool. Second, handle the client who opts out: your consent form promised manual notes as an option, so make sure that path still works when you are busy.
If a draft is bad enough that editing takes longer than writing from scratch, write from scratch. The plan is for AI to be the default, not a rule.
Week 3 checklist
The last week answers the question you started with: did this help? It also closes the loose ends that make an audit painful later.
Measure. Compare the average minutes per note from week one's audit with this week's numbers, counting input time, editing, and the review checklist. A common result for chat-based drafting is that the writing itself drops sharply while review stays constant, so the total might be half of what paper took rather than a tenth. That is still hours per month. Compare that against what you pay; our pricing page lists PsyFiGPT at $15/month for the base plan.
Tidy retention. Your old paper notes stay subject to your state's retention period; switching changes nothing there. Decide whether they stay in current storage or get scanned into the EHR, and record the decision. For the new workflow, your EHR should remain the chart of record, and check whether the AI tool retains chat history you should periodically clear.
Check backups. Paper's one advantage was that it did not depend on a login. Make sure your EHR export or backup covers this month's notes and that you can find a specific client's note without opening the AI tool.
Decide. With four weeks of data, either continue, extend the trial to a second template or a second clinician, or go back to paper knowing exactly why.
Week 4 checklist
If you continue, the next step is usually widening: add a second template for intakes or treatment plans, or bring a second clinician through the same four weeks. In a group practice the consent and BAA work from week one transfers directly; only weeks two and three need repeating per clinician.
If you want to run the plan with a real tool, PsyFiGPT starts with a 7-day free trial and no card required, which covers the week one setup. If you create the account on day one, the trial ends as week two begins, so either create it late in week one or expect to start a paid plan before the parallel-drafting sessions.
