Best AI Note Taker for Therapists (2026): An Honest Comparison

Two years ago, "AI notes" meant one or two startups and a lot of skepticism. Now every EHR has an AI add-on, a dozen standalone scribes want $30–120 a month, and the marketing all sounds identical: record the session, get a note, go home earlier.

I build one of these products, so read me with the appropriate salt. But I'm also a clinical psychologist who spent years writing notes at 9 p.m., and I think the honest map of this market is genuinely useful — because the products are not identical, and the differences that matter are not the ones in the ads.

The Three Shapes of AI Note Taker

1. Standalone scribes (Mentalyc, Freed, Upheal, and friends)

These do one thing — turn sessions into notes — and the best of them do it well. Mentalyc and Upheal are therapy-specific; Freed came from general medicine and expanded. Pricing runs roughly $15 to over $100 a month depending on volume and features, on top of whatever you already pay for your EHR.

The structural catch isn't quality. It's that the note is born in the wrong place. You record in one app, the note appears in that app, and then it needs to get into your EHR — usually by copy-paste, per session, forever. That's five minutes of ferrying per note, a second system with PHI in it to vet and pay for, and a documentation trail split across two vendors. Some have integrations with the big EHRs; most therapists I talk to are still pasting.

2. EHR add-ons (the "+$30/month" checkbox)

SimplePractice, TherapyNotes, and Sessions Health all now sell AI note-taking as a paid add-on, typically in the $25–40/month range on top of your subscription. The integration problem is solved — the note lands where it lives — and for many practices that convenience is decisive.

Two things to check before you pay: whether the add-on's AI actually understands therapy sessions (some are general-medical engines with a mental-health template bolted on), and what your all-in monthly cost becomes. A $69 EHR plus a $35 AI add-on is a $104/month documentation stack — real money for a solo practice, and worth comparing against alternatives on total cost rather than sticker price.

3. AI-native platforms (where the scribe IS the EHR)

The newest shape: platforms built around AI documentation from day one, where recording, transcript, note, and chart are one system. This is what Practice Harbor is — AI notes aren't an add-on with a separate price; they're how the product works, included at $19/month for licensed clinicians and free for pre-licensed. The honest trade-off of the category: younger platforms, shorter track records, and you should evaluate the EHR fundamentals (billing, scheduling, portal) as skeptically as the AI. An AI-native platform with weak claims processing is a bad deal no matter how good the notes are.

How to Actually Choose

Whichever shape you're drawn to, the questions that separate the serious products from the demos are the same. (I wrote a full guide to the compliance side — HIPAA-compliant AI notes: 7 questions to ask — but here's the short version, plus the clinical ones.)

• The compliance floor. Signed BAA covering every AI subprocessor; contractual no-training on your data; clarity about what the model actually sees. Bonus points for de-identification before AI processing — the architecture we chose, and the strongest answer to "what does the model see."
• Therapy fluency. Ask for a sample note from a therapy session — not a primary-care visit. Does it capture process, affect, and interventions, or does it read like a discharge summary? Does it know what a mental status exam is? General-medical scribes routinely produce notes that are accurate and clinically useless.
• Your formats, your voice. SOAP, DAP, BIRP, intake vs follow-up — and how easy it is to edit before signing. The draft should feel like a competent intern wrote it in your structure, not like a form letter.
• The workflow seam. Count the clicks from "session ends" to "signed note in the chart." Standalone scribes lose here; add-ons and natives win. Five minutes of pasting per session is 20+ hours a year.
• Who signs. Every serious product treats AI output as a draft for clinician review and signature, with a locked-note audit trail after. If the pitch implies you can skip the review, run.
• Total monthly cost. EHR + AI, together. The market spans free (our pre-licensed tier) to $150+/month stacks. Most of that range buys roughly the same hour of your evening back; the difference is what else comes with it.

The Honest Bottom Line

If you love your current EHR and it has a competent therapy-aware add-on: buy the add-on. The integration is worth more than a marginally better note. If your EHR's AI is a general-medical engine, a standalone therapy scribe like Mentalyc or Upheal will write you a better note — and you'll pay for it twice: in dollars and in pasting.

And if you're choosing an EHR anyway — starting out, or fed up with your current one — it's worth looking at the AI-native shape before you commit to a stack of subscriptions. That's the bet we made with Practice Harbor: notes drafted from the session itself, de-identified before AI ever processes them, reviewed and signed by you, in the same system that runs your scheduling, billing, and portal. One subscription, one chart, no ferrying. (How the pipeline actually works, including why it won't replace your clinical voice, is its own article: How AI notes actually work.)

Albert Wong, PhD, is a clinical psychologist and the founder of Practice Harbor.