Dictation for Psychologists and Allied Health: Session Notes Without an AI Scribe in the Room

Almost all the new speech technology being marketed into healthcare assumes one thing: that you want a machine in the room, listening to the whole appointment and turning it into a note for you. Ambient AI scribes are where the investment and the sales effort have gone, and for a lot of clinicians that trade is worth it. For psychologists, counsellors and a good chunk of allied health, it is not — and the market has been slow to notice that the objection is not reluctance about technology.

If you write session notes for a living and you have quietly decided you do not want an ambient AI scribe recording your clients, you have not opted out of technology. You have opted out of one technology. Dictation is the other one, and it solves a large part of the same problem without ever putting a recorder between you and the person in front of you.

Why are psychologists and therapists reluctant to use AI scribes?

AI scribes have been adopted quickly in general practice and hospital medicine, and there are good reasons for that. A GP consultation is largely a structured information exchange, and turning it into a summary is genuinely useful.

A therapy session is a different kind of conversation. Reporting through 2026 has documented substantial hesitancy among mental health practitioners about introducing recording into the therapeutic relationship — concerns that a third party in the room changes what clients are willing to say, and that a verbatim transcript of a psychotherapy session is a uniquely sensitive artefact to have created and stored somewhere (NPR, May 2026; Proof News).

There is also a practical point that gets less attention. Psychological notes are frequently not meant to be a summary of what was said. They are the clinician’s formulation — impression, risk, hypothesis, plan — written deliberately, and often written to be defensible if they are ever subpoenaed. An ambient tool that summarises the conversation is solving a problem many psychologists do not have, while creating one they would rather avoid.

None of that makes AI scribes wrong. Plenty of allied health practitioners use them happily, particularly in high-volume physical disciplines where the appointment content is objective and repetitive. It just means the question is not “which AI scribe should I use” but “do I want one at all”.

Do I need client consent to use an AI scribe?

If you do decide to use an ambient scribe, there is a body of professional guidance you should read first rather than take second-hand from an article like this one — and it differs by country.

In the United States, the American Psychological Association has published ethical guidance on AI in clinical practice, HIPAA governs how session data is handled, and state law is moving quickly: Texas introduced explicit AI disclosure requirements from January 2026, and a large number of AI-and-mental-health bills are in progress across other state legislatures. In the United Kingdom, the HCPC has published AI guidance, and UK GDPR applies to any recording of a client. In Australia, AHPRA has published guidance on AI in practice and the Australian Clinical Psychology Association has published practice guidance on AI-assisted tools in clinical psychology (ACPA, September 2025). In Canada, PIPEDA and provincial legislation such as PHIPA apply.

Check the current version of whatever applies to you, and check with your own professional association and indemnity insurer. This area is moving fast enough that guidance issued a year ago may already have been superseded.

The themes running through all of it are consistent, wherever you practise: the registered practitioner remains responsible for the accuracy of the record, informed consent needs to be obtained and documented, clients must be free to decline without any effect on their care, and you are expected to understand where the tool sends and stores the data — including whether it crosses a border.

The relevant point for this article is simpler. Every one of those obligations exists because the tool is recording the client. Dictation does not record the client, so the consent, storage and offshore-processing questions do not arise in the same form. That is not a loophole — it is just a narrower tool that never creates the artefact those obligations are written to govern.

What is the difference between dictation, transcription and an AI scribe?

It is worth clearing up, because “speech to text” now covers three quite different products.

Ambient AI scribe. Listens to the whole appointment, then generates a structured note. You review and sign.

Audio file transcription. You record a memo, upload it, get a transcript back later.

Cursor-based dictation. You put your cursor in a field, press a key, and speak. Your words appear as text, punctuated, exactly where the cursor is — in whatever program is in front of you. Nothing is listening unless you press the key. Nothing is summarised. Nothing is interpreted.

Speech Recognition Cloud is the third kind. You write the note; you just write it with your voice instead of your hands. Two things follow from that, and they are the two things this article is really about: it is verbatim, and it needs no integration with anything.

Modern dictation is also nothing like the Dragon-era experience some practitioners remember. There is no voice profile to train, no reading passages aloud for twenty minutes before it works, and no per-application plugin to install. Speak a full paragraph naturally and the punctuated paragraph lands at the cursor — and you control the punctuation, headings and line breaks by voice as you go, which is covered in the spoken punctuation guide. Getting from download to first dictated note takes a couple of minutes; the getting started guide walks through it.

Does dictation software work with my EHR or practice management system?

This is always the first question, and the honest answer is that it is the wrong question — because with cursor-based dictation there is nothing to check.

Mental health and allied health practitioners are spread across dozens of platforms, and the list looks completely different depending on which country you practise in. A rough map of what is in common use, purely descriptively:

United States. SimplePractice and TherapyNotes are widely used by solo and small practices for scheduling, notes, telehealth and billing. Valant is a behavioural health EHR built around measurement-based care and outcome measures. ICANotes offers structured narrative note-building with a large library of built-in clinical rating scales. Ensora Mental Health (formerly TheraNest) provides EHR, scheduling, telehealth and billing for practices and clinics. CounSol and Sessions Health are web-based systems used by independent counsellors. Practitioners inside health systems are typically working in Epic or Oracle Health (Cerner) behavioural health modules instead.

United Kingdom and Ireland. WriteUpp is a UK-built practice management system used across therapy and allied health. Semble is commonly used where private medical insurance billing is involved. Pabau is positioned as a broader clinic operations platform for larger teams. Cliniko is also used widely in the UK.

Canada. Jane App is used across both physical and mental health disciplines. Owl Practice was built specifically for therapists, counsellors, psychologists, social workers and speech-language pathologists, and stores data in Canada.

Australia and New Zealand. Cliniko, Halaxy, Zanda (formerly Power Diary), coreplus, Splose and Nookal.

Those descriptions are deliberately factual. Which platform suits your practice depends on your discipline, your size, your billing model and your jurisdiction, and that is your call to make — not something a dictation vendor should be ranking for you.

The point is what they have in common. Almost all of them are browser-based, and almost none of them will ever appear on a dictation vendor’s certified-integration list. The enterprise dictation products were built around deep integrations with a handful of large hospital systems, so the first thing you traditionally had to do was check whether your platform was supported. For a two-person practice on a browser-based platform, it never was, and it was never going to be.

Cursor-based dictation removes the question entirely. Speech Recognition Cloud types at the cursor at the operating-system level. It does not connect to your EHR, does not need an API, does not install a plugin, and does not need to know what program is in front of it. If you can click into a text field on Windows and type into it, you can dictate into it.

One honest limitation: Speech Recognition Cloud is Windows software. If your practice runs entirely on Mac or iPad, this is not the tool for you.

That has three consequences worth spelling out.

No integration, so no compatibility list. You are not waiting for your vendor and your dictation vendor to build something together. Nothing has to be certified. Nothing has to be approved by anyone’s IT department. You install it and it works in what you already use.

No lock-in, in either direction. Because there is no integration, switching practice management software does not break your dictation, and switching dictation does not touch your clinical records. Practices that have been burned by a platform migration understand exactly what that is worth. Your dictation licence is not a hostage to your EHR decision, and your EHR decision is not constrained by your dictation.

One tool for everything, not just the EHR. The same hotkey works in your practice management notes, a Microsoft Word assessment report, an Outlook referral letter, a funder or insurer progress report, a PDF form, a payer portal, a court report template, and your email. Clinicians consistently find the report writing is where the hours actually disappear — and that work happens outside the EHR, where an EHR-integrated tool cannot follow.

Why does verbatim matter in clinical and therapy notes?

This is the part that gets underrated until the first time it bites you.

Generative AI does not transcribe. It predicts. An ambient scribe or an AI-summarised note is producing a plausible version of the encounter — tidied, restructured, sometimes rephrased into language you would not have chosen, and occasionally containing something you never said at all. That is not a bug in a particular product; it is what generative models do. It is precisely why the guidance in every jurisdiction insists that the clinician reviews and corrects every AI-generated note before signing it.

Which means the time an AI scribe saves you in writing, it partly hands back in proofreading — and proofreading someone else’s plausible prose for things that are subtly wrong is slower and more cognitively demanding than most people expect.

Verbatim dictation does not have that failure mode. You say the sentence, the sentence appears. Punctuated, formatted, at the cursor — but your words, in your order, in your clinical language. Nothing is interpreted, nothing is summarised, nothing is invented.

For a lot of allied health documentation that precision is not a nice-to-have:

• Risk documentation. What was assessed, what was said, what was decided. This has to be exact, and it has to be yours.
• Direct quotes from the client. A paraphrase is not a quote, and an AI-smoothed paraphrase presented as a quote is a real problem.
• Medico-legal and insurer reports. Written to be read adversarially. Every sentence is a sentence you may have to stand behind.
• Court and tribunal reports. Same, with higher stakes.
• Standardised assessment language. Instrument names, subscale scores and scoring conventions have to be reproduced precisely, not approximated.

And because it is verbatim, there is no second question about whether the note reflects your clinical reasoning. It does, because you dictated it.

The audio side matches. Speech is transcribed and the audio is then erased — it is not retained, not stored, and not used to build a profile. Combined with the fact that nothing is recorded during the session in the first place, there is simply no audio artefact of your client’s appointment sitting anywhere.

Which allied health professions get the most out of dictation?

Psychologists and counsellors. Session notes and case formulation written after the client leaves, plus the long-form work — assessment reports, court and insurer reports, treatment summaries, letters back to referring GPs. The reports are where the hours actually go, and they are the part an ambient scribe cannot help with at all.

Physiotherapists, chiropractors and osteopaths. Objective findings, treatment applied, plan. Highly repetitive language, which makes text replacements and templates unusually effective.

Occupational therapists. Functional capacity assessments and disability funding reports — NDIS reports in Australia, similar assessments for insurers, workers’ compensation schemes and disability programmes elsewhere — are long, structured, and brutal to type. This is arguably the single strongest use case in allied health.

Speech pathologists and dietitians. Assessment write-ups, goal setting, progress reporting against plans.

Social workers and case managers. Case notes, file notes, reports to agencies — usually written in volume, usually written late.

Can dictation handle clinical terminology, drug names and assessment instruments?

General-purpose dictation stumbles on the words that matter most in a clinical note: drug names, diagnostic terminology, anatomical language, assessment instrument names. The Medical Ultra plan adds continuously updated medical vocabularies and an Ultra Accuracy medical mode for exactly this reason.

If your notes are heavy on psychometric instruments, medication names or specialist terminology, that vocabulary is the difference between dictation saving you time and dictation costing you time in corrections. Custom vocabulary lets you add the terms specific to your practice — including the acronyms and shorthand your discipline uses that no general model will know — and text replacements let a short spoken trigger expand into a full standard paragraph.

Deliberately, the Medical Ultra plan excludes the AI Modes available on other tiers. AI Modes send text to an AI model for rewriting or summarising, which is an unnecessary data flow when the text is clinical — and it is the one route by which generated wording could get into a note. On the Medical Ultra plan that pathway is closed, which is what keeps the output verbatim end to end.

What can dictation not do?

Dictation is not a scribe and does not pretend to be. It will not:

• Sit in the session and write the note for you
• Summarise, structure or format a note into SOAP or DAP on its own
• Extract goals, risks or actions from a conversation
• Remove the need to think about what the note should say

If what you want is to walk out of a session with a draft note already written, an ambient scribe is the product that does that, and you should use one — with consent, and with review.

Dictation is for practitioners who intend to write the note themselves and want the writing to take a fraction of the time. It is designed to assist documentation workflows, not to make clinical judgements. It is not a medical device and it does not provide clinical advice.

How much does dictation software cost for a therapist or allied health practitioner?

The Medical Ultra plan is US$439 per year, flat, per user — one annual licence rather than a monthly per-seat subscription, with no lock-in contract. It includes the medical vocabularies, Ultra Accuracy medical mode, dedicated remote support and one-on-one training, and the mobile companion app that turns your phone into a wireless microphone. It is English only.

Practitioners who do not handle clinical information — practice managers, admin staff, or clinicians dictating non-clinical correspondence — may be better served by the Professional plan at US$159 per year, which includes AI Modes.

There is also a free Starter tier with 20 minutes a month and no credit card required, which is enough to find out whether dictation suits how you work before you spend anything. You can compare plans and pricing directly.

Pricing accurate at time of publication. Prices in USD and will convert to your local currency at checkout. Verify current pricing before purchasing.

How do I start using dictation without disrupting my week?

The realistic way in is not to convert your whole workflow on a Monday morning.

1. Install it and dictate three or four session notes at the end of one day. Notes are short and forgiving. The getting started guide covers microphone setup.
2. Add your ten most-used clinical terms to custom vocabulary. This is the step people skip, and it is the step that makes the difference.
3. Once notes feel natural, use it on one long report. That is where you will actually feel the time come back.
4. Set up text replacements for the phrases you type identically every single time.
5. If anything behaves unexpectedly, the troubleshooting guide and FAQ cover the common causes.

Most practitioners find the adjustment is not technical — it is learning to speak in finished sentences rather than composing as you type. That takes a few days.

Should I use dictation or an AI scribe?

Work through the three questions in order, because they narrow quickly.

One: do you actually want something listening to your sessions? If the answer is yes — and for many clinicians it legitimately is — use an ambient scribe, with consent and with review. If the answer is no, you have not opted out of technology. You have ruled out one category.

Two: does your documentation have to be verbatim? Formulations, risk notes, direct client quotes, insurer reports and court reports all have to say precisely what you meant, in your language, because you may have to stand behind every line. Generated prose cannot give you that, however good it is — it can only give you a plausible version you then have to police.

Three: does it need to type directly into whatever you already use? Not into one certified EHR. Into your practice management notes and Word and Outlook and PDF forms and payer portals — because that is where the long documents actually get written.

Very few products satisfy all three at once. Enterprise clinical dictation gives you verbatim but ties you to a supported-systems list. Ambient scribes give you convenience but record the session and generate the prose. Speech Recognition Cloud is built for the intersection: verbatim, direct into any Windows application, and nothing listening to the room.

So here is the ask: install the free version and dictate four session notes with it. That is a twenty-minute test, it costs nothing, and it will tell you more than any comparison table.

Download Speech Recognition Cloud free →  No credit card, no voice training, installs in about two minutes.

If you want more detail before you install, the dictation for clinicians page covers clinical workflows in depth, dictation for healthcare teams covers multi-practitioner settings, and plans and pricing sets out what each tier includes. If you are currently weighing up Dragon, the Dragon Medical One comparison covers that decision directly, and Australian clinicians can find local pricing, trial and support information for dragon medical one itself. And if you would rather ask a person, get in touch.

Who wrote this?

I’m Russell Bewsell, and I built Speech Recognition Cloud.

I have spent 28 years deploying speech recognition — across hospitals, GP and specialist practices, allied health, courts and government departments, with customers in Australia, the US, UK, Canada, New Zealand and across Europe. I ran what became the largest Dragon reseller in the Southern Hemisphere, supporting more than 40,000 customers and over 15,000 Dragon users, and in 2017 I presented the national Dragonology clinical training series on behalf of Nuance.

Most of that time was spent fixing other people’s dictation software for clinicians. SRC is the simpler, cheaper tool I wished I could hand people instead.

So the argument above is not a marketing brief. It is what I have watched actually happen in practices: the people who write formulations and long reports want their own words on the page, and a lot of them do not want a recorder in the room. If your setting does not fit the patterns I have described, get in touch and I will tell you honestly whether dictation is likely to suit your work.

— Russell Bewsell, Founder, Speech Recognition Cloud

Sources: NPR — Mental health therapists who use AI to take notes face questions about trust · Proof News — Why AI scribes, widely embraced by doctors, spook therapists · ACPA Practice Guidance: Use of AI-Assisted Tools in Clinical Psychology