Dictation for Nurses, Therapists and Allied Health

Nurses and therapists can dictate most of the text they write, but the parts that take the most time are the parts that dictation does least well. Nurses chart a large share of each shift in structured flowsheets, which are grids of tick boxes and short fields. Therapists usually write a narrative note after a session, which is the easiest kind of text to speak. The evidence for both groups is much smaller than the evidence for physicians, and this guide separates what has been measured from what is only assumed. For the wider subject, see our overview of dictation for work. For the physician side of the same question, read the guide to medical dictation and the guide to AI medical scribes.

Nurses spend a large share of every shift documenting, much of it in flowsheets

Nurses spend roughly a fifth to a third of their practice time on documentation, and the figure depends on how it was measured ¹. Time-motion studies cited by Collins and colleagues report that nurses spend between 19 and 35 percent of nursing practice time documenting care ¹. An EHR vendor's time logs at one academic health system put flowsheet work at 31.11 percent of a 12-hour shift ².

The volume of charting is large. On acute care units at one academic center, nurses made between 631 and 875 manual flowsheet entries per 12-hour shift, not counting device data ³. Each entry is a single value, such as a pulse, a pain score or a skin check, so the count is a sign of how many small decisions a nurse records, not how long each one takes.

Self-reports and observations also disagree. In a Western Australian hospital, observers found that nurses and midwives spent 28.1 percent of morning shifts on clinical documentation ⁴. In the same study, nurses and midwives self-reported 50.4 percent of morning shifts on documentation ⁵. That is a gap of 22.3 percentage points ⁶, so a time study and a survey can give very different answers for the same shift.

The useful reading is that documentation takes a large share of practice time (19 to 35 percent in the time-motion studies above ¹), and the self-reported figure runs well above what observers and logs record. Any tool that promises to cut documentation time should be measured against one of those baselines, not against how long the work feels.

Flowsheets resist dictation, while free-text notes and handoffs suit it

A flowsheet is a set of discrete fields. A nurse taps a value, picks an option from a list or selects a normal finding. Speaking a value such as "pain four out of ten" into a grid can be slower than tapping it, and recognition errors in numbers and units are costly in a clinical record. Our reading of the evidence is that flowsheet entries are the weakest case for dictation, and free text is the strongest.

Free-text work suits speech better: a shift summary, a handoff to the next nurse, a narrative about a conversation with a family, or a therapy session note written as connected prose. These are sentences, which is what speech recognition handles well. Where an EHR allows dictation straight into a progress note, the workflow is simpler. A 2020 tip from a Canadian health system says that Epic's Haiku and Canto apps support Nuance Dragon dictation directly into progress notes started in the mobile app ⁷.

Templates are the main risk in free text. Clinicians can dictate into pre-filled templates, and a 2024 review found that excessive use of DotPhrases can lead to an excessively long and cumbersome note ⁸. A dictation template should hold the fields you must record, and nothing more.

Speech recognition studies in nursing show small time savings so far

The nursing evidence is small. A 2024 literature review searched PubMed, CINAHL and Google Scholar, found 156 studies, and kept seven that met its criteria ⁹. The review concludes that speech recognition, especially when combined with artificial intelligence, can speed up documentation by up to 15 percent ¹⁰. That 15 percent is the top of the range the reviewed studies found, not a typical result.

In a 2022 survey of 73 ICU nurses in Iran, "paperwork reduction", "performance improvement" and "cost reduction" were the most common benefits they named ^{11 12}. The scores were close together on a five-point scale, so the ranking shows little ¹². The survey measured opinion, not time saved.

Accuracy takes time to learn. In a radiology study, new users of speech recognition corrected 11 characters per report on average ¹³. That study is about radiologists, not nurses, but it shows why early use of voice charting can feel slower than typing.

We did not find a randomized trial of speech recognition for nursing charting. Until one exists, the honest summary is that nurses have reason to try dictation for free text and no measured basis to expect large savings on flowsheets.

Therapy notes are narrative and suit dictation, but Medicare adds required fields

Therapy notes are narrative, so they suit speech. Medicare adds fields that a dictated note must still carry. The CMS fact sheet on outpatient therapy documentation says that progress reports justify medical necessity and must be written at least once every 10 treatment days, with the provider's signature, professional identification and date ¹⁴. It also asks therapists to document the total minutes of timed-code treatment for each treatment day, because those minutes support the number of units and codes billed ¹⁵.

For a physical or occupational therapist, that means a dictated note should end with the minutes, the interventions billed and the signature step, all spoken or typed as separate fields. The minutes are a number, so check them before you sign. For the billing rules that apply to your setting, read the Medicare manual and your contractor's local requirements, not a summary.

The evidence for therapists comes mostly from allied health. A 2025 mixed-methods study in Australian private practice surveyed 119 allied health clinicians before and after using an AI scribe, and it also surveyed patients ¹⁶. The authors report a significant reduction, compared with baseline, in self-reported time on notes, letters and out-of-hours notes ¹⁷. Clinicians also reported that their productivity increased by an average of 5.8 percent ¹⁸.

These numbers need care. The productivity figure is self-reported, the study was not randomized, and several authors work for the company that supplied the scribe. The paper does not report physiotherapists and occupational therapists separately, so it does not show which discipline gained most.

AI scribes in therapy raise consent questions stricter than in a medical visit

A medical scribe records a visit. A therapy session is more sensitive, and the law treats some therapy records more strictly. Under the HIPAA Privacy Rule, psychotherapy notes are notes a mental health professional makes about a counseling conversation, and they must be separated from the rest of the medical record ¹⁹. An AI scribe that records a whole session creates a transcript, and whether that transcript counts as psychotherapy notes under this definition is a question for your compliance team.

Consent is also more personal. In a 2026 NPR report, a client learned partway through a session that her therapist had been recording it with an AI note tool, and she later realized the conversation had been recorded without her consent ²⁰. The report also says therapists who use these tools tell clients and let them opt out ²⁰. That is the practice to copy: say what the tool records, where it is stored and how the client can refuse, before the first recording, and record that agreement.

The allied health study gives a more reassuring view. Patients in the Australian study were comfortable consenting to an AI scribe in their own appointments, but some said other patients might need more information about data storage and security before they could make an informed decision ²¹. The concern raised was about storage and security.

Accuracy matters as much as consent. In a UCLA randomized trial of 238 outpatient physicians ²², physicians said clinically significant inaccuracies in the AI scribe notes occurred occasionally ²³. The Nabla scribe cut time-in-note by 9.5 percent against the control group ²⁴. That trial was in medicine, so it shows the failure mode, not the size of the effect for therapy.

Before a therapist uses a cloud scribe, check three things. First, the vendor will sign a business associate agreement. Microsoft says that using its services does not on its own achieve HIPAA compliance ²⁵, and HHS says a cloud service that holds health information must sign one ²⁶. Second, you review every draft before you sign it. Third, you check your state licensing board and employer policy, since they can add rules beyond HIPAA.

SOAP, DAP and BIRP notes can be dictated with a spoken template

Most clinicians write in one of three layouts. The format is set by the clinic and the payer, and the dictation tool does not choose it.

To dictate a SOAP note, the steps are short:

1. Build a dictation template that holds only the headings you must record.
2. Say each heading aloud as a cue, so the words land in the right place when you review them.
3. Dictate measured values in full, such as the degrees of range of motion or the number of repetitions, and say the units.
4. Read the draft back and fix names, doses and numbers, which are the errors to check most closely.
5. Add the minutes and the signature fields if your payer requires them, then sign.

The template rule from the flowsheet section applies here too. Excessive use of pre-filled phrases can make a note excessively long and cumbersome ⁸. For a broader view of how to dictate clean text, read our guide to how to dictate, and see the question on how to dictate SOAP notes for the format-specific steps.

Home health and community clinicians often document away from a desk, where dictation fits

Home health and community clinicians often write notes after a visit, in a car or at a kitchen table, with no reliable connection. Dictation suits that setting, because the note can be spoken when the visit ends. Microsoft's Embedded Speech SDK is designed for on-device speech to text where cloud connectivity is intermittent or unavailable ²⁷. That is a description of how the feature is built, not a measurement of how clinicians benefit.

We found no study that measured documentation time for home health clinicians before and after dictation. The argument is plausible, but it rests on the workflow, not on data, and the same limits as the other settings apply.

Free text is where dictation is most likely to help, and the evidence for the rest is thin

Four points summarize the evidence.

• Nurses document a fifth to a third of their time ¹, and flowsheets take a large part of it, which dictation handles least well ².
• Nursing speech recognition has few studies, with modest reported savings and a 15 percent upper figure ¹⁰.
• Therapy notes suit dictation, because they are narrative, but the minutes and signature fields still need checking.
• AI scribes need consent and a signed agreement before they touch a therapy session, and drafts still need review.

The comparison with physicians is useful. In the Epic data, physicians who mainly used transcription and dictation spent 4.8 minutes per visit on notes, while physicians who mainly copied and pasted spent 13.1 minutes ^{28 29}. That is a gap of 8.3 minutes per visit ³⁰, and it is an observational comparison, not a trial. Nurses and therapists should expect the gain to depend on how much of their work is free text.

For the first steps, see the guide to how to dictate, and the question on whether nurses can use speech recognition for charting. For a question on time, read how much time nurses spend on documentation. For therapy notes and AI, see whether therapists can use AI scribes. If you write in other settings, the guides on medical transcription and how to dictate emails cover the adjacent work. Hand and wrist strain from typing is a separate issue, covered in our guide to dictation for RSI and carpal tunnel.

Where the evidence runs out

The gaps are specific, and they matter for purchasing decisions.

• No randomized trial of speech recognition for nursing charting turned up in our search.
• No measured study of documentation time for physical or occupational therapists turned up, and the figures we could find were from vendor pages, which we did not use.
• The allied health AI scribe study we found was not randomized and has company authors.
• No outcome study of AI scribes in mental health practice turned up.

Clinicians who want to test a tool should measure their own baseline first, for a week, with the minutes they already spend, and compare the same week after the change. Yap, which we make, is one of several dictation tools you could test in a text field, and the free dictation built into your computer is worth trying first, starting with voice to text on Mac or dictation on Windows.