Voice typing for doctors: faster notes, patient data protected

A doctor's day does not end when the last patient leaves. It ends when the notes are done. A well-known 2017 study in the Annals of Family Medicine used electronic health record logs from primary care physicians and found nearly two hours of EHR work for every hour of direct patient care, with close to six hours a day spent in the record. A 2024 follow-up by the same group found that EHR time kept growing. Most of that work is typing.

Voice typing for doctors is the obvious answer, and also a risky one, because the thing you dictate is patient information. This guide covers where dictation helps, how to do it well, and what to settle before you trust any tool with clinical text.

Where dictation saves a doctor time

• Notes right after the appointment. Two minutes of speaking while the details are fresh beats twenty minutes of reconstruction at 9 pm.
• Referral letters and reports. These are structured, repetitive and mostly prose. Dictate the substance, then check dosages and dates by eye.
• Replies to patients and colleagues. Messages in the portal or by email are first drafts you can say out loud.
• Plans and handover notes. Speaking in order (history, findings, assessment, plan) gives you structure for free.
• Teaching and research writing. Case write-ups, audit notes and lecture drafts are long text that suits talking.

Dictation is weaker for numbers, drug names, structured order entry and anything that must be picked from a dropdown. Say the narrative, then handle the exact items by hand.

How to dictate clinical text well

1. Speak in full thoughts. Finish the sentence in your head, then say it. Good software removes the «um» and false starts.
2. Use a fixed order. If you always say history, examination, assessment, plan, your notes become consistent and your brain stops hunting for the next sentence.
3. Spell out the risky bits. Drug names that sound alike, left and right, units and doses are where an error does harm. Read them back before you sign.
4. Build a personal dictionary. Names of drugs, procedures, local clinics and colleagues are where dictation stumbles. A dictionary that remembers corrections pays off within a week.
5. Always review. Unedited speech recognition makes mistakes. Treat dictated text as a draft you are responsible for, exactly like a note typed by a trainee.

If you work in more than one language, say in Russian, Polish or Ukrainian with patients and English in your notes, choose a tool that handles mixed speech without a settings switch. The overview of voice typing in 2026 explains how that works now.

The patient data question

This part decides which tool you can use at all, and the answer depends on your country, your employer and your role.

In the United States, HIPAA treats a cloud provider that handles protected health information for you as a business associate. According to law-firm summaries of the HHS cloud computing guidance (also covered by Mintz), that means a signed business associate agreement is required, even when the provider stores only encrypted data. We could not open hhs.gov from the session that prepared this article, so read the HHS page itself. In the EU, GDPR treats health data as a special category with stricter rules, and other countries have their own. Nothing here is legal advice, and your hospital's or clinic's policy may be stricter than the law.

Before you dictate anything identifiable, check five things:

1. Where is the audio processed? On your device, in a vendor's cloud, or both?
2. Is there an agreement? If your rules require a business associate agreement or a data processing agreement, does the vendor offer one for the plan you use?
3. Is anything kept? Retention of audio and transcripts, and whether you can delete them.
4. Is your data used to train models? The terms should say so plainly.
5. What does your employer allow? Many hospitals have an approved-tools list, and a personal app may not be on it.

A simple safe habit while you decide: dictate without patient names, dates of birth or record numbers, and add identifiers by hand in the record. It costs a few seconds and removes most of the risk.

How Tiho fits

Tiho is a dictation app for Mac and iPhone. You hold a key, speak, and finished text appears in the app you are using: your notes tool, email, a browser form. For clinical work, the points that matter:

• On-device mode. An offline model keeps audio on your machine and works with no internet at all.
• Cloud only with your consent. Transcription and formatting use the cloud only after you allow it in the app.
• It listens only when you tell it to, and it shows when it is listening.
• Delete anytime. Recordings can be removed and the account deleted.
• Dictionary and snippets for terms and the paragraphs you repeat.

What Tiho is not: a medical record system, a certified clinical documentation product, or a tool that ships with a business associate agreement we can promise here. If your rules require one, ask before you use it with identifiable patient data, and use the on-device mode or de-identified text until you have an answer. The on-device model is more limited than the cloud one, and there is no Windows or Android version, which matters in many hospitals. See how data is handled on the Tiho privacy section. The free plan gives 2,000 words a week, enough to test it on letters and non-identifiable drafts.

A simple way to start

1. Pick one task: referral letters, or replies to colleagues.
2. Use on-device processing and no identifiers.
3. Dictate for a week and add every misheard term to the dictionary.
4. Check your employer's policy and the vendor's terms before you widen use.

Dictation will not write your assessment for you. It removes the slowest part, getting words out of your head, and leaves you the part that is medicine.

FAQ

Can I dictate patient notes with a normal dictation app?

Only if your rules allow it. Identifiable patient data usually needs a vendor agreement, a lawful basis and your employer's approval. Until that is settled, dictate without identifiers or use fully on-device processing.

Is voice typing accurate enough for medical terms?

It is good with common terms and weaker with rare drug names and similar-sounding words. A personal dictionary helps a lot, but you must still read every note before signing it.

Does dictation work without internet in a clinic?

An on-device model does. It is more limited than cloud transcription, but it works in a basement office, on a ward with poor signal or on a plane to a conference.