Key takeaways
- There are five main routes from spoken words to a finished letter; most clinics end up with a mix.
- Medical secretaries typically earn £23,000 to £30,000 a year[1], so compare options on total time and turnaround, not licence cost alone.
- Tools intended solely for transcription or drafting letters are not regulated as medical devices, but the clinician must still check every output[2].
- Audio and letters contain health data[3]; outsourced and cloud services are processors and need a contract, a DPIA and clear retention rules.
The letter bottleneck
In many private clinics the consultation is not finished when the patient leaves. A letter to the GP, a report for an insurer or a summary for the patient still has to be written, checked and sent. When letters take days, referrers chase, insurers delay payment and patients wonder whether anything is happening. When clinicians type their own letters in the evening, they pay for it in time and fatigue.
The good news is that there are now more ways to turn speech into a finished letter than ever before. The choice depends on your volume, the complexity of your letters, how quickly they must go out and how much checking each clinician is willing to do.
Five routes compared
| Route | How it works | Turnaround | Watch out for |
|---|---|---|---|
| In-house typing from dictation | Clinician records audio; a secretary types and formats the letter | Hours to days, depending on workload | Backlogs during leave; secretary time is the main cost |
| Outsourced transcription | Audio is sent securely to a typing service, which returns a draft | Often same or next day, by agreement | Data protection terms, where typists work, quality checks |
| Front-end speech recognition | Clinician dictates and corrects the text on screen themselves | Immediate | Clinician time spent correcting; accuracy with accents and terms |
| Back-end speech recognition | Software drafts the letter from audio; a secretary corrects it | Faster than typing | Corrections still need someone who knows the language |
| Ambient AI scribe | Software listens to the consultation and drafts notes and letters | Immediate draft | Consent, governance, checking every output, regulatory status of extra features |
What each route really costs
Licence prices are only part of the picture. The bigger cost is people's time: clinicians dictating and checking, secretaries typing and correcting, and managers chasing backlogs. A useful comparison counts all of it.
The National Careers Service puts typical medical secretary pay at £23,000 to £30,000 a year for 37 to 39 hours a week[1]. Employers also pay secondary Class 1 National Insurance at 15% on earnings above £5,000 a year[4], plus pension contributions. Clinician time is far more expensive: typical GP pay is £79,000 to £119,000 a year[5], and private consultants often earn more.
Source: National Careers Service (GOV.UK), 2026[1]
Show the data
| Item | Value |
|---|---|
| Starter | £23,000 |
| Experienced | £30,000 |
Example: a fictional clinic produces 60 letters a week. A secretary takes about ten minutes to type and format each from dictation, so typing alone uses around ten hours a week. Back-end speech recognition might cut that to a few minutes of correction per letter. Front-end recognition removes secretary time but adds several minutes of clinician correction to each letter. Work out the hours for each route with your own volumes before comparing quotes.
Accuracy and clinical safety
Every route makes mistakes. Typists mishear; speech recognition confuses similar drug names, numbers, left and right, and negatives such as "no evidence of". AI scribes can also summarise wrongly or include things that were not said. MHRA guidance published in July 2026 says that products intended solely for transcription, summarising clinical conversations, drafting letters or suggesting clinical codes for a clinician to review are not regulated as medical devices, while products that support diagnosis or treatment, or take automated action without clinician review, are[2]. It also says clinicians remain responsible for reviewing and verifying AI-generated transcripts, summaries and other outputs[2].
- Make clinician sign-off a mandatory step before any letter is sent.
- Use templates with fixed sections so missing information is obvious.
- Train dictation habits: spell unusual names, state laterality clearly, say "full stop" and "new paragraph" if the system needs it.
- Audit a sample of letters each month for errors, especially medicines, doses and sides.
- Record errors that reach the GP or patient as incidents and learn from them.
57%
of GPs using AI use it for clinical documentation and note-taking[6]
Data protection for audio and letters
Dictations and letters contain health information, which is special category data under UK GDPR[3]. Any service that stores or processes them on your behalf, whether a typing bureau, a cloud dictation platform or an AI scribe, is a processor. You need a contract that meets UK GDPR, clarity about where data is processed and by whom, and rules on how long audio is kept. For AI scribes and other new technologies processing health data at scale, a data protection impact assessment is expected before you start[7].
- Ask where audio and text are stored and processed, including any staff or subcontractors outside the UK.
- Check whether the supplier uses your data to train its software, and whether you can opt out.
- Set retention: delete audio once the letter is signed off, unless you have a reason to keep it.
- Make sure devices used for dictation are encrypted and locked, and apps log out automatically.
- Tell patients in your privacy notice how consultations and dictations are processed, and for ambient tools, tell them at the start of the consultation.
Templates that save more time than software
Whatever route you choose, good templates do much of the work. A consistent structure means the clinician dictates less, the typist or software has less to guess, and the reader finds the important part quickly. A simple structure for most clinic letters:
- Patient identifiers and date of consultation.
- Summary in two or three lines: diagnosis or working impression, and the plan.
- Actions for the GP, if any, listed separately and clearly worded.
- History, examination and investigations, kept brief.
- Medicines started, stopped or changed, with doses.
- Follow-up arrangements and what the patient has been told.
Agree separate templates for insurer reports and patient letters. Patient letters should be in plain English and avoid abbreviations.
Getting letters out of the door
A letter is not finished until it reaches the reader. Decide how each type is sent: secure email for GP practices, the insurer's portal or secure email for reports, and email or post for patients according to their preference. Keep a sent log that shows the date each letter went and to whom. Many delays that look like typing problems are really sending problems: signed letters sitting in a folder, or GP letters sent to an old address. Check the sent log weekly and chase anything older than your target turnaround.
Ambient scribes in a private clinic
Ambient AI scribes are the newest option. The government's 10 Year Health Plan commits to rolling out ambient voice technology to cut clinical admin across the NHS[8], and many private clinicians are already trialling it. For a private clinic the questions are the same as for the NHS: is the product clearly limited to transcription and drafting, is patient consent handled properly, who checks the output, and how are recordings stored? Treat a trial as a governed project with a named lead, not as a clinician quietly installing an app. Agree in advance which consultations are excluded, for example safeguarding discussions or patients who decline, and check whether new features added by the supplier, such as suggested diagnoses or automatic orders, would change the product's regulatory status.
Choosing in five steps
Count your letters
Volume per week, by clinician and type, and current turnaround time from consultation to sending.
Decide who checks
Agree whether clinicians will correct drafts themselves or rely on a secretary, because this drives the choice.
Trial two options
Run a four-week trial with willing clinicians, measuring turnaround, correction time and errors.
Do the paperwork
DPIA, processor contract, retention rules and privacy notice updates before any real patient data is used.
Roll out with training
Short training on dictation habits and the sign-off step, with a review after three months.
Example: a mixed model
Example: Riverside Physio and Skin (fictional) had three consultants with very different habits. One preferred to type and moved to front-end speech recognition; another kept dictating, with back-end recognition drafts corrected by the secretary; the third trialled an ambient scribe for routine follow-ups. Turnaround fell from several days to the same or next day for most letters, and the secretary's time shifted from typing to checking and sending.
Common mistakes
- Buying on licence price without counting clinician correction time.
- Skipping the sign-off step because the drafts "look right".
- Leaving audio files on phones or shared drives indefinitely.
- Forgetting insurer report formats and GP letter requirements when building templates.
- Assuming one route suits every clinician. A mix is often the most efficient answer.
Frequently asked questions
Is speech recognition accurate enough for clinical letters?
It is often good, but not perfect. Drug names, numbers, laterality and negatives cause errors, so a clinician must check every letter before it is sent.
Are AI scribes medical devices?
MHRA guidance says products intended solely for transcription, summarising or drafting letters are not, but products that support diagnosis or treatment, or take automated action, are[2].
Can we send dictations to an overseas typing service?
Only with a UK GDPR compliant contract and appropriate safeguards for international transfers. Ask exactly where typists work and how audio is protected[3].
Should we keep the audio after the letter is sent?
Usually not. Set a short retention period and delete audio once the letter is signed off, unless you have a specific reason to keep it.
References
National Careers Service (GOV.UK) (2026). Medical secretary job profile (opens in a new tab) ↑
GOV.UK (Medicines and Healthcare products Regulatory Agency) (2026). MHRA clarifies regulatory status of ambient voice technologies used in the NHS (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2026). UK GDPR Article 9: Processing of special categories of personal data (opens in a new tab) ↑
GOV.UK (HMRC) (2026). Rates and thresholds for employers 2026 to 2027 (opens in a new tab) ↑
National Careers Service (GOV.UK) (2026). GP job profile (opens in a new tab) ↑
Nuffield Trust (2025). How are GPs using AI? Insights from the front line (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2026). UK GDPR Article 35: Data protection impact assessment (opens in a new tab) ↑
GOV.UK (Department of Health and Social Care) (2025). Fit for the future: 10 Year Health Plan for England (opens in a new tab) ↑