Patient Recording Standard
What people deserve when a conversation that matters is recorded, transcribed or summarised. Open for anyone to use, challenge and improve.
Where this is
This is a working draft. Nothing here is final, and every principle can be changed, merged or dropped by the people who shape it.
We invite clinicians, defence organisations, regulators, patient groups and anyone building recording tools to comment. A principle someone else improved is worth more than one written alone.
The Standard is published under an open licence (proposed: CC BY 4.0), so anyone can use, adapt or build against it, whether or not they ever use MyScribe.
Why it exists
People already record their appointments, often covertly. Around 7 in 10 patients say they would like to record, and roughly half of them would do it without asking. The question is no longer whether recording happens, but whether it happens openly, safely and fairly.
This Standard sets out what safe patient recording looks like, so both people in the room can be at ease with it.
In scope
Tools a person uses to record, transcribe or summarise their own conversations, such as an appointment with a clinician, a care review or a benefits meeting. That includes apps, websites and features built into devices.
Out of scope
- Clinician-side ambient scribes that write into a clinical record. They serve the organisation's record and are covered by other guidance.
- Recording by organisations for their own purposes, such as call recording.
- Recording without the other person's knowledge. The Standard offers an open alternative to covert recording rather than trying to make it safer.
Words we use
- Person
- The individual who starts the recording and owns the record.
- Other participant
- Anyone else in the conversation, such as a clinician, carer or advisor.
- Recording tool
- Any app, website or device feature in scope.
- Verbatim record
- The transcript of what was said, word for word, with speakers separated.
- AI-written content
- Anything the tool composes rather than transcribes, such as a summary.
- Moment
- The point in the verbatim record where something was said.
The nine principles
A conforming tool meets all nine. Each one comes with the safety reason behind it, so it can be judged on its merits.
-
Ask in the room
Requirement: the tool's first step is asking the other participant whether they agree to the recording. Recording cannot start without that ask being shown, and the tool offers wording the person can read out.
Why: open consent is the basis of medical defence organisation guidance on patients recording consultations. Asking turns a covert act into a shared one.
-
Audio is not kept
Requirement: original audio is deleted once the verbatim record is made. It is never stored, shared or played back.
Why: audio can be clipped, shared out of context or used against a participant. Clinicians object to audio far more than to a transcript, and a voice is identifying data in its own right.
-
Answers come only from what was said
Requirement: when the person asks about the conversation, the tool answers only by pointing to the words spoken, at their moment. It never writes its own answer.
Why: an AI paraphrase of clinical advice is new advice that nobody gave. Pointing to the real words keeps the other participant's meaning intact.
-
No medicines or doses in AI-written content
Requirement: medicine names and doses appear only in the verbatim record, never in summaries or other AI-written content.
Why: a wrong medicine or dose is among the most harmful errors a summary can make. Sending the person to the exact words removes that risk.
-
Uncertainty is shown, never smoothed
Requirement: unclear audio and low-confidence words are visibly flagged in the verbatim record. The tool does not silently guess or tidy them.
Why: a confident wrong word is more dangerous than a visible gap. Flagging tells the person when to check with the other participant.
-
Every summary sentence traces to its moment
Requirement: each sentence of AI-written content links to the moment in the verbatim record it came from.
Why: the person can check any line, and the other participant can see exactly what was said behind it.
-
The record belongs to the person
Requirement: the record is never sold, never used to train AI models and never reviewed or monitored by the tool's provider. The person can export and delete it at any time.
Why: people will only record openly if they trust where the record goes. Ownership is also what separates a personal memory aid from surveillance.
-
Nothing lands on the other participant
Requirement: the tool makes no clinical claims, never writes into a clinical or organisational record, and does not present itself as endorsed by the other participant.
Why: clinicians should not carry liability for a tool they did not choose. This keeps the record a personal aid, not a second clinical record.
-
A route to say something is not right
Requirement: every summary offers a simple way to report an error, and the provider acts on reports and publishes how.
Why: every safety system needs a reporting channel. Errors that are reported can be fixed. Errors that are hidden repeat.
How a tool shows it conforms
A tool conforms when it meets all nine principles in its released version, not in its plans.
- The provider publishes a self-assessment against each principle, saying how it is met and linking to evidence.
- The self-assessment sits where people choosing the tool can find it, and is updated with every release that affects a principle.
- Anyone can challenge a self-assessment. Challenges and responses are published.
- Independent assessment and a conformance mark may follow in a later version, once the principles have settled.
MyScribe is being built to this Standard and will publish its own self-assessment at launch.
Who shapes it
The Standard is versioned (0.1, 0.2 and so on) with a public change log. Version 1.0 will only be published once at least three organisations from outside its original authors have contributed.
We are seeking early contributors from a medical defence organisation, a patient voice organisation, a health innovation network, clinician representatives and at least one other recording tool provider. No single provider may hold a majority of contributors, MyScribe included.
Open questions
These are the questions we most want help with.
- Should principle 1 require the other participant's answer to be recorded in the tool, or only the ask to be shown?
- Is a full ban on medicines in AI-written content right, or should a verified, word-for-word quote be allowed?
- Should the Standard say anything about where data is processed and stored?
- How should it treat recording for people who lack capacity, or recording by carers on their behalf?
- Should there be a minimum age for people using conforming tools?
- Who should hold the Standard long term: a coalition, an existing body, or a new one?
Evidence
- Elwyn G, Barr PJ, Grande SW (2015). Patients recording clinical encounters: a path to empowerment? BMJ Open 5(8): e008566. doi:10.1136/bmjopen-2015-008566. Around 69% of patients would like to record encounters, roughly half openly and half covertly.
- O'Hara et al. (2017). Setting the record straight. Clinical Medicine 17(2): 188. doi:10.7861/clinmedicine.17-2-188. Open and covert recording rates, and medico-legal survey findings.
- Dommershuijsen, Dedding, Van Bruchem-Visser (2021). Consultation recordings in patients aged 50 and over: a systematic review. Health Communication 36(2): 168-178. doi:10.1080/10410236.2019.1669270. Recordings improve recall, retention and satisfaction, and are often shared with family.
- Kessels RP (2003). Patients' memory for medical information. Journal of the Royal Society of Medicine 96(5): 219-222. Between 40% and 80% of medical information is forgotten immediately.
Help shape it
Comments, challenges and better wording are all welcome. Tell us which principle you would change, and why.
Send your thoughts