It's 12:40 on a Thursday. Sarah's halfway down the corridor with a sandwich she hasn't touched when James, one of the psychiatrists, stops her. "This new dictation tool," he says. "Where exactly are my patients' words going?"
Sarah opens her mouth, and nothing crisp comes out. She knows the tool was approved. She knows it saves time. What she can't tell James, with the waiting room filling up behind him, is where the audio goes, what's kept, or who decides what lands in the record.
If this sounds familiar, you're in good company. AI charting confidentiality is now a governance question as much as a technical one, and the people asking it are your own clinicians. This post gives you a short list of questions to take into any vendor conversation, so the next time someone stops you in the corridor, you've got an answer ready.
Psychiatry, nutrition, aesthetics and private GP's work all involves patients telling you things they wouldn't tell most people. A psychiatrist who isn't confident about where those conversations go will quietly stop using the tool, or start leaving things out of the record that should have been included a coding to your SOP. Either way, the investment stalls and the clinic loses the time savings it bought the tool for.
Good clinical AI governance means you can explain, in plain language, what happens to a patient's information from the moment a clinician starts speaking. You don't need to be an engineer to do that. You just need to have asked the right questions before you signed anything.
Keep this in mind at the procurement phase, or keep it on your phone for the next demo. These work for any AI charting tool, not just ours.
Ask exactly what is captured (a full consultation recording, or the clinician's own dictated summary), where it's processed, and whether it leaves the platform you're directly using.
Ask a vendor where the audio goes and you'll hear the word "cloud" a lot. It's a lovely word. It suggests something fluffy, distant, and somebody else's problem entirely. It isn't an answer. Ask whose cloud, and which country.
Some tools keep recordings or transcripts to "improve the service". That's corporate speak for "we'd like a copy, please." It may be fine for a shopping app; it's a different conversation for a psychiatric history. Ask whether audio, transcripts or derived text are stored after the suggestions are produced, and get the answer in writing.
This is the big one. If the system can write a diagnosis, a prescription into the management plan or a history item into the chart without a person saying yes, you've handed part of your clinical record to software. Ask whether every item needs individual human approval, or whether clinicians can (or must) accept suggestions in bulk.
SNOMED CT coding software and ICD 10 suggestions save real time, but coding drives referrals, reporting and sometimes billing. Ask who signs off each code, and whether there's a clear trail showing that a named clinician accepted it.
Every extra tool is another login, another contract, another set of backups and another place sensitive information can sit. If the dictation tool, the transcription service and the EHR are three different products stitched together, count them all as places your patients' information lives. After a while, your clinic starts to feel less like a practice and more like a data warehouse that happens to do consultations on the side.
For UK clinics, data residency is a straightforward question with a straightforward answer, so expect one. If the vendor hedges, or talks about "primarily" UK hosting, ask what happens to the rest.
We built F365 Medical Assistant Dictation Mode with exactly these questions in mind, so here's how it holds up against the list.
The source is the clinician's own summary. Charting suggestions come from either the audio from the call itself, or from the practitioner's dictated summary of the consultation, converted in real time. In dictation mode the clinician chooses what to say, so they're in control of what the assistant hears in the first place.
Nothing goes in unless a human says so. Every suggestion has to be individually approved by the practitioner before it reaches the patient's record. F365's AI can't invent facts and add them on its own. If James dictates a summary and the assistant offers a medication history item he doesn't recognise, he simply doesn't approve it, and it goes nowhere.
Coding stays with the clinician. Suggestions cover SNOMED CT or ICD 10 diagnosis coding, medication history and personal medical history. The assistant does the look-up work; the clinician makes the call.
Data stays in the UK. Data residency is strictly UK only, which gives you a clean answer to question six.
Here's what that looks like in practice. Hannah, a dietitian, finishes a follow-up and dictates two minutes of summary while the details are fresh. By the time she's stopped talking, she's looking at a short list of suggested items: an updated diagnosis code, a new supplement in the medication history, a relevant family history note. She approves two, corrects nothing, rejects one that doesn't apply, and moves on to her her consultation summary document that builds on what she just confirmed, then the next patient. The chart and clinical noted reflect what she confirmed and nothing else.
Confidentiality isn't only about what a tool does with data. It's also about how many places that data ends up.
Plenty of clinic software options handle the record well, then asks you to bolt on a separate dictation app, which hands off to a separate transcription service, which pastes text back into your notes. Each handover is another system holding patient information, and another vendor on your data map. By the time you're updating your records of processing, you're listing four suppliers for one consultation.
F365 Medical Assistant Dictation Mode sits inside the Function 365 charting system built directly on AWS, so there's no stitching separate tools together. It's also designed for zero data retention by design, which means the assistant isn't building up a store of patient conversations in the background. Fewer systems and nothing retained add up to fewer places your patients' most sensitive details can sit.
It's worth being straight about this, because a cautious CMO will spot any overclaim from across the room.
Software design can support your confidentiality approach. It can't be your confidentiality approach. Whether your clinic meets UK GDPR or HIPAA obligations depends on how you run things: your policies, who has access to what, how you train staff, how you handle a subject access request, and what happens when someone leaves.
What a well-designed tool does is make the right behaviour the easy behaviour. Item-by-item approval, zero data retention by design, UK-only residency and one platform instead of four all reduce the number of things that can go wrong. Your governance still has to do the rest, and the question list above is a good starting point for your own internal policy too.
Confidentiality is the reason to ask the questions; time is often the reason the tool got shortlisted. Clinics tell us that F365 Medical Assistant Dictation Mode can cut charting time in half, which matters when clinicians are finishing notes at 8pm. Because approval happens in the moment, that time saving doesn't come at the cost of control, and practice managers get clinical hours back across the day without adding admin.
That's the trade-off you actually want: faster charting, with every fact still signed off by the person who was in the room.
If you're weighing up AI charting and want to see how a vendor handles audio, retention, approval and UK residency before your clinicians start asking, bring this list to a conversation with us. We'll answer each question on screen, using Dictation Mode itself, and you can decide whether it fits the way your clinic works.
Book a one-to-one demo and test our answers against your toughest corridor question.
It's 12:40 on a Thursday. Sarah's halfway down the corridor with a sandwich she hasn't touched when James, one of the psychiatrists, stops her. "This new dictation tool," he says. "Where exactly are my patients' words going?"
Sarah opens her mouth, and nothing crisp comes out. She knows the tool was approved. She knows it saves time. What she can't tell James, with the waiting room filling up behind him, is where the audio goes, what's kept, or who decides what lands in the record.
If this sounds familiar, you're in good company. AI charting confidentiality is now a governance question as much as a technical one, and the people asking it are your own clinicians. This post gives you a short list of questions to take into any vendor conversation, so the next time someone stops you in the corridor, you've got an answer ready.
Psychiatry, nutrition, aesthetics and private GP's work all involves patients telling you things they wouldn't tell most people. A psychiatrist who isn't confident about where those conversations go will quietly stop using the tool, or start leaving things out of the record that should have been included a coding to your SOP. Either way, the investment stalls and the clinic loses the time savings it bought the tool for.
Good clinical AI governance means you can explain, in plain language, what happens to a patient's information from the moment a clinician starts speaking. You don't need to be an engineer to do that. You just need to have asked the right questions before you signed anything.
Keep this in mind at the procurement phase, or keep it on your phone for the next demo. These work for any AI charting tool, not just ours.
Ask exactly what is captured (a full consultation recording, or the clinician's own dictated summary), where it's processed, and whether it leaves the platform you're directly using.
Ask a vendor where the audio goes and you'll hear the word "cloud" a lot. It's a lovely word. It suggests something fluffy, distant, and somebody else's problem entirely. It isn't an answer. Ask whose cloud, and which country.
Some tools keep recordings or transcripts to "improve the service". That's corporate speak for "we'd like a copy, please." It may be fine for a shopping app; it's a different conversation for a psychiatric history. Ask whether audio, transcripts or derived text are stored after the suggestions are produced, and get the answer in writing.
This is the big one. If the system can write a diagnosis, a prescription into the management plan or a history item into the chart without a person saying yes, you've handed part of your clinical record to software. Ask whether every item needs individual human approval, or whether clinicians can (or must) accept suggestions in bulk.
SNOMED CT coding software and ICD 10 suggestions save real time, but coding drives referrals, reporting and sometimes billing. Ask who signs off each code, and whether there's a clear trail showing that a named clinician accepted it.
Every extra tool is another login, another contract, another set of backups and another place sensitive information can sit. If the dictation tool, the transcription service and the EHR are three different products stitched together, count them all as places your patients' information lives. After a while, your clinic starts to feel less like a practice and more like a data warehouse that happens to do consultations on the side.
For UK clinics, data residency is a straightforward question with a straightforward answer, so expect one. If the vendor hedges, or talks about "primarily" UK hosting, ask what happens to the rest.
We built F365 Medical Assistant Dictation Mode with exactly these questions in mind, so here's how it holds up against the list.
The source is the clinician's own summary. Charting suggestions come from either the audio from the call itself, or from the practitioner's dictated summary of the consultation, converted in real time. In dictation mode the clinician chooses what to say, so they're in control of what the assistant hears in the first place.
Nothing goes in unless a human says so. Every suggestion has to be individually approved by the practitioner before it reaches the patient's record. F365's AI can't invent facts and add them on its own. If James dictates a summary and the assistant offers a medication history item he doesn't recognise, he simply doesn't approve it, and it goes nowhere.
Coding stays with the clinician. Suggestions cover SNOMED CT or ICD 10 diagnosis coding, medication history and personal medical history. The assistant does the look-up work; the clinician makes the call.
Data stays in the UK. Data residency is strictly UK only, which gives you a clean answer to question six.
Here's what that looks like in practice. Hannah, a dietitian, finishes a follow-up and dictates two minutes of summary while the details are fresh. By the time she's stopped talking, she's looking at a short list of suggested items: an updated diagnosis code, a new supplement in the medication history, a relevant family history note. She approves two, corrects nothing, rejects one that doesn't apply, and moves on to her her consultation summary document that builds on what she just confirmed, then the next patient. The chart and clinical noted reflect what she confirmed and nothing else.
Confidentiality isn't only about what a tool does with data. It's also about how many places that data ends up.
Plenty of clinic software options handle the record well, then asks you to bolt on a separate dictation app, which hands off to a separate transcription service, which pastes text back into your notes. Each handover is another system holding patient information, and another vendor on your data map. By the time you're updating your records of processing, you're listing four suppliers for one consultation.
F365 Medical Assistant Dictation Mode sits inside the Function 365 charting system built directly on AWS, so there's no stitching separate tools together. It's also designed for zero data retention by design, which means the assistant isn't building up a store of patient conversations in the background. Fewer systems and nothing retained add up to fewer places your patients' most sensitive details can sit.
It's worth being straight about this, because a cautious CMO will spot any overclaim from across the room.
Software design can support your confidentiality approach. It can't be your confidentiality approach. Whether your clinic meets UK GDPR or HIPAA obligations depends on how you run things: your policies, who has access to what, how you train staff, how you handle a subject access request, and what happens when someone leaves.
What a well-designed tool does is make the right behaviour the easy behaviour. Item-by-item approval, zero data retention by design, UK-only residency and one platform instead of four all reduce the number of things that can go wrong. Your governance still has to do the rest, and the question list above is a good starting point for your own internal policy too.
Confidentiality is the reason to ask the questions; time is often the reason the tool got shortlisted. Clinics tell us that F365 Medical Assistant Dictation Mode can cut charting time in half, which matters when clinicians are finishing notes at 8pm. Because approval happens in the moment, that time saving doesn't come at the cost of control, and practice managers get clinical hours back across the day without adding admin.
That's the trade-off you actually want: faster charting, with every fact still signed off by the person who was in the room.
If you're weighing up AI charting and want to see how a vendor handles audio, retention, approval and UK residency before your clinicians start asking, bring this list to a conversation with us. We'll answer each question on screen, using Dictation Mode itself, and you can decide whether it fits the way your clinic works.
Book a one-to-one demo and test our answers against your toughest corridor question.
