UK aesthetic clinics handle sensitive clinical data under UK GDPR. A security audit published in early 2026 found a widely used AI scribe tool was sending patient data to Google Analytics via URL parameters. Not a theoretical risk. A real finding, documented and reproducible. That is what happens when a product is built for growth first and compliance second.
Four questions worth asking before you connect an AI tool to your patient records.
Does it record the consultation, or only process the notes you write?
Most AI note tools use audio transcription. A microphone captures the consultation, the tool transcribes it, and AI structures the output. The notes are good. The trade-off is that audio of your patient's consultation travels to a third-party server.
That creates a consent obligation. Your patients need to know their consultation is being recorded and processed externally. Your privacy notice needs updating. Your data processing agreement with the provider needs to be in place and UK GDPR-compliant before you switch it on.
Some tools work differently. They take the shorthand a clinician writes afterwards and reformat it into structured text. No audio. No transcript. No recording of the patient or the conversation.
If you are not certain which category your tool is in, check the data flow documentation before enabling it.
Does it log what the notes say?
Processing a note and storing it are different things.
Processing means the text travels to a server, gets reformatted, returns. Storing means a copy of what the note said stays in a log file on the provider's infrastructure. Those logs may be retained for debugging, model training, or quality assurance.
Ask your provider: do you retain copies of note content after processing? What is the retention period? Are note contents used to train models? If the documentation is vague or the provider cannot answer quickly, that tells you something.
Does a clinician approve every summary before it saves?
The clinical record is a legal document. In an audit or complaint, it is the evidence. A note generated by AI and saved without a clinician reading it first is a governance problem regardless of accuracy.
The correct flow: AI produces a draft, clinician reads it, clinician approves or edits, it saves. The clinician is responsible for what ends up in the record. The AI assists the drafting. It does not replace the review.
If a tool saves automatically, or makes the review step optional, treat that as a governance risk before a convenience feature.
Is your patient data isolated from other clinics at the database level?
Multi-tenant platforms serve many clinics from shared infrastructure. Proper isolation means each clinic's records are walled off at the database level. Not by an access policy. By the schema itself, so that even a misconfiguration cannot expose one clinic's patients to another.
Ask your provider how isolation is enforced. The answer should reference database-level controls. Application-level permissions can be bypassed by bugs. Database constraints enforced at the schema level are significantly harder to breach accidentally.
How Calyx Refine is built
Calyx Refine takes the shorthand notes a clinician writes and suggests a structured version. No microphone. No consultation recording. No audio transcript sent anywhere.
The system never logs what a note said. Each clinic's records are walled off from every other clinic at the database level. Every output is a draft, a clinician reads it and approves it before anything saves.
The four questions above are the ones we asked ourselves when building it.
Refine is on the Calyx Professional plan. To see it running in a full clinic workflow, book a demo or read how it works on the Refine page.
