Classification in under a second
The Safety Agent classifies every review across 11 dimensions, including the presence of a treatment, refund request, or legal risk. Signals are flagged by construction.
Feature · Safety Agent
Alerts and escalations for sensitive situations
The Safety Agent blocks medically sensitive situations (treatment mention, refund request, identifications, legal threat) and prevents any automatic publishing.
For every sensitive review, the practitioner receives an alert by email and in the dashboard, with the measured reply draft, timestamped evidence, and the Google reporting procedure ready to launch.
What the Safety Agent does
No sensitive review published without you
Any review that contains a sensitive signal is locked. No automatic reply is sent, and the Google listing is never updated without human validation.
Measured draft, ready to decide
For every alert, a measured draft is prepared, compliant with the code of ethics, that neither admits nor denies any care.
Google reporting procedure
When a review crosses the line (misconduct, defamation, identifications), the Google reporting procedure is ready, with timestamped evidence and a pre-filled form.
How an alert is triggered
When the Safety Agent detects a sensitive signal on a review, it locks publishing, prepares a draft, and notifies the practitioner by email and in the dashboard.
- 1
Detect the sensitive signal
The Safety Agent scans the review in under a second across 11 dimensions (treatment, refund, identifications, legal, etc.).
- 2
Lock the publishing
If one of the 11 signals is present, the review is locked: no reply is published, no automatic draft is handed to the Patient Agent.
- 3
Prepare the measured draft
A measured draft is written, compliant with the code of ethics, that neither admits nor denies any care.
- 4
Notify the practitioner
The practitioner receives an alert by email and in the dashboard, with the review, the draft, and the Google reporting procedure ready to launch.
Why now
Framework current as of September 2026.
Three bodies of law govern how you reply to an online review.
Revydo does not reinvent the framework. It absorbs it into every agent, so compliance does not depend on a forgetful assistant.
Legal framework
Consumer right to publish an online review
An online review is protected speech, not a marketing artefact. Hosting platforms and the reviewed professional both have obligations to the publisher, including the right of reply and the limits on removal.
EU Directive 2000/31/EC and national transpositions
Personal data
GDPR in force since May 2018
Every review contains personal data. The act of drafting a reply, storing it, or displaying it alongside the review is processing under GDPR. A careless reply is a separate data event, not a footnote.
Regulation (EU) 2016/679
Professional ethics
Codes of professional conduct
Where the practice is regulated (medicine, law, dentistry, physiotherapy, etc.), the professional code of ethics stays enforceable inside the reply. A draft that confirms a client relationship or reveals a covered fact is a disciplinary risk, not a marketing choice.
National codes of professional conduct
Frequently asked questions
What are the 11 signals scanned by the Safety Agent?
Treatment mention, refund request, identifications (patient or third party), clinical complaint, comparison with a colleague, legal threat, accusation of misconduct, life-threatening urgency mentioned, identified minor, conflict of interest, and procedural irregularity. The list is versioned and public.
Can I configure the signals that alert me?
Yes. The Safety Agent always classifies across the 11 dimensions, but you can choose the alert channel (email, SMS, phone) and the threshold (every alert / critical-only) by category.
What happens if I do not respond to an alert?
No reply is published without you. The review stays locked, the alert stays open in the dashboard, and a reminder is sent every 24 hours until you have decided.
Does the alert expose patient data to the AI?
No. The Safety Agent classifies the review locally on the basis of deterministic rules and classification models trained on public data. No identifiable patient data is sent to an AI provider.
Go further
Back to the Voice Agent
The Safety Agent is the lock before publishing. The Voice Agent never drafts a reply while a sensitive signal is still active.