A plain-language guide for healthcare practices using the Nimbiq website assistant — what happens to a conversation from the moment a visitor types, to what your staff actually see.
Our assistant answers visitor questions and captures contact information for follow-up. Staff are notified by email and text with a summary, not the full conversation. This is a deliberate privacy choice, not a limitation we overlooked. While we support full conversation logs and dashboard access for clients in other industries, we disable those paths for healthcare-flagged accounts to prevent any exposure of sensitive patient detail. That deliberate choice is the reason we're comfortable operating on a pediatric, medical, dental, or any other healthcare practice's website in the first place.
What your front desk works from is the email and text notification— deliberately narrow. We do not offer a practice workflow that opens the full word-for-word chat as the primary way to follow up. Notifications include:
Email and text messages are convenient, but they are not a controlled environment. A single notification can be forwarded, auto-synced to a personal phone, backed up to a shared inbox, or read over someone's shoulder — none of that requires anyone to do anything wrong, it's just how email and SMS work in a busy front office.
So we made a deliberate choice: if a visitor mentions symptoms, medication, dental history, or other medical detail while chatting, that information stays inside our secure system and is never placed into an email or text — regardless of how sensitive or routine the message actually was.
We considered a version where our software would try to detect whether a message contains health information, and only redact it when detected. We rejected that approach:
Automatic detection is not reliable enough for health information. People describe symptoms and concerns in everyday language — "she hasn't been herself since Tuesday," "same issue as last time," "the tooth still hurts after the filling"— that software can miss entirely. A detector that misses even occasionally means a patient's health information ends up in an email by mistake. We don't consider that an acceptable trade-off for a pediatric, medical, dental, or any other healthcare practice, so we treat every healthcare conversation the same way rather than gambling on detection accuracy.
Your front desk gets enough from the notification to call the visitor back and pick up the conversation — name, number, and a clear sense of why they reached out. That phone call is the intended path, the same way a voicemail or a front-desk note works today.
Follow up using the contact details and non-clinical intent summary provided in the email or text notification.
Ever, for any healthcare-flagged conversation — not just ones that seem sensitive.
We never send links that open a full conversation transcript in a browser. A forwarded or bookmarked link would be a copy of visitor information outside your control — we don't create that risk.
This is not a substitute for your patient portal or EHR. Visitors are told this in the chat itself.
The assistant directs anyone describing a medical emergency to call 911 or your office directly, not to continue chatting.
Raw chat transcripts for healthcare-flagged accounts are automatically deleted after 30 days. Short-lived storage supports system operation and abuse review; it is not offered as a staff tool for reading full patient conversations day to day. We do not keep raw transcripts indefinitely, because the ongoing risk of storing health-adjacent conversations outweighs any convenience.
The lead record itself — contact information and a non-clinical summary of why the visitor reached out — is kept as your ongoing record, so you retain a usable history of inquiries without retaining raw health information indefinitely.
Chat messages are processed by third-party AI models to generate responses and match content. We evaluate many different models for this use case — not to chase a leaderboard, but to pick what works for a busy practice website. Selection weighs factors such as how fast the model responds, how verbose or concise it tends to be, whether answers stay on-task for patient and visitor questions, and whether the path can meet our data-handling rules. Naming a provider is not a permanent vendor lock or a claim that one model is always "the best" in the abstract.
Models such as Google Gemini have been among the tools we use for chat responses, alongside other providers like OpenAI for related AI tasks. We re-evaluate frequently as new options appear; when something fits this workload better, we can switch. What does not change is how we route healthcare traffic through whichever model is selected.
We do not enable features like live web search grounding that would create additional third-party storage of visitor messages.
More context feels more useful — we understand the instinct. But the fastest way to lose a patient's trust, and create real liability for your practice, is a health-related message ending up somewhere it shouldn't, even by accident. We designed follow-up around a non-clinical notification plus a phone call — not around staff reading a full transcript in a dashboard. If the summary is not enough in practice, we improve the summary together; we do not put the transcript back into email, text, or a day-to-day staff workflow.
No. The phone call, using the contact details in the notification, is the intended path — exactly like a voicemail or a written note asking someone to call back. The assistant is built to make that first call easy, not to replace it with a transcript review.
Yes. We don't try to judge in the moment which messages are sensitive and which aren't, because that judgment call is exactly where mistakes happen. Every healthcare conversation is treated the same way, consistently — which is also what makes this policy simple for us to guarantee to you.
Yes, but only deliberately and in writing. If, after real use, this genuinely isn't working for your front desk, we'll revisit it together — the fix would be improving what's in the summary itself, not adding the transcript back into email or text.
Reach out directly to your Nimbiq contact. We're glad to walk through this with your office manager or compliance advisor before go-live.