AI-Drafted Replies

DPC Pro’s AI assistant can draft replies to patient messages, giving clinicians a starting point they can review, edit, and send.

When a patient sends a message, the AI assistant analyzes the message content, the patient’s clinical history, and your practice’s communication style to draft a suggested reply. The draft appears in the message thread for the clinician to review. No AI-drafted message is sent without clinician approval.

This page covers how AI-drafted replies appear in your inbox, how to review and edit them, and how to adjust the AI’s behavior over time. The goal is to reduce the time clinicians spend composing routine responses while maintaining the personal touch patients expect from their DPC provider.

AI-drafted replies work best for common clinical questions, appointment follow-ups, and administrative responses. Complex clinical discussions should always be written by the clinician directly.

AI Transparency

Before using AI-drafted replies, your team should understand what the AI does, what it does not do, and where you remain in control.

What It Does

The AI assistant reads the patient’s incoming message and relevant context from their patient record, then generates a suggested reply. Specifically:

Reads the message content: the AI processes the patient’s message text to understand what they are asking or communicating.
Reviews patient context: depending on your practice’s configuration, the AI may access the patient’s demographics, chronic conditions, allergies, current medications, recent clinical notes, active prescriptions, visit history, and membership status.
Generates a draft reply: based on the message and patient context, the AI writes a suggested response that follows your practice’s communication style.
Presents the draft for review: the draft appears in the conversation thread marked as a suggestion, clearly distinguished from sent messages.

What You Control

You have full control over every AI-drafted reply:

Every draft requires human approval: no AI-generated message reaches a patient without a clinician reviewing and explicitly sending it.
You can edit any draft: change the wording, add details, remove sections, or rewrite entirely before sending.
You can discard any draft: if the suggestion is not helpful, dismiss it and write your own reply from scratch.
You choose which conversations use AI drafts: the AI assistant generates drafts when the feature is enabled for your practice; you decide whether to use each individual draft.
You control what data the AI can access: practice administrators can configure whether the AI has access to patient clinical data or operates with limited context.

What It Does Not Do

The AI assistant has clear boundaries:

Does not send messages: the AI never sends a reply to a patient. Only a clinician can send a message.
Does not make clinical decisions: the AI drafts text for your review; it does not diagnose, prescribe, or make treatment recommendations.
Does not modify patient records: the AI reads patient data for context but cannot create, update, or delete any clinical or administrative records.
Does not access data across practices: all AI context is scoped to the practice you are working in. The AI cannot see patients or data from other practices in your organization.
Does not retain conversation history between sessions: each AI conversation is independent. The AI does not learn from previous conversations or build a persistent profile.

Where to Review What the AI Did

Every AI interaction is logged for audit and transparency:

Conversation history: each AI conversation is stored with all messages, including the system prompt, your questions, and the AI’s responses. Content is encrypted at rest.
Audit log: when the AI accesses patient data, an entry is recorded in the HIPAA audit log showing which user initiated the access, which patient’s data was viewed, and when.
Provider and model tracking: each conversation records which AI provider (Claude or Ollama) and model were used.
Token usage: the system tracks how many tokens each conversation consumed, so you can monitor usage.

To review AI activity, navigate to Settings → AI Assistant or check the audit log under Settings → Audit Log.

How to Adjust AI Behavior

If you want to change how the AI drafts replies:

Enable or disable patient context: practice administrators can toggle whether the AI receives patient clinical data. With patient context disabled, the AI still drafts replies based on the message content but without access to the patient’s medical history.
Switch AI providers: your organization can choose between cloud-based AI (Claude) and a locally-hosted model (Ollama). Local models keep all data on your infrastructure.
Upload reference documents: add clinical protocols, practice policies, and medical references to the AI knowledge base. When document search is enabled, the AI references these documents in its responses and cites which sources it used.
Disable AI features: if your practice prefers not to use AI-drafted replies, the feature can be turned off entirely in practice settings.

How AI Drafts Appear

When the AI generates a draft reply for a patient message, it appears in the conversation thread with a visual indicator distinguishing it from regular messages.

Identifying a Draft

AI drafts are marked with:

A label identifying the message as an AI-suggested draft
A visual style that differentiates it from sent messages (the draft is not yet part of the conversation visible to the patient)
Action buttons to Edit and Send, Send as-is, or Discard

When Drafts Are Generated

The AI generates a draft when:

A patient sends a new message in a conversation.
The AI assistant feature is enabled for your practice.
The conversation is a patient conversation (not a team or broadcast thread).

Drafts are generated asynchronously. There may be a brief delay between when the patient’s message arrives and when the draft appears.

Review and Edit a Draft

When you open a conversation that has an AI-drafted reply:

Read the patient’s message to understand the context.
Review the AI draft below the patient’s message.
Choose one of the following actions:

Send the Draft As-Is

If the draft is accurate and complete:

Select Send on the draft.
The draft becomes a sent message visible to the patient.

Edit Before Sending

If the draft needs changes:

Select Edit on the draft.
Modify the text in the compose area. You can change any part of the message.
Select Send when you are satisfied with the content.

Discard the Draft

If the draft is not useful:

Select Discard on the draft.
The draft is removed from the conversation.
Write your own reply using the standard compose area.

What the AI Considers

The AI builds its draft using information available in your practice’s system. The specific data depends on your practice’s AI configuration.

Message Context

The AI reads:

The patient’s current message
Previous messages in the same conversation thread
The conversation subject line

Patient Record (When Patient Context Is Enabled)

When patient context is enabled, the AI can access:

Demographics: name, date of birth, age, gender
Medical history: chronic conditions, allergies, current medications, surgical history
Recent clinical notes: the last several visit notes, including chief complaint, vitals, assessment, and plan (see Patient Timeline for how these notes are organized)
Active prescriptions: medication names, dosages, directions, and refill status
Visit history: recent and upcoming visits with dates, types, and providers
Membership status: current plan, billing status, and coverage details

Practice Context

The AI also receives general information about your practice:

Practice name and location details
Staff roster with roles
Membership plans and their features
Recent appointment and messaging summary statistics

This context helps the AI generate replies that reference the correct practice name, provider names, and available services.

Document Knowledge Base (When Enabled)

If your practice has uploaded documents to the AI knowledge base and enabled document search for the conversation:

Clinical protocols and guidelines
Practice policies
Medical reference materials

When the AI uses information from uploaded documents, it cites the source in its response.

Training the AI

The AI improves its usefulness over time as your practice configures it. While the AI does not learn from individual conversations, you can shape its behavior in several ways.

Upload Practice-Specific Documents

Add documents to the AI knowledge base to give the AI access to your practice’s protocols and policies:

Navigate to Settings → AI Assistant → Knowledge Base.
Upload documents such as clinical protocols, patient communication guidelines, or practice FAQs.
Documents are processed, indexed, and made available for the AI to reference.

Supported document types include PDFs and text files. Each document is broken into sections and indexed for retrieval.

Configure Patient Data Access

Control whether the AI can access patient clinical data:

Navigate to Settings → AI Assistant.
Toggle the Allow Patient Context setting.
When enabled, the AI receives patient demographics, medical history, clinical notes, and prescription information.
When disabled, the AI drafts replies based on the message text and general practice context only.

Choose Your AI Provider

Your organization can select which AI provider powers the assistant:

Claude (cloud-based): uses a lightweight patient context and can retrieve detailed records on demand through built-in tools. Suitable for practices comfortable with cloud AI processing.
Ollama (local): runs entirely on your infrastructure. Receives the full practice context in each conversation. No data leaves your network.

To change providers, navigate to Settings → AI Assistant → Provider Configuration.

Feedback Through Use

While there is no explicit feedback button, your pattern of using, editing, and discarding AI drafts shapes the value you get from the feature:

Editing drafts helps you identify recurring gaps in the AI’s responses that can be addressed by uploading better reference documents.
Discarding drafts for certain types of messages signals that those topics may require direct clinician authorship.

Need Help?

If you have questions about AI-drafted replies, reach out to the DPC Pro support team at [email protected] or visit the troubleshooting guide.