Setting Up Your AI Chat Agent: A Complete Configuration Guide

Your AI Agent Is Only as Good as Its Configuration

A Tellop AI agent can answer questions, collect intake data, and guide potential clients through your workflow around the clock. But a poorly configured agent creates confusion, gives vague answers, and frustrates the very people you are trying to convert. This guide walks you through every configuration step to build an agent that sounds like you, knows your practice, and stays on track.

Step 1: Name Your Agent

Navigate to your procedure's Agent Settings and start with the basics. Your agent's name appears in the chat interface, so pick something that fits your brand:

• Professional practices: "Dr. Chen's Assistant" or "Rivera Legal Intake"
• Agencies and studios: "Pixel Studio Bot" or "Atlas Intake Assistant"
• Simple and clear: "Booking Assistant" works if you prefer something neutral

The name sets the tone for the entire conversation. A client chatting with "Dr. Chen's Assistant" immediately understands this is an automated helper connected to a real professional.

Step 2: Write the System Prompt

The system prompt is the most important piece of your agent's configuration. It tells the AI who it is, how to behave, and what it knows. Think of it as a detailed onboarding document for a new front-desk employee.

Here is an example of a strong system prompt:

"You are the intake assistant for Greenfield Family Therapy. You help potential clients learn about our therapy services, answer common questions, and guide them through the intake process. You are warm, empathetic, and professional. You never provide therapeutic advice, diagnoses, or crisis intervention. If someone expresses an immediate safety concern, direct them to 988 (Suicide and Crisis Lifeline) and let them know a human will follow up. You speak in clear, jargon-free language."

Compare that with a weak prompt:

"You are a chatbot for a therapy practice. Answer questions and collect info."

The difference is dramatic. The strong prompt gives the agent context, tone, boundaries, and a safety escalation path. The weak one produces generic, unhelpful responses.

Step 3: Build Your FAQ Knowledge Base

Your FAQ entries are the factual foundation your agent draws from. For each entry, provide a question and a thorough answer. Here are examples across different practice types:

Medical practice:

• Q: "Do you accept Blue Cross insurance?" A: "Yes, we accept Blue Cross Blue Shield PPO and HMO plans. We also accept Aetna, Cigna, and United Healthcare. For other providers, please mention your plan during intake and we will verify coverage."

Legal firm:

• Q: "How much does an initial consultation cost?" A: "Our initial case evaluation is $150 for a 45-minute session. This fee is applied toward your retainer if you decide to move forward with representation."

Creative agency:

• Q: "What is your typical project timeline?" A: "Most branding projects take 4-6 weeks from kickoff to final delivery. Website projects range from 6-10 weeks depending on complexity. We will provide a detailed timeline during the project scoping phase."

Add as many entries as you can. The more your agent knows, the more confidently it answers — and the fewer conversations need human intervention.

Step 4: Set Conversation Boundaries

Boundaries define the edges of what your agent will discuss. This is critical for professions where overstepping could be harmful or create liability.

Configure your boundaries to specify:

• Topics the agent can discuss: Your services, pricing, availability, the intake process, general information about your field
• Topics the agent must avoid: Specific diagnoses, legal advice, financial guarantees, competitor comparisons, anything that requires professional judgment
• Redirect phrases: What the agent says when asked about off-limits topics. For example: "That is a great question for your appointment with Dr. Patel. I can help you get that scheduled."

Well-defined boundaries do not make your agent less helpful. They make it reliably helpful within the right scope.

Step 5: Configure Escalation Triggers

There are moments when a conversation needs a human. Escalation triggers tell your agent when to stop and hand off. Common triggers include:

• Urgent safety concerns — The client mentions self-harm, abuse, or an emergency
• Complex questions — The inquiry falls outside your FAQ entries and boundaries
• Client frustration — The conversation is going in circles or the client is unhappy
• Explicit requests — The client asks to speak with a human

For each trigger, define what the agent should say. A clear, empathetic handoff message makes the client feel heard, not dismissed.

Step 6: Test Your Agent

Before sharing your invite link, run through several test conversations. Try to break it:

1. Ask a FAQ question — Does it respond accurately with the right details?
2. Ask an off-limits question — Does it redirect gracefully?
3. Go through the full intake flow — Is the process smooth and logical?
4. Be vague — "I have a problem" should prompt the agent to ask clarifying questions, not guess
5. Be demanding — "I need to see the doctor right now" should trigger an appropriate response about the process

Take notes on where the agent stumbles. Each stumble is a clue to improve your prompt, FAQ entries, or boundaries.

Step 7: Iterate Based on Real Conversations

Once your agent is live, review actual conversations from your dashboard. Look for:

• Questions the agent could not answer — Add them to your FAQ knowledge base
• Moments where the agent went off track — Tighten your boundaries or system prompt
• Drop-off points — Where clients stopped responding. Was the agent too wordy? Too vague?
• Successful intakes — What did those conversations have in common? Reinforce those patterns

The best agents are not configured once and forgotten. They evolve as you learn what your clients actually ask and need.

Good vs. Bad Agent Prompts at a Glance

Aspect	Good Prompt	Bad Prompt
Identity	"You are the intake assistant for Greenfield Family Therapy"	"You are a chatbot"
Tone	"Warm, empathetic, and professional"	(not specified)
Boundaries	"Never provide diagnoses or therapeutic advice"	(not specified)
Escalation	"Direct safety concerns to 988 and flag for human follow-up"	(not specified)
Language	"Speak in clear, jargon-free language"	(not specified)

Every missing element in a bad prompt is a gap the AI fills with generic behavior. Be explicit about what you want.

Final Thoughts

Your AI agent is the first interaction many clients will have with your practice. Investing an hour in thoughtful configuration — writing a clear system prompt, adding comprehensive FAQs, setting firm boundaries, and defining escalation paths — pays off in every single conversation that follows. Start detailed, test thoroughly, and refine continuously.