How to Create Your First AI Customer Support Agent: A Step-by-Step Guide

Setting up an AI customer support agent sounds like it should be complicated. It isn’t, but it does help to know what you’re actually deciding at each step, instead of clicking through a setup wizard and hoping for the best. This guide is both: a decision guide for the choices you’ll make before you build anything, and an implementation guide for actually building, testing, and publishing your first AI chat or voice agent.

Watch the full walkthrough below, or follow the written guide underneath.

1. Do you need an AI customer support agent?

Short answer: yes — if your business has a website, a phone line, or any channel where customers ask questions you’re not always available to answer. We’ve made the fuller case for this already: read Why Every Small Business Website Needs 24/7 AI Customer Support if you want the full argument. The short version is that every hour your business isn’t actively staffed is an hour a customer’s question goes unanswered, and an AI customer support agent closes that gap without asking you to hire around the clock.

If you’re already sure you need one, skip ahead. If you’re still deciding, that post is the place to start.

2. Voice or chat? Choosing the right AI agent for your business

Once you’ve decided to build an AI support agent, the next decision is which channel to start with — and this is a real setting, not just a philosophical question. When you create a new agent, you’ll set its channel to Chat, Voice, or Both. That single toggle is where the decision you make actually gets built.

A few questions make it easier:

• How do your customers currently reach you? If most contact comes through phone calls, Voice (or Both) is probably where the coverage gap actually is. If it’s mostly website visits and messages, Chat is the more natural fit to start.
• How much are you ready to take on for a first agent? Chat is the lower-cost, lower-commitment starting point — simpler to test, cheaper to run, and easier to evaluate before you turn on voice.
• How urgent are the questions you’re getting? Voice suits time-sensitive, in-the-moment questions — someone trying to book an appointment today, or checking if you’re open right now. Chat suits questions people are comfortable typing and waiting a moment for.

You’re not locked in either way — you can build an agent as Chat and switch it to Both later once you’ve seen it perform. Most small businesses start with Chat, get comfortable with how the agent handles real questions, and turn Voice on once they’re confident in it.

3. Building your agent: system prompt, name, and personality

To actually create an agent, you’ll go to Agents → New Agent → Start from scratch. From there, three things define who your agent is:

• The system prompt — this defines the agent’s identity: its personality, its tone, how it addresses and answers questions, what functionality it should use, and general instructions on how it should handle a conversation. This is the single most important field, because everything the agent does traces back to it.
• A name — give your agent an actual name (in our own setup, we used “Priscilla”). Referencing that name inside the system prompt itself — “You are Priscilla, VoiceLink’s support agent” — is what makes the agent introduce itself consistently and sound like a specific person rather than a generic bot.
• The channel setting — Chat, Voice, or Both, as covered above.

Write the system prompt the way you’d brief a new employee on their first day: who they are, how they should sound, what they should always do (confirm details before acting, ask clarifying questions when something’s ambiguous), what they should never do (guess at pricing, promise exceptions, pretend to know something they don’t), and when they should hand off to a human instead of continuing. Vague instructions produce vague, inconsistent answers. Specific ones produce an agent that sounds like it actually works there.

4. Knowledge base: teaching your agent about your business

Your AI agent is only as useful as what it knows, and the knowledge base is where you give it that information. In practice, this is often as simple as a document you already have — a brand overview, a product summary, an FAQ page — pasted directly into the agent’s knowledge base field. The goal is to give it what a new prospect or customer would actually want to know: what your business does, how your product or service works, and answers to the questions people ask most.

Treat this as an ongoing job, not a one-time upload. Start with the questions you get asked most often, document the answers clearly, and add to it as new questions come up. If your agent doesn’t know something, it should say so and hand off — never guess. An accurate, current knowledge base is the difference between an agent that genuinely helps and one that frustrates customers with confident wrong answers.

5. Tools: letting your agent do more than talk

Tools are what let your agent take action instead of just answering questions — checking a real order status, booking an actual appointment slot, processing a payment. This is where an AI support agent stops being a smarter FAQ page and starts being genuinely useful. Not every business needs every tool on day one — start with whatever action your customers ask for most and expand from there.

6. Testing your agent before you publish it

Before your agent goes live anywhere, you can preview it — either as an inline conversation or as a widget, the same format it’ll appear in on your site. This is where you actually talk to it: ask it the questions a real prospect or customer would ask, and see whether it answers the way you intended. In our own test run, we asked the agent what it could do for a SaaS business, what use cases it supported, and whether it could connect to a phone number — and it walked through sales qualification, support, onboarding, billing questions, and telephony support, then offered a clear next step (explore the site or book a demo).

This step matters because it’s the cheapest place to catch a problem — a wrong answer, an off tone, a missing piece of knowledge — before a real customer ever sees it. Test both the questions you expect and a few you don’t, since those edge cases are where a weak system prompt or thin knowledge base shows up first.

7. Publishing and embedding your agent on your website

Once you’re happy with how it performs, publish the agent. From there, go to Sharing and enable Embed — this generates an embed code you can drop into your website. You’ll also see an option to collect guest information (name, email, message) before a conversation starts; turn it on if you want that context up front, or leave it off if you’d rather keep the barrier to starting a conversation as low as possible.

The embed code itself is just HTML, so it works the same way whether your site runs on WordPress, a React app, or plain HTML — anywhere you can drop in a script, you can drop in the widget. Once it’s live, visitors can open it and start a real conversation with your AI chat agent immediately, no rebuild or redesign required.

For a closer look at this step — including platform-specific notes for WordPress and React, and how to think about the guest information toggle — see How to Embed an AI Chat Widget on Your Website.

8. Connecting your AI agent to a phone number

Bringing your agent to voice means connecting it to a phone number through telephony, either a dedicated number for the agent, or your existing business line. Once connected, your AI voice agent answers inbound calls the same way a person would: handling the conversation, working from the same system prompt and knowledge base as your chat agent, and escalating to your team when a call needs a human.

Because it draws on the exact same setup as your chat agent — same identity, same knowledge, same rules for when to hand off — you’re not building a second agent from scratch. You’re extending the one you already built to a second channel.

Where to start

If you’ve read this far, you likely already know whether chat or voice is the right first move for your business, and what your system prompt and knowledge base need to cover. The setup itself is quick — most of the real work is in getting the system prompt and knowledge base right, which is exactly what this guide (and the video above) walks through.

Try the live demo to talk to a fully built AI agent yourself, sign up to start building your own, or book a walkthrough if you’d rather have us build the first version with you.

Still deciding if you even need one? Start with Why Every Small Business Website Needs 24/7 AI Customer Support.