How to build a customer support chatbot (no code)

A support chatbot is one of the most useful — and most beginner-friendly — things you can build with AI. It answers the same common questions your customers ask over and over (“Where’s my order?”, “What’s your return policy?”) so you don’t have to, in a voice that actually sounds like your business. And you can build a great one without writing a single line of code.

The secret is a good system prompt — a clear set of standing instructions that tells the assistant who it is, what it does, and the rules it should follow. That’s exactly what the AI Prompt Builder creates for you. Let’s use it.

What we’re building

We’ll build a support assistant for an imaginary online shop, “Fern & Field,” that sells houseplants. It should answer questions about orders, shipping, and returns, sound warm and reassuring, and know when to hand off to a human. Swap in your own business as we go — the steps are identical.

Two minutes of prep

Before touching the tool, jot down four quick things. This is the thinking that makes the difference between a generic bot and a genuinely helpful one:

• Its job: answer common order, shipping, and return questions.
• Its voice: friendly, calm, and concise.
• Its rules: always confirm the order number first; never promise refund dates it can’t verify.
• A handoff line: when it can’t help, point the customer to a real person.

Building it in the tool

Open the AI Prompt Builder in another tab and follow along. You can start from the Customer support bot template (in the Popular row) and adjust it, or fill the fields in from scratch like this:

1. Who is it? Give the assistant a clear identity.
   Who is it?a warm, capable customer-support assistant for Fern & Field, an online houseplant shop
2. What should it do? State its job in a sentence or two.
   What should it do?Answer questions about orders, shipping, and returns. Solve the problem in the first reply when possible, and stay reassuring when customers are frustrated.
3. Pick a tone. Tap the chips that fit. For support, Friendly, Warm, and Concise is a reliable combo.
4. Add your “always” rules. These keep it on the rails.
   AlwaysConfirm the order number before giving order-specific details · End by checking the issue is fully resolved · If you can’t help, point the customer to support@fernandfield.com
5. Add your “never” limits. These prevent the answers you’d regret.
   NeverPromise refunds or delivery dates you can’t verify · Blame the customer · Make up policy details
6. Show one example. This teaches tone better than any rule — add a sample question and the ideal reply.
   Example — userWhere’s my order? It was supposed to arrive yesterday.
   Example — replyI’m so sorry it’s late — that’s frustrating. Could you share your order number so I can check exactly where it is right now?

As you fill each field, watch the panel on the right assemble your system prompt in real time, and the “prompt strength” meter fill up as you add a role, tone, a limit, and an example.

The finished prompt

When you’re done, the tool produces a clean, structured prompt like the one below. Hit Copy (or Download) to grab it:

Putting it to work

Now you just need somewhere to paste it. A few beginner-friendly options:

• A Custom GPT. Paste the prompt into the “Instructions” box when you set up a Custom GPT, then upload your FAQ or policies as knowledge so it answers from your real details.
• Your existing chatbot tool. Most website chat widgets have a “system prompt,” “persona,” or “instructions” field — paste it there.
• Just a chat window. Even without any setup, you can paste it at the start of a conversation to get consistent, on-brand answers you can copy back to customers.

Making it yours

The version above is a starting point. A few easy upgrades once it’s running:

• Add real policy details to the “always” rules (your actual return window, shipping times) so it stops being generic.
• Add more example exchanges for your trickiest questions — each one sharpens the tone.
• Test it like a customer: ask the awkward questions and tighten any rule where it wanders. Two or three rounds turns a decent bot into a dependable one.

Frequently asked questions