No Developer, No Problem? Making a Chatbot Yourself

Type "how to make a chatbot" into Google and count how long it takes before someone shows you code. Usually it's about one scroll, maybe less. Great if you code. If you don't, it reads a bit like a recipe that starts with "first, build an oven."

Take a woman who sells soy candles out of her spare room and runs the whole thing through Telegram.

Can you ship to Germany? When will it get here? Is the lavender one back yet? She answers some version of those three all day, usually while trying to pour wax.

This guide is for her, and for anyone else starting from nothing.

Step 1: The Questions Come Before the Bot

Leave the chatbot builders alone for now. The first job happens on paper, or the back of an envelope, or the notes app you mostly use for shopping lists. You're writing down what people ask you, again and again, week after week.

For the candle seller, that's shipping, delivery times, stock, gift wrapping, and maybe the difference between two scents. Your list will look different, but it probably won't be long.

There's no need to turn it into a project. An evening going through last month's chats is usually enough, since the repeat questions are pretty hard to miss once you look. If someone else answers the messages, ask them, because they'll know straight away, and they're usually a bit fed up with those questions.

People skip this step all the time, and then wonder why their bot feels useless. A bot built from questions customers really send has a good chance of helping. One built from a rough idea of "support stuff" tends to miss what people ask most.

Step 2: Put the Most Common Ones at the Top

Once you've got the list, you'll see that a couple of questions dwarf everything else. The candle seller probably gets "do you ship to…" ten times as often as anything about candle care.

Count roughly how often each one shows up and reorder the list, biggest first. Five or six questions make the cut. Whatever's below can wait in a separate note for a month or so.

The "let's cover everything" plan is the one that sinks most first bots. Ask it about gift wrap, and you get a half-answer. Ask about returns, and you get something from last year's policy.

Customers try twice, shrug, and go back to messaging the owner, and now she's answering the same questions as before, plus a few about why the bot was weird. If the shipping-to-Germany answer is right every single time, people get curious and start trying it with other questions.

Step 3: Write Out Every Answer Word for Word

Write each reply as if you were sending it from your own phone, to a regular customer, on a decent day. So no "Dear valued customer." Two or three sentences per answer is about the limit, because chat bubbles get unreadable past that.

Then there's a second, shorter list of the stuff you'd never hand to a bot.

Anyone complaining, chasing a refund, or clearly upset goes straight to a human. Someone whose order arrived broken doesn't want a cheerful automated message. They want you.

Step 4: Choose a Builder You Won't Dread Opening

This is usually where tutorials switch to code, and where most shop owners quietly decide it's not for them. You can skip that part completely, because plenty of builders now work without any code at all.

For a bot whose main job is answering "do you ship to Germany?" so you don't have to, a no-code builder is plenty.

SUCH is one I'd look at if you're on Telegram, since it's built for exactly this. You type in your questions and answers, and when your prices change in January, you just go in and edit them yourself.

Step 5: Launch Small

Set the bot up with your five or six questions from Step 2 and nothing else. Send it a few test messages from your own phone, ask a friend to try to confuse it, and see what happens.

You'd think customers would be annoyed by "I'm not sure about that one, let me get someone for you." Most of them aren't. What does annoy them is a bot that sounds very sure of itself and gets the answer wrong.

You'll probably feel the pull to add a few more scenarios before launch, then a few more after that.

It's worth holding off. A bot that answers five questions well starts saving your team time on day one, while a bot still waiting on fifty edge cases saves nothing until it's finished, if it ever is.

Step 6: Keep an Eye on What It Can't Answer

After launch, the interesting part is the questions the bot gets stuck on. For the first month, look at those once a week.

You'll almost certainly spot something you forgot. Maybe it's the "will it arrive before Christmas?" rush in December, or a returns rule that only ever existed in your head.

That's fine, and it happens to pretty much everyone the first time round.

What helps most is boring. A single phone note titled something like "bot didn't know," where you paste every question it got stuck on.

Once a week, maybe Sunday with a coffee, you go through it. Anything that turned up twice gets an answer written for it, and one-offs get deleted. After a few months of that, the replies start sounding a lot like you, which is the whole idea.

What People Mean by "Chatbot"

"Chatbot" covers a strange amount of ground. According to Wikipedia's overview of chatbots, the term stretches from basic rule-based programs that react to keywords all the way to advanced conversational AI.

A candle shop needs the basic end of that. A rule-based bot only knows what you've typed into it. Ask it something off-script, and it can't improvise, so it won't cook up a delivery date or hand out a free-shipping deal you never agreed to.

People call that a drawback. For questions about money and dates, though, I'd honestly rather have the one that can't make things up.

Maybe someday you'll want the clever kind. Many small sellers never get there and seem perfectly happy.

Getting the First Version Live

By now the hard part's mostly done, and none of it involved code. You've got a list of questions, a handful of answers written in your own words, and a tool to put them in.
Switch it on with those few answers and see what people ask over the next week or two. That will tell you what to add next.