How-to

How to build an AI chatbot without writing code

The build itself takes about an hour. What decides whether the bot helps you or embarrasses you happens in the thirty minutes before you start and the twenty minutes before you publish.

By Casper LöfgrenPublished Jul 22, 20268 min readRevdot prices as of November 2026, excluding VAT.

Flat-pack furniture parts laid out on the floor

Building an AI chatbot used to mean guessing every question a customer might ask and hand-writing an answer to each one. Most people gave up long before the list was finished. That is why so many company websites ended up with a chatbot that could do three things badly.

Modern tools work the other way round: they read the website you already have and draft the answers from it. See how retrieval-based bots work for the mechanism. Your job has moved from writing answers to describing the job and checking the result.

This guide is vendor-neutral until the last section, and none of the seven steps below needs code.

Before you open any tool: thirty minutes of preparation

1. Decide what the bot is for

Pick one primary job. A bot that tries to do everything does none of it well. Three options are realistic:

  • Deflect repeat questions: opening hours, delivery, returns, warranty. Easy to build, easy to measure, useful in the first week.
  • Capture leads: qualify the visitor and collect contact details. Worth more per conversation, and it needs more thought about what to ask.
  • Help people choose: recommend the right product from a range. Hardest to get right, and the biggest payoff when the catalog is genuinely confusing.

Write the one you picked on a sticky note. When you are tempted to add a fourth capability, look at the note.

2. Find the questions you already answer

You do not need to invent these questions. They already exist in three places, and collecting them takes about fifteen minutes:

  • Your support inbox. Sort by subject and read the last two weeks. The top five repeats are your bot's core job.
  • Your site search log. Especially the searches that returned nothing: those are questions your website fails to answer today. If that list is long, the problem may be search itself rather than the chatbot.
  • Whoever answers the phone. Ask them what people ask. It takes two minutes and it is consistently the best source of the three.

3. Fix the two or three pages the bot will lean on

A retrieval-based bot answers from your published content. If your delivery page is vague, the bot is vague about delivery. The difference is that it is now vague to every visitor instead of one.

This is the step people skip and later blame the bot for. You do not need to rewrite the site: check that delivery, returns, pricing and contact are accurate and unambiguous. That is usually the whole job.

Everything above happens before you open a single tool. Do only this, then pick a tool at random, and you will still get a better bot than someone who chooses carefully and skips the preparation.

Building it: about an hour

4. Describe the job in plain words

Modern tools take a written description instead of a decision tree. Write it the way you would brief a new hire on their first morning: who you are, who is asking, what to do, what to leave alone.

A description that works looks like this:

"We sell e-bikes in Portland. Answer warranty and service questions, recommend a bike based on how the customer plans to ride it, and ask for contact details if they want a test ride. Do not quote delivery dates. Send those to the shop. Keep answers to three sentences."

Scope, job, exclusion, length: one working bot.

Four things make that brief work: a scope, a job, an explicit exclusion and a length. Vague descriptions produce vague bots. The exclusion is the part people leave out, and telling the bot what not to answer matters more than you would expect.

5. Let it read your site, then check what it built

The tool indexes your pages and drafts the answers. Read them. You are looking for three things:

  • Answers that are right but off-tone: fix those in the description, not one answer at a time.
  • Answers that are confidently wrong: almost always the wrong source page. Find the page, fix the page.
  • Questions it refuses that it should handle: usually the content genuinely is not on your site. That is a website gap the bot just found for you.

6. Try to break it

Twenty minutes here separates a bot you trust from a bot you only hope is right. Ask it these five:

  1. Something your website does not cover. It has to say it does not know. If it improvises, stop and fix that before anything else. This is the single most important test.
  2. A pricing question with a trap. "Can I get a discount if I buy three?" Check that it does not invent a policy.
  3. A complaint. "This arrived broken and I want my money back." It should route to a human, not try to resolve it.
  4. Something off-topic. "What's the weather?" It should decline politely and steer back rather than answer.
  5. The same question three ways. Consistency across phrasings is what your customers actually experience.

7. Publish narrow

Put it on one section first: the support pages, or a single product category. A week of real conversations on a small surface teaches you more than a month of internal testing, and a bad answer reaches far fewer people.

Revdash builds your bot from your site address: Revdash AI reads your pages, builds the look, writes the instructions and emails you the link in 10 minutes. Nothing is installed on your site until you decide. Prices start at 39 € a month for 300 conversations, and the price is the same whether you build the bot yourself or we build it for you.

Try it on your site → See pricing

After publishing: the part that compounds

A bot is not finished at launch. It is at its worst on day one. Two habits change that, and both take about fifteen minutes a week:

  • Read the unanswered questions. Every tool worth using collects the questions it could not handle. That list is your bot's to-do list and your website's content gaps in one place.
  • Read five real conversations. Not the metrics: the transcripts. You will find tone problems and near-misses that no dashboard surfaces.

How to tell it is ready for the whole site: it correctly refuses things it does not know, it has handled a week of real traffic without an embarrassing answer, and the unanswered-question list has stopped growing faster than you can clear it.

What building an AI chatbot without code does not remove

Three honest caveats, since "no code" is often sold as "no work":

  • Your content is still the ceiling. No tool makes a bot better than the material it reads. A thin website produces a thin bot.
  • Someone has to own it. Fifteen minutes a week, indefinitely. A bot that nobody reads goes stale within a couple of months.
  • Some integrations still need a developer. Putting one snippet on a page does not. Pushing leads into a bespoke CRM, or reading a custom product feed, might.

The whole process, with times

Seven steps, one afternoon. The column worth reading twice is the last one.

StepWhat you doTimeWhere
1Decide the one job the bot has5 minBefore any tool
2Collect the questions you already answer15 minBefore any tool
3Fix the two or three pages it will lean on10 minBefore any tool
4Describe the job in plain words10 minIn the tool
5Read what it drafted, fix the source pages30 minIn the tool
6Try to break it with five questions20 minIn the tool
7Publish to one section only10 minIn the tool
Total100 minHalf of it outside the tool

About a hundred minutes end to end, and half of that sits outside the tool. Preparation and testing are the blocks that get cut when the afternoon runs short. They are also the two that decide whether the bot is any good.

Frequently asked

How long does it take to build an AI chatbot without code?+

Roughly an hour of building, plus about thirty minutes of preparation beforehand and twenty minutes of testing before you publish. The preparation, meaning collecting your real repeat questions and fixing two or three key pages, makes more difference to the result than the tool you choose.

Do I need to write question-and-answer pairs?+

Not with a retrieval-based tool. You write a plain description of what the bot should handle, and it reads your existing website to draft the answers. Older rule-based builders still require hand-written pairs, which is the main reason those projects stalled.

How do I train a chatbot on my own website?+

You point the tool at your site and it indexes your pages, then answers from that content. "Training" in this context means checking the drafted answers and correcting them, usually by fixing the underlying page rather than the answer itself, so the fix benefits human visitors too.

What should I test before publishing a chatbot?+

Five things: a question your site does not cover (it must say it does not know), a pricing question with a trap, a complaint (it should route to a human), something off-topic, and the same question phrased three ways. The first is the most important: a bot that improvises when it lacks information will eventually improvise something costly.

Can a non-technical person maintain it?+

Yes, and it should be a non-technical person, ideally whoever answers customer questions today. Maintenance is reading the unanswered-question list and a handful of real conversations each week. Only custom integrations typically need a developer.

The summary

Collect the questions you already answer, fix the pages the bot will lean on, then describe the job in four or five sentences including what not to do. Spend twenty minutes trying to break it before a customer does. The tool matters less than that sequence.

Steps 1 to 3 stay yours. We can do steps 4 to 7 for you: we read your site, write the description and try to break the bot before your customers do. Then we publish it narrow, and you check the answers before it goes live. Building it for you costs 290 € one time (490 € on Business and Scale), and 0 € on annual billing. The subscription is the same either way: Starter 39 € a month for 300 conversations, Pro 79 € for 1,000, Business 149 € for 2,000 and Scale from 249 € for 4,000. Ordering straight from the dashboard opens with the Revdash launch; until then you order with us, on the same terms. Prices are in euros, excluding VAT.

Try it on your site → See pricing

Related reading: how retrieval-based bots work and what AI chatbots actually cost.

Try it on your site

Your own bot tells you more than any article.

Enter your site address, and Revdash builds a bot from your content and emails you the link in 10 minutes. Prices are public, from 39 € a month.

What is your main goal?

Revdash reads your site · knowledge base · bot in your inbox

The Revdot team answers in person:WhatsAppBook a call+358 45 231 1420eelis@revdot.ai