How to make an AI chatbot, step by step
A plain-language guide to building a chatbot that answers from your own content: the no-code and code routes, real free plans as of October 2026, and how to keep it from making things up.
How do you make an AI chatbot?
Decide the one job your chatbot does, such as answering customer questions, then choose a no-code builder like Botpress or Tidio, or call an AI model’s API from your own code. Give it your own help content to answer from, set rules for what it must not say, add a handover to a person, and test it before launch.
- Search demand
- About 450 US Google searches a month for “how to make an ai chatbot” (Ahrefs, October 2026)
- Two routes
- A no-code chatbot builder, or your own code calling an AI model’s API
- Free to start, no code
- Botpress: 100 conversations a month and 3 AI agents free; Tidio: first 50 AI conversations free (October 2026)
- What makes it accurate
- Answering from your own documents (retrieval, or RAG), plus a clear “I don’t know” rule
- Must-have
- A handover to a person for anything the bot can’t or shouldn’t answer
What is an AI chatbot, and how does it work?
An AI chatbot is a chat window powered by a large language model that reads a person’s message and writes a reply in natural language. Unlike older chatbots with fixed buttons and scripts, it understands questions phrased any way and can answer ones you never wrote out.
Behind the window, three things decide how good it is: the model that writes the replies, the instructions that set its job and limits, and the knowledge it answers from. Most of the work in a good chatbot is the last two. If it also takes actions, like booking or refunding, it is closer to an agent: see how to build an AI agent.
- Customer support: answers questions about orders, accounts and policies, and hands the rest to a person.
- Sales and lead capture: answers product questions on your site and collects contact details from interested visitors.
- Bookings: explains services and times, and points people to your booking page.
- Expert assistant as a product: a coach, tutor or advisor chatbot that people pay to use.
- Internal helper: answers staff questions from your handbook and procedures.
Should you build a chatbot with a builder or with code?
Use a no-code builder if the chatbot sits on a website and answers questions from your content. Write code if the chatbot is the product itself, needs your own data and logic, or must look and behave exactly your way.
| Your situation | Better route | Why |
|---|---|---|
| Support or FAQ bot on your website | No-code builder | Upload your help content, paste a snippet on your site, done |
| You don’t code | No-code builder | Visual setup and a ready chat widget |
| The chatbot is what customers pay for | Code, or an app builder with sign-in and payments | You need accounts, usage limits and billing around the chat |
| It must read your database or private systems | Code | Builders connect to common apps; anything custom needs code |
| Many conversations a day | Code | Builder plans charge per conversation or credit; with code you pay the model provider directly |
You can also start on a builder and move to code later. Keep your instructions, your source documents and your list of test questions in your own files, so moving means rebuilding the plumbing, not rewriting what makes the bot good.
What tools can you use to build a chatbot?
Here are common options, from fully no-code to code, with free starting points checked on each company’s own pricing page or docs on October 24, 2026.
| Tool | Route | Free or starting option |
|---|---|---|
| Botpress | No-code / low-code | Free plan: 100 conversations a month, 3 AI agents, 3 seats; Plus is $150 a month billed yearly |
| Tidio | No-code | Free plan; its Lyro AI agent includes the first 50 conversations free, once |
| Chatbase | No-code | No free plan listed; 7-day trial; Hobby is $32 a month billed yearly with 700 message credits |
| OpenAI API | Code | Pay per use; its file search tool lets the model answer from files you upload |
| AI SDK by Vercel | Code | Free, open-source TypeScript library for chat interfaces and AI apps; you pay your model provider |
With any code route, the model provider bills you per use. For what that looks like across providers, see what AI model calls cost per user.
Whichever you pick, judge it on three things before price: how easily you can update its knowledge, how the handover to a person works, and whether you can read every conversation. Those decide how good the bot gets after launch.
How to make an AI chatbot, step by step
- 1
Write down the one job
For example: “Answer questions about our classes, prices and bookings, and hand anything else to a person.” List the 20 questions people ask most, from your inbox or DMs.
- 2
Gather the content it should answer from
Your help pages, price list, policies and past answers. Fix anything out of date first: the bot will repeat whatever you give it.
- 3
Choose a builder or a code route
Use the decision table above. On a no-code builder, start on the free plan. On the code route, pick one model provider and one library.
- 4
Write the instructions
Set the job, tone and audience. Tell it to answer only from your content and to say it doesn’t know otherwise. Say what it must never do, such as promise refunds or give medical advice.
- 5
Add your knowledge
Upload documents or connect your help center so the bot retrieves the right passage before replying. This is what keeps answers specific to your business.
- 6
Set up the handover
Decide when a person takes over: when the customer asks, when the bot isn’t confident, or for topics like refunds and complaints. Make sure someone is notified.
- 7
Test with real questions
Ask your 20 common questions, then try to break it: off-topic questions, angry messages, questions in another language. Fix the instructions or content after each wrong answer.
- 8
Launch and read the conversations
Add it to your site or app, read the first week of conversations, and turn every unanswered question into new content.
How do you make a chatbot answer from your own content?
You make a chatbot answer from your own content with retrieval: before each reply, the system searches your documents for the most relevant passages and gives them to the model along with the question. This is called retrieval-augmented generation, or RAG, and it is how most “train it on your data” chatbots work.
- It is not retraining the model: you are giving it the right page to read at the right moment.
- Short, clear documents work better than long ones: one topic per page, with the question in the heading.
- Keep the source up to date: change the document, and the bot’s answers change with it.
- Ask the bot to say which document it used, so you can trace wrong answers back to their source.
Anthropic’s guide to building effective agents makes the same point from the other side: for many applications, a single model call with retrieval and good examples is enough, without a complex agent.
A simple test shows whether retrieval is working: ask a question whose answer appears in only one document, then change that document and ask again. If the answer changes, the bot is reading your content. If it doesn’t, the bot is answering from the model’s general knowledge, and you need to fix the setup before launch.
How do you stop an AI chatbot from making things up?
You can’t make made-up answers impossible, but you can make them rare and harmless. Limit the bot to your content, give it a clear way to say “I don’t know,” and keep anything with consequences behind a person.
| Problem | Example | Fix |
|---|---|---|
| Invented facts | Quotes a price or policy you never had | Answer only from your documents; tell it to say “I don’t know” otherwise |
| Promises it can’t keep | “Sure, I’ll refund you today” | Forbid commitments on refunds, discounts and dates; hand those to a person |
| Off-topic use | Writing essays or code for visitors on your bill | Keep it to your topic and cap messages per person per day |
| Risky advice | Medical, legal or financial guidance | Refuse and point to a professional; add a visible note |
| No way out | The customer keeps asking for a human | Always offer a handover, and notify someone when it happens |
Test these failures on purpose before launch. Ask about a product you don’t sell, demand a refund, ask for medical advice and ask to speak to a person. Each should end in a clear, safe reply or a handover, never a confident guess.
Where should your chatbot live, and how do you know it works?
Put the chatbot where people already ask the question: on the pricing and help pages of your website, inside your app next to the feature it explains, or in your support inbox as a first reply. A chatbot on a page nobody visits answers nobody.
You know it works when it resolves questions without a person and without complaints. Look at real conversations every week, not just the dashboard.
| Measure | What it tells you | What to do if it’s poor |
|---|---|---|
| Questions answered without handover | How much work the bot saves | Add content for the questions it hands over most |
| Wrong or made-up answers | Whether people can trust it | Tighten instructions and fix the source documents |
| Customer rating after the chat | Whether people leave satisfied | Read the low-rated chats first |
| Cost per conversation | Whether it pays for itself | Shorten replies, cap messages, or use a smaller model for simple questions |
Can you sell an AI chatbot as a product?
Yes, and many AI products are a chatbot with expert knowledge inside: a nutrition coach, a language tutor, an interview practice partner. To sell one, you need more than the chat: sign-in so each person has their own history, payments, a daily usage limit so one user can’t run up your model bill, and support.
That is where an AI app builder with the business included saves time. For the full process from idea to paying users, read how to build an AI app.
Be clear with users about what the chatbot is: say it is an AI, say what it can and can’t help with, and add a note on advice topics such as health or money. Clear limits build more trust than a bot that pretends to know everything.
Make an AI chatbot people can sign up and pay for
Starty is our product. Describe the chatbot in one sentence, such as a meal-planning coach or a tutor, and Starty builds it as an app with sign-in, payments, emails, legal pages and analytics, all in test mode until you review.
- AI capabilities from a catalog, including agents, with no API keys and no model to set up
- Each user’s plan, AI usage and AI cost in one view
- A customer agent sees each user’s plan, payments and AI usage, drafts the reply and waits for your approval
- Starty Payments built in for subscriptions or packs, with no Starty fee on your revenue
Related guides
Where every fact comes from
Official pricing pages, docs, announcements and named studies, checked on November 10, 2026. Where a company doesn’t list something, we say so rather than guess.
Questions people ask
Start on a free plan. As of October 2026, Botpress gives 100 conversations a month and 3 AI agents for free, and Tidio’s free plan includes its first 50 AI conversations once. Free plans are enough to build and test; real traffic usually needs a paid plan.
Yes. No-code builders let you upload your content, write instructions in plain language, test in a preview and paste a chat widget onto your website. You only need code if the chatbot must reach private systems or is the product you sell.
Upload your documents or connect your help center in the builder, or use retrieval in your own code. The bot then searches your content before each reply. This doesn’t retrain the model; it gives the model the right passage to read.
Building can cost nothing on a free plan. Running it costs either a builder subscription, usually priced by conversations or credits, or the AI model’s per-use fees if you code it yourself. Cost grows with the number and length of conversations.
You can make a chatbot that uses the same kind of model through an API, with your own instructions and content. You can’t easily recreate the model itself, which takes huge amounts of data and computing power. Most products build on top of existing models.
A simple FAQ chatbot on a no-code builder can be live in a day if your content is ready. Most of the time goes into writing clear content, testing real questions and fixing wrong answers, not the setup.
Enter your email to join Starty. We’ll email you your invite, and you can start building for free.
No. You describe what you want, or fork an idea, and Starty builds it. You review and approve; you never have to open the code.
No. Starty takes no fee on your revenue. Payouts go to your bank account, and Starty earns from plans and credits.