Table of Contents
- Key takeaways: how to make a chatbot at a glance
- What does it take to make a chatbot?
- How to make a chatbot in 7 steps
- Step 1: Define the job the chatbot does
- Step 2: Choose the chatbot type
- Step 3: Choose the build route
- Step 4: Prepare the content the chatbot answers from
- Step 5: Build and configure the chatbot
- Step 6: Test with a fixed question set
- Step 7: Launch the chatbot and measure it
- Worked example: making a chatbot for a bike shop website
- The plan arithmetic for the bike shop
- Watch the content setup in SiteGPT
- Which route should you pick to make a chatbot?
- What SiteGPT does not do
- Common mistakes when making a chatbot
- How to make a chatbot FAQ
- Basics
- What is the easiest way to make a chatbot?
- Can you make your own chatbot without coding?
- What is the difference between a rule-based chatbot and an AI chatbot?
- Are Siri and Alexa chatbots?
- Building
- How do you make a chatbot for your website?
- How do you train a chatbot on your own content?
- Can you make a chatbot with ChatGPT?
- Is it legal to make and use an AI chatbot?
- Cost and time
- How long does it take to make a chatbot?
- How much does it cost to make a chatbot?
- Can you make a chatbot for free?
- Sources
Created by
Do not index
Created time
Aug 22, 2023 04:46 PM
To make a chatbot, define its job, choose a rule-based or AI type, pick a build route, prepare your content, configure it, test it on 30 real questions, and launch it with a bot label and a route to a person.
Most business websites can follow all seven steps without code in an afternoon to a few days. The steps stay the same whether you use a no-code platform, a flow builder, or your own code.
TL;DR
- Start with the job. Write one sentence on who the chatbot serves and list 20 real customer questions before you open any tool.
- Pick the type next. A rule-based chatbot follows paths you design. An AI chatbot answers open questions from your content.
- For a website chatbot that answers from your own pages and files, a no-code platform such as SiteGPT is the shortest route. Starter costs $59 a month billed monthly, or $39 a month billed yearly.
- Fix outdated pages before the chatbot learns from them. A grounded chatbot repeats whatever its sources say.
- Test 30 questions before launch, including 5 the chatbot should decline. Then read the unanswered questions every week.
Key takeaways: how to make a chatbot at a glance
Step | What you do | What you have at the end |
1. Define the job | Write the audience, the task, and 20 real questions | A one-sentence brief and a question list |
2. Choose the type | Pick rule-based, AI, or a hybrid | A type that fits the questions |
3. Choose the route | Match your needs to one build route | One tool or one code stack |
4. Prepare the content | Update pages, remove contradictions, gather files | A clean source set |
5. Build and configure | Add sources, rules, fallback, handoff, and branding | A working chatbot in a test window |
6. Test | Run 30 questions and check each answer against its source | A pass or fail score |
7. Launch and measure | Label it as a bot, embed it, review weekly | A live chatbot and a weekly fix list |
What does it take to make a chatbot?
Making a chatbot takes a clear job, a source of answers, a tool to build it with, and a test before launch. The tool is the smallest part of the work. Most failed chatbots fail on scope or content.
A chatbot is software that holds a conversation in a chat window on a website, in an app, or in a messaging channel. It either follows paths you design or answers in its own words from content you give it.
This guide covers the steps. To compare the tools themselves, see the AI chatbot creator guide. For the four layers that make a chatbot your own (knowledge, behavior, brand, and channels), see custom AI chatbot.
How to make a chatbot in 7 steps
Step 1: Define the job the chatbot does
Defining the job means writing one sentence that names the audience and the task. An example is "answer pre-sales, shipping, and returns questions from website visitors."
- Do this: pull 20 real questions from your inbox, live chat history, sales calls, and site search. Mark each one as "the chatbot answers" or "a person answers."
- Check: every question on the list has an owner. If a question has no answer anywhere on your site, write the answer before you build.
Step 2: Choose the chatbot type
The chatbot type depends on how visitors ask. A rule-based chatbot shows buttons and follows paths you design. An AI chatbot reads your content and answers open questions in its own words.
A rule-based chatbot is predictable, which suits a booking form or a fixed checkout step. An AI chatbot suits pre-sales, policy, and how-to questions, which visitors phrase in many different ways. Many teams use a hybrid: an AI chatbot for questions, plus one short flow for lead capture.
Step 3: Choose the build route
The build route is the kind of tool you build on. There are five common routes, and they differ in code needed, setup time, and what the chatbot can do.
Route | Example tool | Best for | Entry price (USD) | Code needed |
1. No-code AI platform trained on your content | A website chatbot that answers from your pages, files, and cloud documents | Starter $59 a month billed monthly, or $39 a month billed yearly. 7-day trial, card required | No | |
2. AI agent platform with actions | A choice of AI models and actions such as Stripe or Zendesk lookups | Hobby $40 a month billed monthly, or $32 billed yearly. Stripe and Zendesk integrations start on Standard at $150 billed monthly, or $120 billed yearly | No | |
3. Visual flow builder | Landbot | Scripted lead capture with buttons and forms | Free plan with 100 chats and no AI chats. Starter $45 a month billed monthly, or $36 billed yearly | No |
4. Live chat suite with an AI agent | A team that already answers in live chat and wants one inbox | Starter $24.17 a month billed yearly. Lyro AI agent from $32.50 a month billed yearly | No | |
5. Developer platform or your own code | Botpress, or a model API | Custom logic, or a chatbot inside your own product | Botpress Free has 100 conversations. Plus is $150 a month billed annually | Some, or all |
A sixth option, a custom GPT, runs inside ChatGPT for people signed in to it. It suits internal staff, not website visitors. The guide to training ChatGPT on your own data compares a project, a custom GPT, and the OpenAI API.
For first-year totals on each route, including a freelance build on a model API, see chatbot cost. For drag and drop canvases in detail, see drag and drop chatbot builder.
Step 4: Prepare the content the chatbot answers from
Preparing the content means fixing your sources before the chatbot reads them. An AI chatbot repeats what its sources say. If the returns page says 30 days and the FAQ says 14, the chatbot will state one of them with confidence.
- Do this: list every page and file that answers a question from step 1. Update outdated pages, delete drafts, and settle any contradiction.
- Check: each of the 20 questions maps to one current page or file. Write a short answer page for any question that has none.
The chatbot training guide covers how to structure that content so answers come out right.
Step 5: Build and configure the chatbot
Building the chatbot means loading the content, setting the rules, and adding the safety nets. On a no-code platform this part takes minutes to a few hours.
- Add the sources. Enter the website URL or sitemap, upload files, and connect cloud documents.
- Write the instructions. Put the scope, the tone, and the "do not guess" rule in plain lines. Copyable starting sets are in chatbot templates.
- Set a fallback message. Tell the chatbot what to say when the content has no answer, and include a contact route.
- Add a route to a person. Decide who is alerted when a visitor asks for a human.
- Brand the widget. Set the name, colors, welcome message, and a few starter questions.
- Add lead capture if you need it. Ask for contact details only when the visitor wants a follow-up.
Step 6: Test with a fixed question set
Testing means running the same set of real questions before launch and after every change. Use the 20 questions from step 1, add 5 more phrased in new ways, and add 5 the chatbot should decline.
- Do this: check each answer against its source page. Mark it right, wrong, or missing.
- Check: all 25 in-scope answers match a source, and all 5 out-of-scope questions get the fallback message. Fix the content, not only the instructions, when an answer is wrong.
The guide to stopping chatbot hallucinations covers the guardrails for answers that drift from the source.
Step 7: Launch the chatbot and measure it
Launching means embedding the chatbot, labelling it as a bot, and reviewing it every week. Most platforms install with one script tag on your site.
- Do this: say in the first message that the visitor is talking to an AI chatbot. In the EU, Article 50 of the AI Act has required this since 2 August 2026, unless it is obvious.
- Check: every week, read the questions the chatbot could not answer. Add a page or a Q&A for the common ones, then rerun the step 6 test.
The full list of practices for running a chatbot after launch is in chatbot best practices.
Worked example: making a chatbot for a bike shop website
This worked example applies the seven steps to a hypothetical independent bike shop with an online store. The shop has 60 product and policy pages, 1 PDF warranty guide, and one staff member who will run the chatbot.
Step 1, the job. "Answer sizing, shipping, returns, warranty, and service questions from website visitors, and collect contact details for custom builds." Repair quotes and order problems go to a person.
Step 2, the type. Visitors ask open questions such as "will a medium frame fit me at 5 foot 9?" That points to an AI chatbot, with one short lead form for custom builds.
Step 3, the route. The answers already exist as pages and a PDF, and nobody will maintain flows. That points to route 1, a no-code AI platform. The shop uses SiteGPT Starter, with no chat integration because the owner reads leads in the dashboard.
Step 4, the content. The shop finds that the returns page says 30 days and an old FAQ says 14. It deletes the old FAQ and writes a new page on frame sizing.
Steps 5 to 7. The owner adds the website URL and the PDF, writes five instruction lines, sets a fallback message with the shop's phone number, and turns on escalation to a person. The test set has 30 questions, including 5 about other shops' prices that the chatbot must decline.
The plan arithmetic for the bike shop
The content size and the team decide the plan. SiteGPT counts a page as 2,500 cleaned characters of content.
Item | Calculation | Result |
Website text | 60 pages at 3,000 cleaned characters each | 180,000 characters |
Warranty PDF | 1 file at 20,000 characters | 20,000 characters |
Total content | 180,000 + 20,000 | 200,000 characters |
Size in SiteGPT pages | 200,000 divided by 2,500 | 80 pages, which is 8% of the 1,000 pages on Starter |
Files | 1 PDF against the Starter limit of 10 files | Fits |
Starter, billed monthly | $59 x 12 | $708 a year |
Starter, billed yearly | $39 x 12 | $468 a year, which saves $240 (34%) |
Message allowance | As published on the pricing page | 4,000 GPT-4.1-mini messages or 400 GPT-4.1 messages a month |
The character counts are assumptions for the example. Starter fits on size, files, and team. The shop moves to Growth ($129 a month billed monthly, or $79 billed yearly) only if it wants monthly auto refresh, more team members, or a chat integration such as Crisp or Slack.
On Starter, refresh is manual. After the shop changes a policy page, the owner refreshes the source by hand so the chatbot does not repeat the old rule.
Watch the content setup in SiteGPT
This walkthrough shows how to add website pages and other content to a SiteGPT chatbot.
For the full no-code walkthrough, from account to embed code, see how to create an AI chatbot.
Which route should you pick to make a chatbot?
The route to pick depends on how visitors ask, whether the chatbot must take actions, and who maintains it. The matrix below maps each need to one route.
If you need... | Pick... | Why |
A website chatbot that answers from your pages, files, and cloud documents | It reads URLs, sitemaps, files, Google Drive, Dropbox, OneDrive, Box, Notion, GitHub, and YouTube with no code | |
A chatbot that takes actions, such as a Stripe or Zendesk lookup | Chatbase Standard | Every plan includes AI actions, and the Stripe and Zendesk integrations start on Standard |
Scripted lead capture with buttons | Landbot | Its visual builder is made for fixed paths, and it has a free plan |
An AI agent inside your live chat inbox | Tidio | Live chat, ticketing, and the Lyro AI agent sit in one tool |
AI answers inside Slack, Crisp, or Zendesk | Slack and Crisp are on Growth. Zendesk AI Agent starts on Scale | |
Custom logic with developer time | Botpress | Its free plan covers 100 conversations, and developers can add code to flows |
A chatbot for staff only | A custom GPT in ChatGPT | Staff already signed in to ChatGPT can use it with no website install |
A chatbot inside your own app | Your own code on a model API | You control the data flow and the interface |
What SiteGPT does not do
SiteGPT is the route-1 pick here, and it has real limits. It has no free plan, and the 7-day trial needs a credit card. It has no visual flow canvas, so a multi-step scripted flow belongs in a flow builder.
Starter includes 1 member, manual refresh, and no chat integrations. There is no Intercom or HubSpot integration today, and the pricing page lists no payment or order lookup actions. Webhooks start on Scale.
Common mistakes when making a chatbot
The most common mistakes are a vague job, stale content, and no route to a person. Each one below has a fix you can apply in the steps above.
- Starting with the tool. A team that opens a builder first ends up with a chatbot that answers nothing in particular. Write the step 1 sentence and the 20 questions first.
- Training on outdated pages. The chatbot repeats old prices and old policies. Clean the sources in step 4 and refresh them after every change.
- Letting the chatbot guess. Without a fallback message, an AI chatbot fills gaps with plausible text. In Moffatt v. Air Canada (2024), a Canadian tribunal held the airline liable for wrong bereavement fare information its website chatbot gave a customer.
- Hiding the human. Visitors with a billing problem or a complaint need a person. Add the escalation route in step 5 and test it in step 6.
- Skipping the bot label. Visitors should know they are talking to AI. In the EU, Article 50 of the AI Act makes this a legal duty for providers.
- Asking for contact details too early. A form before the first answer blocks visitors who only wanted a quick fact. Ask only when the visitor wants a follow-up.
- Launching without a baseline. Without the step 6 score, nobody can tell whether a change helped. Rerun the same 30 questions after every content or instruction change.
How to make a chatbot FAQ
Basics
What is the easiest way to make a chatbot?
The easiest way is a no-code AI platform that learns from your existing website. You enter your URL, the platform reads the pages, and you embed the chatbot with one script. SiteGPT and Chatbase both work this way.
Can you make your own chatbot without coding?
Yes. All four no-code routes in this guide (a content-trained platform, an agent platform, a flow builder, and a live chat suite) need no code to build or embed. Code is only needed for custom logic or a chatbot inside your own app.
What is the difference between a rule-based chatbot and an AI chatbot?
A rule-based chatbot follows buttons and paths you design and gives fixed answers. An AI chatbot reads your content and answers open questions in its own words. A hybrid uses AI for questions and a short flow for forms.
Are Siri and Alexa chatbots?
Siri and Alexa are voice assistants. They hold conversations, but their main job is to run tasks on a device. A website chatbot answers typed questions about one business.
Building
How do you make a chatbot for your website?
Pick a platform that installs with an embed script, add your website URL or sitemap as a source, configure the instructions and fallback, test it, and paste the script into your site. The AI chatbot for website guide covers placement and setup.
How do you train a chatbot on your own content?
Add your pages, files, and documents as sources on a content-trained platform. The platform indexes them and retrieves the relevant passage for each question. Keep the sources current, because the chatbot repeats what they say.
Can you make a chatbot with ChatGPT?
You can build a custom GPT inside ChatGPT, but it runs for people signed in to ChatGPT, not on your website. For a website chatbot, use a platform or your own code on the OpenAI API.
Is it legal to make and use an AI chatbot?
Yes. AI chatbots are legal, but the business is responsible for what its chatbot tells customers. In the EU, Article 50 of the AI Act requires that people are told they are talking to an AI system unless it is obvious.
Cost and time
How long does it take to make a chatbot?
A no-code AI chatbot can be live in an afternoon if the content is ready. Preparing content and testing usually take longer than the build. A coded chatbot takes days to weeks.
How much does it cost to make a chatbot?
No-code platforms charge a monthly fee and no build fee. SiteGPT Starter costs $59 a month billed monthly, or $39 a month billed yearly ($468 a year). Landbot, Tidio, and Botpress have free plans with low limits. The chatbot cost guide works out first-year totals for each route.
Can you make a chatbot for free?
Yes, with limits. Botpress Free covers 100 conversations, and Landbot Free covers 100 chats with no AI chats. SiteGPT has no free plan, but every self-serve plan starts with a 7-day trial, with a credit card required.
If your answers already live on your website and in your documents, start a SiteGPT trial and run the step 6 test on your own 30 questions before the 7 days end.
Sources
Pages checked on 16 October 2026.
Written by
