In this article
- What data can you train an AI chatbot on?
- How do you build it, in five steps?
- Step 1: How do you create your Dante AI account?
- Step 2: How do you add your data?
- Step 3: How do you check what it learned?
- Step 4: How do you make it sound like your business?
- Step 5: How do you put it on your website?
- How long does this actually take?
- Where does ChatGPT fit?
- What should you test before you launch?
- Common mistakes when training on your own data
- What changes after you launch
- Train your agent on your own data today
How do you build an AI chatbot trained on your data?
Use a no-code platform that does the training for you. Create a free Dante AI account, paste your website URL or upload your files, test the answers in the sandbox, then paste one script tag into your site. Your AI agent is live and answering visitors from your own material in under 60 seconds.
What data can you train an AI chatbot on?
Almost anything you have already written down:
- Your website, crawled page by page
- Files: PDFs, documents, spreadsheets, slide decks
- Help centre and knowledge base articles
- Product, delivery and pricing pages
- Pasted text, for policies that live in someone's head
You do not need to clean any of it first. The agent retrieves the relevant passage at answer time rather than memorising your documents, so a messy page costs you far less than a missing topic. If your knowledge sits mostly in files, see training an AI chatbot on your documents. If it is scattered across tools, start by building an AI knowledge base.
How do you build it, in five steps?
Five steps, all of them in one dashboard, no engineering involved.
Try it yourself. Train an AI agent on your website, docs, or files. Live in 60 seconds. No code needed.
Step 1: How do you create your Dante AI account?
Start by creating your free Dante AI account. No card is required. The free plan includes 100 message credits per month, and up to 700 during onboarding, which is enough to train an agent and interrogate it properly before you decide anything. Signup drops you straight into the knowledge step, so the first thing you do is hand over your data.
Step 2: How do you add your data?
Paste your website URL and the crawler walks the site page by page. Upload files alongside it for anything that never made it onto the site, and paste raw text for the rest.
One thing matters more than volume: crawl depth. Let it go past the homepage and the top navigation. Customer questions get answered on inner pages, in delivery terms, in the fourteenth help article. An agent trained only on top-level pages sounds confident and is useless.

For a longer walkthrough of source selection, see how to train a chatbot on your own data.
Step 3: How do you check what it learned?
This is the step almost everyone skips, and it decides whether the agent is any good. Open the sandbox and ask the ten questions your customers actually ask. Not clever ones. The boring ones: what does it cost, do you ship to my country, how do I cancel.
- Right answer, right source. Nothing to do.
- Thin or vague. The content exists but is buried. Add that page or file as its own source.
- No answer. You never wrote it down. Paste it in as text, and you have just found a gap on your website too.
When we look at agents that answer well against agents that do not, the difference is almost never the model. It is whether the crawl reached the pages holding the answers.
Step 4: How do you make it sound like your business?
With the data in place, customisation takes a couple of minutes:
- Set the tone and the opening message
- Turn on lead capture so useful conversations leave you a contact
- Enable human handoff for questions that should reach a person
- Add scheduling buttons pointing at your existing booking tool
- Choose the model, including the GPT-5.6 tiers
Changing the model does not require retraining.
Step 5: How do you put it on your website?
Copy the embed snippet and paste it into your site builder, CMS or theme. That is the whole deployment.
If you want the chat bubble in the corner rather than a full page agent, see how to add AI chat to your website.
How long does this actually take?
With your website live and your files to hand:
- Account creation: under 1 minute
- Pointing it at your site: under 30 seconds
- Crawling and indexing: depends on your page count
- Sandbox testing: 10 to 20 minutes if you do it properly
- Embedding: 2 minutes
Training is the fast part: hand it your site and it is answering in under 60 seconds. The twenty minutes of honest testing is what turns a working agent into a good one.
Where does ChatGPT fit?
It is one route, and it solves a different problem. A custom GPT, a Project, or a GPT shared across a Business or Enterprise workspace gives your team a private assistant that reads your uploaded files. Useful internally. It does not answer your customers, because it sits behind a login rather than on your site, so a visitor reading your pricing page at 11pm never reaches it.
Fine-tuning is not the answer either. It teaches a model a style or a pattern rather than your facts, and every price change means another training run. Retrieval, where the model reads your own content at answer time, is what a customer-facing agent needs. Use a custom GPT for your team and an embedded AI agent for your visitors. Plenty of businesses run both.
What should you test before you launch?
- Your ten most common customer questions, in customer language
- One question you know is not answered on your site, to see how it declines
- Pricing and policy answers, checked against the real page
- Human handoff, triggered on purpose
- The widget on mobile
Every conversation is logged, so the first week of real questions tells you which sources to add next.
Common mistakes when training on your own data
- Crawling only the homepage, then blaming the model for shallow answers
- Uploading an outdated price list and never replacing it
- Adding hundreds of irrelevant files, which makes retrieval noisier, not smarter
- Rewriting the prompt to paper over a missing source
- Launching without reading a single conversation log
What changes after you launch
- Visitors get an answer in seconds instead of a contact form
- Repetitive support questions stop reaching your inbox
- You see, in the logs, what people actually want to know
- Conversations that show buying intent leave you a contact
- A well trained agent answers most customer questions on its own, and the rest reach you with the context already attached
Build and test on the free plan, then move to a paid plan from $40 per month when your volume grows.
Train your agent on your own data today
You do not need a project plan, a brief, or a developer. The whole loop is four moves:
- Create a free account and paste your website URL
- Let the crawl run past the homepage, and write down your ten most common customer questions while it works
- Ask all ten in the sandbox, and add the missing source wherever an answer is thin
- Paste the embed snippet into your site
Train an AI agent on your own data for free, then ask it the questions your customers asked you last week and judge the answers yourself. If the answers are good, embedding takes two minutes. If they are not, you have just found the gaps on your own website, which is worth knowing either way.