How to train a chatbot on your website in under 5 minutes

A step-by-step walkthrough of going from "I have a website" to "I have a chatbot that answers questions about my business" — using nothing but a URL.

•By Fabrile Team•5 min read
TutorialGetting StartedKnowledge BaseRAG

The single most common question we get is some variation of: "Do I need to write FAQs? Upload documents? Train a model?"

No, no, and no. If your website is up to date, that's the only training data you need. Here's how to get from URL to working chatbot in under five minutes.

Before you start

You need:

  • A website with the content your customers usually ask about (product info, pricing, support docs, policies).
  • A Fabrile account. Free, no card required.

You do not need:

  • An OpenAI key (we handle the model).
  • Engineering help.
  • A list of "intents" or "utterances." That's the old way.

Step 1 — Create an agent (30 seconds)

In the Fabrile dashboard, click New Agent. Give it a name (your company name is fine), pick a tone, and continue. The agent is created with sensible defaults — you can tweak everything later.

Step 2 — Point it at your website (2 minutes, mostly waiting)

Open the agent's Knowledge Base tab and add a new source of type Website. Paste your URL.

Fabrile will:

  1. Discover the pages on your site (sitemap first, then crawled links).
  2. Show you the list before scraping anything — you can exclude pages you don't want indexed (think /internal/, draft blog posts, legal boilerplate).
  3. Scrape the selected pages, extracting the actual content and skipping nav, footers, and ads.
  4. Chunk the content into retrieval-sized pieces and upload them to the vector store.

For most marketing sites (~30–200 pages), this finishes in a couple of minutes. Larger sites take longer — you can leave the tab and we'll email you when it's done.

Tip: if your site has a sitemap at /sitemap.xml, Fabrile will use it automatically. If you have a docs site separate from your marketing site, add both as sources to the same agent.

Step 3 — Test it (1 minute)

Open the Test tab and start asking the kind of questions your customers actually ask. Some examples:

  • "What's your refund policy?"
  • "Do you integrate with [tool]?"
  • "How much does the team plan cost?"
  • "Is there a free trial?"

You'll see two things in each response:

  • The answer itself.
  • The source — the specific page (or pages) the agent used. Click through to verify.

This is the part where you'll find gaps. If the agent says "I don't have information about X" and you know it's on your site, two things are likely:

  1. The page wasn't included in the scrape (check your include list).
  2. The page is there, but the wording is ambiguous. Rephrase a heading or add a short paragraph that uses the same words a customer would.

Step 4 — Publish and embed (1 minute)

Click Publish to copy the draft config to live. Then grab the embed snippet from the Deploy tab:

<script src="https://fabrile.app/scripts/widget.js?agentId=YOUR_AGENT_ID&tok=YOUR_WEBSITE_TOKEN"></script>

Drop it into your site's footer. The widget appears in the bottom-right corner. If you're on Shopify, Squarespace, or any platform with a native Fabrile integration, skip the script tag and use the integration instead — it's cleaner and survives theme updates.

What if my website doesn't have all the answers?

Common, and easy to fix. You have two options:

  1. Update your website. Often the best move — if your customers are asking, future visitors are too. Add the info to your FAQ or product pages and re-scrape.
  2. Add another knowledge source. Upload a PDF, connect Google Drive, or sync a Notion page. The agent will combine everything.

The point is that you don't have to choose. The website gets you 80% of the way for free, and you can layer on private docs from there.

Cadence after launch

Once the agent is live, check in once a week for the first month:

  • Read 5–10 real conversations.
  • Look for questions the agent didn't answer well.
  • Either add the information or rephrase what's already there.

This loop is where the magic happens. After a month, you'll have an agent that's better at answering your customers' questions than most of your team is, because it has all of your content in its head and infinite patience.

That's it. Five minutes to launch, a few weeks of light tending, and you've replaced a chunk of repetitive support work with something that runs while you sleep.