Train an AI chatbot on your website.
In about a minute.
"Training" sounds like a data-science project. It isn't. There's no dataset to label, no model to fine-tune, no developer to hire. Here's what actually happens between pasting your URL and chatting with a bot that knows your business.
You paste your website URL. The pages are scraped, the text is split into chunks, and those chunks are embedded — turned into numbers a search can match against — so the bot answers questions from your own content. A typical small-business site finishes in about a minute. Better source pages make a better bot, and after launch you improve it by filling the content gaps your visitors reveal.
What the minute
is actually spent on.
Four steps, all automatic. None of them involves you uploading anything or writing a line of code.
Your pages are read
The crawler walks your site from the URL you gave it — homepage, services, about, contact, FAQ — and pulls out the readable text. Navigation, footers and cookie banners are noise; the words that describe your business are signal.
The text is split sensibly
Whole pages are too big to match against a one-line question, so the content is broken into chunks — roughly a topic each. "Do you cover Leeds?" should find the coverage paragraph, not your entire services page.
Chunks become searchable
Each chunk is embedded: converted into a long list of numbers that captures its meaning. Chunks about similar things get similar numbers — which is why "how much is a callout?" can match a paragraph that says "our visit fee is…" without sharing a single keyword.
Questions meet knowledge
When a visitor asks something, their question is embedded the same way, the closest chunks are pulled, and the AI writes an answer grounded in those chunks — your words, not the open internet. Before you ever go live, a pre-flight exam checks the bot can actually do this.
That's the whole trick. No fine-tuning, no months of training — retrieval over your own content, which is why it's fast and why the bot doesn't need to make things up.
The bot is only as good
as your pages.
The pipeline is the same for everyone. What varies is the raw material. Three things separate a useful bot from a shrug:
A clear services page
Say plainly what you do, where you do it and who it's for. "Bathroom fitting across Greater Manchester" trains a bot. "We craft bespoke solutions for discerning clients" trains nothing — the bot can only repeat vagueness back.
Real FAQs
Not filler questions for SEO — the ones customers actually ask on the phone. Do you charge for quotes? How long does a job take? Are you insured? Every genuine question with a written answer is a conversation the bot can now handle at 11pm.
Prices, if you publish them
Visitors ask about cost more than anything else. If your prices are on the site, the bot quotes them accurately. If they're not, it says so honestly and takes the enquiry — it will never invent a figure. Either way beats a contact form.
The test is simple: if a new employee couldn't answer a question from your website alone, neither can the bot. Fix the page and you fix the answer.
Training doesn't end
at minute one.
Launch is when the real training data arrives: actual visitor questions. Three tools turn them into a smarter bot:
Content gaps
Questions the bot couldn't answer are clustered into a "content gaps" list in your portal. It's the most honest market research you'll get — the things visitors wanted that your website doesn't say. More on this in what to do with the questions your chatbot can't answer.
The teach-your-bot box
Type the missing answer once and it's in the knowledge base instantly — no rebuild, no redeploy, no developer. Useful for anything you'd never put on a public page: a minimum job value, a cancellation policy, a trade discount rule.
The weekly digest
One email a week rounds up new gaps, captured leads and how visitors rated the answers. Ten minutes of teaching a week keeps the bot current — accuracy as a habit, not a launch-day project.
Changed your services or prices on the site? Retrain and the knowledge base rebuilds — with the pre-flight exam re-run, so an edit can't quietly break your bot.
Training, asked
plainly.
Do I need to prepare training data or documents?
How long does training actually take?
What if my website is thin or out of date?
Will the bot make things up if the answer isn’t on my site?
What happens when I change my website?
Keep reading: the pre-flight check · why chatbots make things up · pricing · all posts
Your website is already the training data.
Paste your URL and chat with the bot it builds. Free, no card, about a minute — keep it if it earns the place.
Train my chatbot · free