"Training" a chatbot sounds technical. It is not. For a modern website chatbot, training simply means giving it the right content to read, checking what it learned, and filling the gaps. If you can organise a folder of documents, you can do this.
This guide walks you through the whole process, from choosing which pages to include to keeping your chatbot accurate months from now.
What "training" a chatbot really means
When you train a chatbot on your website, you are not teaching it to think. You are building it a library. The chatbot reads your pages and documents and stores them in a way that lets it find the right passage quickly. When a visitor asks a question, it looks up the most relevant passages and writes a short answer from them.
That has two big consequences:
- The quality of answers depends on the quality of your content. Clear, current, specific content gives clear, current, specific answers.
- Gaps in your content become gaps in the chatbot. If something is not written down anywhere, the chatbot cannot know it.
So most of the work is really about your content, not about the chatbot.
Step-by-step: training a chatbot on your website
Step 1: Decide what the chatbot is for
Before you add anything, write down the five to ten jobs you want the chatbot to do. For example:
- Answer questions about prices and packages.
- Explain opening hours, location and parking.
- Explain how to book and how to cancel.
- Answer delivery and returns questions.
- Take enquiries for quotes.
This list tells you which content matters most, and it becomes your test list later.
Step 2: Choose the right website pages
Start with your public website. Most chatbot tools can read your pages for you, either one by one or from your sitemap (a page that lists all your website's pages). If you are not sure whether you have a sitemap, our free sitemap checker will find it and tell you whether it works.
Include pages that answer customer questions:
- Home, about and contact pages.
- Service or product pages.
- Pricing pages.
- FAQ, delivery, returns and policy pages.
- Location or branch pages.
Leave out pages that add noise:
- Old news posts and expired offers.
- Thin pages with only a heading and a photo.
- Duplicate pages, like print versions or tag archives.
- Login pages, carts and account pages.
Step 3: Add documents that hold the details
A lot of useful information never makes it onto a website. It sits in a price list, a menu, a terms PDF or a welcome pack. Upload those too. Patchwhistle accepts PDF, Word (.docx), plain text, Markdown and CSV files.
Good candidates:
- Full price lists or rate cards.
- Menus, including allergen information.
- Terms and conditions, cancellation and refund policies.
- Product spec sheets.
- Onboarding guides or "what to expect" leaflets.
A CSV file works well for structured information, such as a product list with sizes and prices, or a list of branches with addresses and hours.
Step 4: Paste in the things nobody wrote down
Some knowledge lives only in your head. Think of the questions you answer on the phone without looking anything up. Paste short notes for these, or better, add them as questions and answers. Examples:
- "Is there parking?" Yes, free parking behind the building for up to two hours.
- "Do you take card?" Yes, all major cards and contactless. No cheques.
- "Can I bring my dog?" Dogs are welcome on the terrace, not inside.
Questions and answers are the most precise way to teach a chatbot something, because you control the exact wording.
Step 5: Add videos if you explain things on video
If you have YouTube videos that explain your product, show how to use something, or answer common questions, you can add those as well. The chatbot learns from the video's captions, so check the captions are accurate first, especially for product names and numbers.
Step 6: Review before anything goes live
Once your content is added, look through what the chatbot has learned. Check that the right pages came in and the wrong ones did not. Remove anything outdated. In Patchwhistle, nothing goes live until you review and publish it, so you can take your time here.
Step 7: Test with real questions
Now use the list from Step 1. Ask each question the way a customer would, including vague and misspelled versions. A good test set includes:
| Type of question | Example | What a good answer looks like |
|---|---|---|
| Simple fact | What time do you close on Friday? | The correct time, plus a link to the page |
| Price | How much is a full set of nails? | The price from your list, with any conditions |
| Policy | Can I return something I bought in a sale? | Your actual returns rule, not a generic one |
| Process | How do I book? | The steps and the booking link |
| Off-topic | What is the capital of Peru? | A polite note that it can only help with your business |
| Not covered | Do you do home visits? | An honest "I do not have that information" and a way to reach you |
The last two rows matter as much as the first four. A chatbot you can trust is one that admits when it does not know.
Step 8: Fix the gaps
Every weak answer points to a fix:
- Wrong answer: usually an outdated page or document. Correct the source.
- Vague answer: the content is too general. Add a specific question and answer.
- No answer: the information is missing. Add it.
- Mixed-up answer: two sources disagree, for example an old and a new price list. Remove the old one.
Then test again. Two or three rounds is normal.
Step 9: Go live and watch the first week
Add the chatbot to your website and check in daily for the first week. Read some real conversations. You will learn quickly how customers phrase things, which is often different from how you would.
Keeping your chatbot up to date
Training is not a one-off job. Your business changes, so the chatbot needs to keep up.
Refresh when your website changes
If you change prices or hours on your website, the chatbot needs to read the new version. Some plans do this automatically. In Patchwhistle, the Care plan refreshes weekly and the Complete plan refreshes daily. Otherwise, refresh by hand after any important change.
Use the questions it could not answer
This is the single most useful habit. When the chatbot cannot answer something, Patchwhistle puts that question in an Unanswered inbox. Once a week, go through it and add the missing answers as questions and answers. Over time, the gaps shrink.
Look at what people ask about
Your dashboard shows what visitors ask about most and when they are busiest. If lots of people ask about something your website barely mentions, that is a sign to improve the website itself, not just the chatbot.
Retire old content
When an offer ends or a service is discontinued, remove the related page or document. Old content is the most common cause of wrong answers.
Common training mistakes
- Adding everything. More content is not always better. Old blog posts and duplicate pages can muddy answers.
- Skipping the review. Always check what came in before you publish.
- Testing only easy questions. Test the awkward ones too.
- Writing answers that are too long. Short, specific questions and answers work best.
- Forgetting about other languages. If your customers speak several languages, test a few questions in each. See our guide to a multilingual website chatbot.
A simple weekly routine
Once your chatbot is live, ten to fifteen minutes a week is enough for most small businesses:
- 1Open the Unanswered inbox and add answers for anything new.
- 2Skim a few conversations to check answers are still accurate.
- 3Update any page or document that changed this week.
- 4Glance at what people asked about most.
Start training your own chatbot
If you want to try this on your own website, Patchwhistle walks you through each step: add pages, documents, text, questions and answers, and YouTube videos, review everything, then publish. It answers only from your content and tells you what it could not answer.
You can start a 14-day free trial with no card. Plans start at $19 a month; see Pricing. For a guided tour, read the Patchwhistle walkthrough, or start with the bigger picture in our practical guide to AI chatbots for small business.
Frequently asked questions

Written by
Akash Limbani
Akash Limbani builds Patchwhistle, the AI chatbot that learns from your website and turns visitors into enquiries. These guides share practical, plain-language advice for small businesses on chatbots, websites and winning more customers online.
Try it on your own website
Start 14-day free trial