How to train a chatbot on your own website and PDFs (no coding)

You can train an AI chatbot on your own business in minutes: point it at your website, upload your PDFs, and it answers customer questions from that content. You do not write scripts or type out question and answer pairs. This guide explains how that works, what to feed it, and how to fix wrong answers.

How "training" actually works

Despite the word, you are not training a model from scratch. Your content is split into small chunks, converted into numerical representations, and stored in a searchable index. When a visitor asks something, the system finds the chunks most relevant to the question and gives them to the language model, which writes an answer from that material.

Two things follow from this. The bot can only answer from what you gave it, and fixing a wrong answer means fixing the content, not rewriting prompts. That is also why a content-trained bot is easier to maintain than a scripted one.

Step 1 - Add your website

Enter your website address and start a crawl. The crawler reads your sitemap and follows internal links to collect pages, up to the page limit of your plan (15 on the free plan, 100 on Starter, 500 on Growth, 2,000 on Pro). If your site renders content with JavaScript, FiveMinChat re-crawls it in a headless browser so the text is still captured.

Step 2 - Upload documents

Upload anything customers ask about that is not on your site: price lists, brochures, menus, syllabi, policy documents, product sheets. Supported formats are PDF, DOCX, CSV, XLSX and TXT, with a 5 MB limit per file on Free, Starter, and 10 MB on Growth, 25 MB on Pro.

Use text-based files. A PDF exported from Word or Canva works well; a photograph of a printed page saved as PDF has no readable text for the bot to learn from.

Step 3 - Write content the bot can use

Answer quality depends mostly on content quality. A few habits make a big difference:

• State facts in plain sentences. "Admission fee is Rs 5,000 for the 2026 batch" beats a table image.
• Put prices and timings in text, not images. A banner with fees baked into the picture cannot be read.
• Keep one source of truth. If an old PDF says Rs 4,000 and the site says Rs 5,000, the bot may quote either.
• Add the questions you hear on calls. A short FAQ page covering refunds, timings, locations and "do you do X" is the highest-value page you can add.
• Remove outdated pages or delete old uploaded files so they stop being used.

Step 4 - Test with real questions

Write down the ten questions customers ask you most. Ask all ten in the preview. Mark each answer as correct, incomplete or wrong. For incomplete and wrong answers, find the page that should have contained the answer. Either it is missing the detail or it says something different. Fix the page or file, re-crawl, and ask again.

Check how it handles what it does not know

Also ask a few things your content does not cover. A good setup says it does not have that information and points the visitor to contact you, instead of guessing. If your business has a WhatsApp number, add it in Settings so the widget can offer a "Continue on WhatsApp" link.

Step 5 - Keep it current

Training is not one-off. When prices, timings or offers change, update the page or file and re-crawl. Paid plans re-crawl on a schedule (monthly on Starter, twice a month on Growth, weekly on Pro) and include manual re-crawls each month (2 on Starter, 5 on Growth, 10 on Pro). After each crawl the bot also refreshes its suggested opening questions from your latest content.

Common mistakes

• Crawling a site that is mostly images, then wondering why the bot knows little
• Uploading a 200-page catalogue when customers only ask about ten products
• Leaving old offers on the site after they expire
• Judging the bot on one odd question instead of the ten most common ones

Frequently asked questions

Can I train a chatbot on a PDF?

Yes. Upload a text-based PDF and the bot learns from its content alongside your crawled website pages. PDFs, DOCX, CSV, XLSX and TXT files are supported, up to the file size limit of your plan.

Do I need to write questions and answers manually?

No. The bot learns from your pages and documents directly. You only need to make sure the content contains the information customers ask about.

What if the chatbot gives a wrong answer?

Find the page or file that should hold the correct answer. Usually it is missing, outdated or conflicts with another source. Fix the content, re-crawl, and test the question again.

How often should I update the chatbot's knowledge?

Whenever prices, timings, policies or offers change. Paid plans also re-crawl automatically on a schedule, and you can trigger a manual re-crawl at any time.