How to Train an AI Discord Bot on Your Website
· Wallu Team · 4 min read · Tutorials
Point a Discord bot at your website and let it build its own knowledge base. What gets indexed, what to exclude, and how to check the answers are actually coming from your content.
Table of Contents
- Why Your Website Is the Right Source
- What Actually Gets Indexed
- What to Exclude
- Doing It
- Checking the Answers Are Real
- What It Costs
- When Your Site Is Not Enough
- FAQ
Why Your Website Is the Right Source
Your website already contains the answers. Pricing, features, refund policy, shipping times,
compatibility. It is written, reviewed, and someone keeps it current because customers read it.
That last part is what makes it the right source. A knowledge base built from uploaded PDFs
decays the moment someone edits a page and forgets the bot exists. A knowledge base built by
crawling your live site inherits your existing habit of keeping the site accurate.
What Actually Gets Indexed
The crawler follows links from the URL you give it and extracts readable text: headings, body
copy, lists, tables, FAQ entries. Navigation, footers and boilerplate get stripped, because
indexing your cookie banner helps nobody.
Well-structured pages index best. A page with a clear H1, real headings and prose beats a page
that is three columns of marketing fragments. If a human would struggle to extract the answer,
so will the crawler.
What to Exclude
Blog archives, if they contain outdated pricing or superseded advice. The bot cannot tell a
2024 post from today's.
Careers pages, press releases, legal boilerplate. Nobody asks your Discord bot about your
privacy policy, and indexing it dilutes retrieval.
Anything contradicting your current pricing. This is the big one. If an old landing page
still quotes a plan you discontinued, the bot will quote it too, to a customer, as fact.
Before you point a crawler at your site, read it as if you were the bot. Most companies find
something wrong. That is a useful exercise regardless of whether you ship the bot.
Doing It
In Wallu Assist you give it your website URL and it handles the crawl and indexing. There is no
CSV, no question-and-answer pairs to write, and no training period. Once it has crawled, the
same knowledge base serves Discord and every other channel you connect, so you are not
maintaining separate answers for your website widget, email and Instagram.
Then set the per-channel modes so it answers in support and stays silent in off-topic, and
enable knowledge-base-only mode if you want it to refuse rather than improvise. The step-by-step
version is in
making a bot answer from your documents.
Checking the Answers Are Real
Ask it five questions whose answers you know, and check each reply against the page it came
from, word for word. You are looking for answers subtly reworded into something no longer true,
which is the failure mode that survives casual testing.
Then ask something your site does not cover and confirm it declines instead of inventing.
Finally, ask in another language. A Spanish question should get a Spanish answer built from
your English documentation.
What It Costs
| Option | Price | Billing basis |
|---|---|---|
| Carl-bot Premium | $3.99/mo | Per server |
| MEE6 Premium | $11.99/mo | Per server ($499.95 lifetime, all servers) |
| Dyno Premium | $4.99/mo | Per server, tiered 1/3/5. No AI features |
| Intercom Fin | $0.99 per resolution | Per outcome; $49 plan includes 50 |
| Self-hosted RAG bot | $0 licence | Plus hosting and model API costs |
| Wallu Assist | from $4.99/mo | Per account, 350 AI responses, 9 channels |
Prices from each vendor's public pricing page, August 2026.
When Your Site Is Not Enough
Some answers only exist in your support history: edge cases, workarounds, the thing that always
goes wrong on Windows. If members keep asking something your site does not cover, that is not a
bot problem, it is a documentation gap, and the bot has just found it for you.
The useful habit is to review the questions the bot declined to answer. That list is your
documentation backlog, generated for free, ranked by how often people actually ask.
FAQ
How do I train a Discord bot on my website?
Give the bot your site URL and let it crawl. Wallu Assist indexes the content itself and builds
the knowledge base, with no CSV or question-and-answer pairs required.
How long does the crawl take?
Usually minutes for a normal marketing or documentation site. Very large sites take longer, and
you can point the crawler at a documentation subdomain instead of the whole site.
Does it re-crawl when my site changes?
Yes, which is the main advantage over uploading files. Update your pricing page and the answers
follow, rather than needing a manual re-upload.
Will it answer in other languages?
Yes. Wallu Assist detects the language automatically and answers in it, including right-to-left
rendering for Arabic, using the same underlying content.
How much does it cost?
From $4.99/month per account, including 350 AI responses across eight channels. That is per
account rather than per server, unlike MEE6 at $11.99/server and Carl-bot at $3.99/server.
What if my website has outdated information on it?
The bot will repeat it, confidently, to customers. Audit your site before pointing a crawler at
it, particularly old landing pages quoting discontinued prices.
---
Ready to give your Discord community 24/7 AI support? See Wallu Assist pricing or start with Assist
