How to Train Your First Convolyn Chatbot
Your bot is only as good as what it knows. A practical loop for giving Convolyn the right content, testing it like a customer, and refining from the real questions it misses.
Your chatbot is only as good as what it knows. Convolyn doesn't make up answers — it responds from the content you give it. So "training" your bot really means giving it the right material, in the right shape, and refining it over time. Here's how to do it well.
What "training" means in Convolyn
There's no machine-learning step you have to manage. When you add a knowledge source, Convolyn breaks it into small passages ("chunks") and creates embeddings — a way of finding the most relevant passage for any question. When a visitor asks something, the bot retrieves the closest matching content and answers from it. This is called retrieval-augmented generation, and the practical upshot is simple: good source content in, grounded answers out.
Step 1 — Start with your highest-value content
Don't try to upload everything at once. Begin with the content that answers the questions people actually ask:
- Your FAQ or help articles
- A product or pricing sheet
- Your "About" and "How it works" pages
- Common pre-sales questions and their answers
A focused, accurate 5-page knowledge base beats a sprawling, contradictory 50-page one.
Step 2 — Choose the right source type
Convolyn accepts three kinds of sources:
- PDF / text files — great for brochures, manuals, price lists.
- Pasted text — fastest for short, specific answers (hours, policies, contact info).
- Website content — point it at pages and let Convolyn pull the text.
Mix and match. You can add, edit, or remove sources at any time without rebuilding anything.
Step 3 — Write content the way people ask
The bot answers best when your source content mirrors how visitors phrase things. Instead of a dense paragraph, structure key facts as clear question-and-answer pairs:
Do you offer refunds? Yes — within 14 days of purchase, no questions asked.
Short, direct, factual content retrieves more cleanly than marketing prose.
Step 4 — Set the response style
Under your bot's settings, choose how it replies:
- Recommended (Bot Mastery) — the bot adapts answer length to the question: short for simple asks, fuller when the question needs it, always grounded in your content.
- Fixed lengths — pin it to one line, two lines, a short paragraph, moderate, or long if you prefer a consistent feel.
Most businesses should start with Recommended and adjust only if they want a specific tone.
Step 5 — Test like a visitor
Open your bot and ask it the questions your customers ask — including the awkward ones. If it answers wrong or says it doesn't know, that's not a failure; it's a signal. Either the content is missing or it's written in a way the bot can't match. Add or rewrite that passage and test again.
Step 6 — Improve from real conversations
Once your bot is live, Convolyn's analytics surface the questions it couldn't answer. This is the single most useful list you'll get — it tells you exactly what content to add next. Train, watch, refine: every cycle makes the bot more useful and more likely to convert visitors into leads.
The takeaway
Training a Convolyn bot isn't a one-time setup — it's a short loop: add your best content, test it like a customer, and feed it the real questions it misses. Do that a few times and you'll have a bot that genuinely sounds like your business and answers like your best team member.
Start training your bot — you can add your first source in under five minutes.