Ask a good sales associate for "a table" and they won't read you the whole catalogue. They'll ask one thing first: for the kitchen or the garden? Then maybe one more: how many people? A store chat should work the same way. This post explains how Claspwell decides what to ask, why it searches your Magento categories before anything else, and what you can do in your catalog to make those questions sharper.
Category first, then search inside it
When a shopper writes a message, Claspwell first works out which Magento category the request belongs to. Only then does it search for products, and it searches inside that category, typically with no more than two catalog queries per turn. Your nested category tree stays intact, so "garden tables" doesn't get mixed up with "table lamps" just because both contain the word table.
This matters because a flat keyword search across the whole store returns noisy results, and a noisy result list is exactly what makes shoppers close the chat.
When the assistant asks a clarifying question
Once the category is clear, the assistant looks at what is still open. If the shopper's request leaves an important choice undecided, it asks about one shopper-facing attribute, such as material, size or load class, instead of showing a long, unsorted list. A typical exchange looks like this:
- Shopper: "I need shelving for my garage."
- Assistant: "Sure. Roughly how much weight should each shelf hold?"
- Shopper: "Heavy stuff, tool boxes."
- Assistant: shows product cards for heavy-duty shelving from your catalog, with real prices and links.
The goal is to narrow the choice in as few turns as possible. If the shopper has already given enough detail, the assistant skips the question and goes straight to products.
Why not just let the model ask whatever it wants?
A generic GPT chatbot can ask clarifying questions too, but it doesn't know which attributes your store actually has. It might ask about a colour you don't stock or a size range that doesn't exist. In Claspwell the questions come from the attributes synced from your Magento catalog, and the products shown afterwards are pulled from that same catalog in code, not written by the model. So every option the assistant offers is one a shopper can actually buy.
How to get better clarifying questions
The assistant can only ask about what your catalog describes. A few practical checks:
- Keep categories meaningful. A clean category tree is the first filter. Huge catch-all categories make the first step less precise.
- Fill in the attributes shoppers decide on. If material or load capacity is missing on half your products, the assistant can't use it reliably to narrow results.
- Use consistent values. "Oak", "oak wood" and "OAK" as three separate values split one choice into three.
- Mark attributes as shopper-facing on purpose. Internal codes and supplier fields shouldn't become questions.
Our attribute checklist for AI product discovery goes through this in more detail, and the post on three-tier AI routing explains when the model is involved at all.
Where Claspwell fits
Claspwell is a catalog-aware AI sales assistant for Magento 2 and Hyvä. During the install we map your attributes with you and run the first catalog sync, typically in 3 days to 1 week once scope and access are ready. See the features, or request an integration.