The costliest mistake a shopping chatbot can make is a confident one: quoting last week’s price, or offering a product that sold out this morning. Most of these errors come from four causes, and each has a different fix.
Cause 1: the chatbot is answering from a snapshot
Most chatbots copy your catalogue into their own knowledge once and answer from the copy. A copy is only right until something changes. A price edit or a sale ending makes it wrong, and nobody is told.
The fix is to separate two jobs. A snapshot is a good way to find the right product when a shopper describes what they want. It is a poor way to know what that product costs and whether it is in stock. Price, stock and availability should be read from the store when the question is asked. TalkTill does this for WooCommerce through the public Store API, and the assistant is instructed to prefer that answer over its own knowledge for those three things.
How to tell: change the price of a test product, then ask the chatbot about it straight away. If it quotes the old price, it is answering from a copy.
Cause 2: a leftover flag makes a simple product look variable
WooCommerce products often carry a “used for variations” flag left over from earlier edits. On one real catalogue, 23 of 88 products had it. A chatbot that trusts the flag asks shoppers to pick a size that does not exist and breaks Add to cart on best-sellers.
The fix is to treat a product as having options only when it is genuinely a variable product, not when a flag says so.
How to tell: ask for a product you know has no options. If the chatbot asks you to choose a size or colour, it is reading the flag.
Cause 3: sizes sold as separate products
Many stores sell 200g and 500g as two products rather than one product with options. A chatbot that treats them as unrelated offers one, or shows two near-identical cards. The better behaviour is to recognise the family and join the sizes into one card with a size choice, so the shopper switches size without leaving the conversation and the cart receives the right product.
How to tell: ask for a product you sell in two pack sizes. You should see one card, not two.
Cause 4: titles full of HTML entities
WooCommerce can return titles like Mahani Pickle – 300g. Left alone, that reaches the shopper
as written. The fix is to decode the entities before they are shown. The store’s own search may not
find such a product by its full name either, so a lookup that retries without the brand and pack size
finds it anyway.
How to tell: look at a product card for any title with a dash, an ampersand or an accent in it.
A quick way to check any chatbot
Pick three products: one on sale, one sold out, and one with several sizes. Ask about each in the chat and compare the answer with the product page. Any difference tells you which of the causes above you are looking at.
The setup guide covers connecting a store so price and stock are read live, and the WooCommerce integration page lists what the assistant can do with a connected store.
Found something here that is out of date, or wrong? Tell us, a guide with a review date on it is a promise to keep it current.