How an AI chat answers about price and availability: where the data comes from and why it can be stale

October 13, 2026

Of all a shopper’s questions, the one about price and availability is the most dangerous for an AI assistant. A slip in a product description is not a disaster, but a wrong price means an argument and an unhappy customer. So it matters to understand where the bot gets this data and when it can’t be trusted.

Where prices and availability come from

The bot doesn’t guess and doesn’t remember prices “in general”. When you enter your site address, it assembles the catalog from one of these sources:

An XML (YML) feed. For every product it has a price, an old price (if discounted) and an availability flag. This is the most reliable source.
The WooCommerce API. Price and “in stock” status come straight from the store.
schema.org markup on product pages. Price and availability are read from the page. If the markup is incomplete, the bot won’t see part of the data.

The bot states the price verbatim from the data it found and does not recalculate it.

Where price and availability can be wrong

The catalog was read long ago. The bot knows the state of the site at the time of study. If you changed a price, a promotion ended or a product sold out, and the catalog wasn’t re-read, the bot will answer from old data. Scheduled refresh is planned for the dashboard; for now, re-read the catalog after significant changes.

The price depends on the variant. For products with sizes and colors the price is sometimes set only on variations. The bot may not see the price of the main product. Check a few such products in the test chat.

A product has no price. “Price on request” or an empty field: the bot will notice this and won’t invent anything. It will say the price needs to be confirmed.

Discounts and personal prices. Prices for registered shoppers, promo codes and promotion terms that aren’t in public data are unknown to the bot.

Availability in a particular city. If you have different stock by warehouse and city but the feed has one shared flag, the bot can’t tell them apart.

How to protect yourself from mistakes

Compare before launch. Choose 10 products, including variable and discounted ones, and compare the bot’s answers with the product pages.
Keep the feed current. If the feed refreshes automatically, that is the best option.
Don’t promise too much in shipping descriptions. Terms that depend on the region are better described in plain text in a “Shipping” section; then the bot answers from it.
Agree who watches the leads. When the bot doesn’t know the answer it offers to leave contacts, and a manager must reply in a reasonable time.

What the bot says when it isn’t sure

The rule is simple: no data, no number. The answer will be “unfortunately I don’t have the exact price” and an offer to leave contacts. For a store that is safer than a nice but wrong answer.

The general pre-launch check is described in How to test an AI chat before launch. If you connect on Tilda or WooCommerce, see the integrations pages: they say where exactly your platform’s price comes from.

Frequently asked questions

Can the bot state a wrong price?

It states the price from the data it read when it studied your site. If you changed the price later and the catalog wasn't re-read, the bot will state the old one. So refresh the catalog after price changes and compare the bot's answers with your site before launch.

What does the bot say if a product has no price?

It names the product but won't state a price, says honestly that it needs to be confirmed and offers to leave contacts for a manager.

Does the bot see stock in the warehouse?

It sees the availability flag that the source provides: the available field in an XML (YML) feed, the "in stock" status in WooCommerce, or availability markup on the page. Sources rarely give exact quantities.