When your AI agent gives a wrong answer, check the page it read

A shopper asks whether their order ships free. Your agent says free shipping starts at $50, the shopper's cart is at $46, and they close the tab. Your real threshold is $40. You lowered it in the summer and updated the shipping page, but the FAQ still says $50, and that's the page the agent read.

When a grounded agent gets an answer wrong, the page behind it is usually what's wrong. The answer links its source, so you can open that page, find the old sentence and fix it the same day.

Made-up answers and out-of-date ones

A model with nothing to go on can make up something plausible. Grounding deals with that: the agent answers from your pages and says it doesn't know when they don't cover the question. We cover that in our post on why AI chatbots make things up.

An out-of-date answer is harder to spot, because the agent did its job. It repeated your own content accurately, and the content was old. No model upgrade changes that, since the fix is a sentence on your own site.

Product answers come from your live catalog

On a connected Shopify or WooCommerce store, the agent reads products, prices and variants from your catalog when the shopper asks, so product answers reflect the store as it is. Policies work differently. Shipping rates, return windows, delivery cut-offs and size advice live on pages the agent indexes, and an index is a snapshot of those pages.

That's why the re-crawl schedule matters. Agentmatica re-indexes your pages on a schedule, and the Re-crawl button in the dashboard refreshes them straight after a change. Fix the FAQ on Tuesday, press Re-crawl, and the next shopper who asks gets the $40 threshold with the page linked under the answer.

How Agentmatica keeps the answer selling

Back to the shopper at $46. With the FAQ fixed, the agent tells them free shipping starts at $40, so their order already qualifies. Put the threshold in Sales notes too, and when a shopper is a few dollars short the agent can suggest something from your catalog that takes them over it, shown as a product card. On Shopify and WooCommerce it adds the item to the cart and hands over a checkout link.

When nothing on your site answers a question, the agent says it doesn't know and offers to bring in a person. That conversation lands under Needs attention in Insights, along with replies that got a thumbs-down and chats that escalated.

Turn the misses into a weekly fix list

Open Needs attention once a week and fix the page behind the top item. Each fix helps the agent and every shopper who reads that page without asking.

Start with the pages that touch money: shipping rates, free-shipping thresholds, return windows and anything with a price in it. Check them against what you actually charge, then press Re-crawl.