The AI answers from the product's knowledge base, and the knowledge base is built\
from your product content. So a bad answer is almost always one of these three\
things.

## The answer describes an older version of the product

Check the **Products** page. If the badge says **Needs update**, you've changed the\
title, description, vendor or product type since it was indexed, and the knowledge\
base still reflects the old version.

Regenerate that product. It costs one AI generation.

## The answer is vague or talks around the question

The detail isn't in your product content. Indexing only reads your title,\
description, vendor and product type — it doesn't go and research your product. If\
the description never mentions what the strap is made of, regenerating won't\
conjure it up.

Add the missing detail to the product description in Shopify, then regenerate and\
check the result on the product's **Knowledge base** tab.

## The answer is confidently wrong

Look at the **Knowledge base** tab for that product. Usually the spec sheet has\
picked up something ambiguous in the description — a measurement with no unit, or a\
compatibility note that can be read two ways.

You can fix it either way. Editing the knowledge base directly is quicker and it\
sticks, though a later regeneration will replace it. Clarifying the description and\
regenerating takes longer but fixes the underlying problem, and improves the product\
page while you're at it.

## If the shopper used an external provider

If they went to ChatGPT or Claude rather than Arus AI, that provider generated the\
answer. The app sends them your product's details to work from, but what they do\
with it happens outside your store, and it won't appear as a transcript in your\
Questions list.

If you want control over the answer, make Arus AI your default provider. See\
[Arus AI — answering shoppers inside your store](/hc/askanyai/en/articles/arus-ai-chat).

## Still wrong

Email [support@aruslabs.co](mailto:support@aruslabs.co) with the product URL and the question that\
produced the bad answer.