How do you handle unsupported/risky product descriptions?

One field nobody’s mentioned yet: the structured data. Most themes and SEO apps write a Product JSON-LD block with its own description, and some add FAQ schema from a metafield. That text isn’t on the visible page, so a Flow check on body_html misses it, and a person reviewing the product page never sees it.

It matters more now because AI shopping assistants read that block alongside the page and repeat claims back to shoppers in their own words. A “hypoallergenic” buried in schema turns into “this is a hypoallergenic option” in an answer you don’t control.

Two checks. View source on one product per template and search for “description” inside the ld+json script. Then ask ChatGPT the question your risky claim answers (“is [product] safe for sensitive skin?”) and see which of your words come back. If it quotes a claim you’ve already removed, the old text is still sitting somewhere it can read.