I rewrote 47 product descriptions for AI search and here's what actually changed

Last Tuesday I was staring at a 34% drop in referral traffic from ChatGPT and Perplexity for a mid-market DTC skincare client. Same products, same pages, same keywords ranking fine on Google. But AI engines had quietly stopped citing us.

I pulled up their top 20 product pages and read them aloud. Every single one read like a catalog entry written by a committee: "Our revolutionary formula combines advanced peptides with botanical extracts to deliver visible results." That's not content. That's filler that no human—and increasingly no AI—wants to quote.

So I rewrote 47 product descriptions over three days. Not for Google. For the machines that are now deciding whether your product exists.

Here's the framework I used, and the numbers I tracked.

The rewrite rules I actually followed

Before: "This moisturizer contains 2% hyaluronic acid and niacinamide."

After: "If your skin feels tight by 2pm even though you moisturize in the morning, this is the fix. The 2% HA sits on the surface layer for 12 hours—I measured transepidermal water loss before and after on 14 users and the average improvement was 31%."

The difference: the second version gives an AI engine something concrete to extract. Specific numbers, specific conditions, specific results. When Perplexity answers "best moisturizer for tight skin," it needs to pull a verifiable claim, not a vague promise.

I ran a GEO Audit Tool on the original pages and found something painful: 68% of our product descriptions didn't contain the exact phrasing people use in AI search queries. We were optimized for "hyaluronic acid moisturizer" but people ask "why does my face feel tight after washing it."

I rewrote the intro paragraph of every page to mirror the actual query language I found in the audit. Not stuffed—just naturally answered.

AI engines love:

I stopped writing flowing paragraphs and started writing in chunks that could be lifted out of context and still make sense. It feels weird. It works.

What happened after 14 days

I tracked citations in ChatGPT, Perplexity, and Google's AI Overview for the 47 rewritten pages vs. the 23 I left as control.

The conversion gap matters. People arriving from AI search are further along in the decision process—they've already had their question answered and now they're comparing options. Our rewritten pages were speaking their language.

The thing nobody tells you about GEO for e-commerce

It's not about gaming the system. It's about being the source an AI engine trusts enough to quote. That means your content needs to be more specific, more structured, and more honest than what you'd write for a human browsing a category page.

If you're trying to figure out where your pages stand, I'd start with an AI Gravity Checker on your top 10 product pages. It'll show you exactly which claims are being extracted and which are being ignored.

The shift from traditional search to generative engines is bigger than most teams realize. The GEO vs SEO breakdown I wrote last month covers the tactical differences, but the short version: if your product pages read like they were written for a shelf tag, AI engines will pass you by.

I'm not saying abandon SEO. I'm saying the bar for being cited just got a lot higher, and the brands that treat AI search as a first-class channel—not an afterthought—are going to pull away fast.