Write for the Shopper and the Machine: The Two Layers Every Product Page Needs

Look at what people actually type into an AI shopping assistant and it’s humbling. Not “a structured Cabernet with grippy tannins and a lingering finish.” It’s “red wine for steak night,” “whisky for my dad’s birthday,” “something under $100 that won’t embarrass me,” “do you have it and can you ship to Ohio.” People ask by occasion, budget, pairing, and product — sometimes just the product. Nobody asks for a vocabulary lesson.

(Others analyzing real AI shopping conversations — like Ian Carroll — have put hard numbers behind exactly this pattern.)

So here’s the trap most product pages fall into: they’re written entirely for the customer who talks like a sommelier — all body, structure, and finish — when the actual customer is buying a bottle for Sunday dinner. The instinct, then, is to rip out the tasting notes and write casual, occasion-first copy.

Don’t. Because there’s a second reader now, and it needs the opposite.

Every product page has two audiences

The human layer — what the shopper asks. Occasion (“great gift,” “for a celebration”), budget (an actual price), pairing (“pairs with shellfish, roast chicken, Sunday roast”), and the boring-but-decisive practical stuff (do you have it, will it ship to my state, can I talk to a person). This is the language people use, and it should lead.

The machine layer — what the AI needs to match you. When someone asks for “a white wine for lobster,” the assistant has to figure out which of your bottles fits. It can only do that if your page spells out the facts: dry or sweet, the grape, the region, the acidity, the alcohol, the flavors — plus a bit of behind-the-scenes code (called structured data) that labels your product, price, and reviews for the machine. The shopper never reads any of it. The AI reading on their behalf reads all of it.

Here’s the whole idea in one example. Someone types “white wine for lobster.” For your bottle to be the answer, your page needs both layers working together:

Human layer (up top, visible): “A crisp, summery white that’s made for shellfish — lobster, oysters, a patio afternoon. Around $28. Ships to most states.”

Machine layer (present as real text + structured data): dry Sauvignon Blanc, Sonoma Coast, 12.5% alcohol, high acidity, notes of citrus and a saline edge.

Delete the machine layer and the AI has no idea your wine belongs with lobster — you’re invisible to the query. Lead with the machine layer and the human bounces off a wall of jargon. You need both, aimed at different readers, on the same page.

How to actually do it

1. Lead with the human layer

Open each product page with occasion, pairing, and price — the way a good shop clerk would talk. This is what the shopper reads and decides on.

2. Keep the factual layer as real text

Not locked in an image or a PDF, because a machine can barely read those. It doesn’t have to dominate the page; it just has to be there and selectable (the kind of text you can highlight with your cursor).

3. Add the “hidden labels” — structured data

Every web page has two parts: the part you see, and a layer of code underneath that builds it. Structured data is a small piece of that underneath-code that never shows on screen — its only job is to tell a machine what each thing on the page is: “this is a product,” “this one is priced at $28,” “this is a 4.5-star review from a real customer.” (Its technical name is schema.org markup.) Picture the tag sewn into a shirt collar — the shopper never notices it, but it spells out “100% cotton, size medium.” Structured data is that tag for your product page.

Why it matters here: without it, the AI has to guess whether “$28” is the price, a vintage year, or a case discount — and a machine that has to guess about you tends to skip you in favor of one it can read cleanly. With it, the assistant can state your product, price, and rating back to a shopper accurately. You won’t write it by hand:

• Decide what to mark up: your product, its price and availability, and your reviews or ratings.
• Have it generated: if your store runs on Shopify, Squarespace, or WordPress, a built-in setting or a plugin (like Yoast) adds it for you. On a custom site, your developer drops in a short snippet — a format called JSON-LD, which is just structured data written as a small block of text in the page’s code.
• Confirm it worked: paste the page’s web address into Google’s free Rich Results Test. It reads the page and tells you exactly what structured data it found.

The one-line instruction for whoever runs your site: “add schema.org structured data for the product, price, and reviews, and verify it in Google’s Rich Results Test.”

4. Answer the practical questions in writing

Shipping, availability, store hours, how to reach a human. Those are real queries too — and almost nobody puts them on the page.

This isn’t just wine

A gift shopper asks “anniversary present under $100”; a parent asks “a gift for a teenager who likes skincare.” Same structure every time: the human asks by occasion and budget, and the machine needs the specs to make the match. Any product page, any category.

The old advice was “know your customer.” The new version is: know your two customers — the person buying a bottle for Sunday dinner, and the machine answering them. Write for both, or you’ll win neither.

Zillah Bahar is the founder of COLAClear and writes about getting found by AI at GetDiscoverable.io.