How AI shopping agents decide what to show a shopper

When a shopper asks an AI agent for "a good dog harness," the agent doesn't open a browser and compare tabs. It searches a product catalog, reads the first few results, and recommends one. Sometimes it buys it.

The catalog Bark Twice measures is shop.dev, which AI agents use to search Shopify stores. We've run hundreds of searches through it while building Bark Twice. Here's what we know about whether your product makes those first few results.

What the catalog does

It hides what can't be bought. By default, search only returns items that can be purchased. If the variant a product leads with is sold out, the product can drop out of results entirely, even if other sizes are in stock.

Filters are strict. When an agent filters by size, color or who a product is for, the catalog treats those as requirements, not preferences. Products whose variants aren't labeled that way don't come back at all. A shirt with sizes named "Smol / Big / Bigger" is invisible to a search for size M.

It shows one variant per product. Each result is a single featured variant with its own price, availability and rating. Ratings are kept separately for the product and for each variant.

Signed-in shoppers see different results. When a shopper is signed in with Shop, results can be ranked by their saved preferences and shopping history. Signed-out results are the same for everyone, which is why we measure those: they're the baseline every shopper starts from.

The exact ranking formula isn't published. Nobody outside the catalog can tell you how much each signal weighs. What you can control is what your listings say.

What we've seen running the searches

Agents match what your listings say, not what your brand is known for. A product called "Soft Chew Ring" won't come up for "puppy teething toy" if nothing in its title, description or details says so, however well it sells.

Product names aren't searches. Shoppers who haven't heard of you search for the kind of thing they want ("wool sneakers"), not your product line ("Tree Runner"). Ranking first for your own product names is table stakes; the growth is in the searches that don't name you.

Thin listings give agents little to work with. A real description, specifics like material and dimensions, and a summary the catalog can quote give an agent something to match and something to say. A title and a photo don't.

Resellers compete with you for your own products. The same product often appears several times from different stores. If a reseller's listing is better described than yours, it can outrank you for your own product.

Colors sold as separate products split your attention. Some stores list every colorway as its own product. That can crowd a results page, but it also splits reviews across near-identical listings, each looking less proven than one product would.

What to do about it

1. Write titles in the words shoppers use. Keep your product's name, and add what it is: "Soft Chew Ring – Puppy Teething Toy."
2. Open each description with one plain sentence saying what it is, who it's for and what sets it apart.
3. Fill in the details a shopper would ask about: material, size or dimensions, fit, care.
4. Use standard option names. Call sizes and colors "Size" and "Color", with values a filter can match.
5. Keep the variant you lead with in stock.
6. Keep reviews coming, and fix the complaint that repeats on your lower-rated products.

None of this is new to anyone who's done SEO. The difference is that there's no page two. An agent shows a shopper a handful of options, and everything outside them might as well not exist.