Amazon Generated a Product You Cannot Buy to Help You Buy a Product

There is a version of this story where it makes sense. You open the Amazon app, you're trying to describe something you half-remember from a dream or a store window, and the app — reading your vague intentions — conjures a visual reference. Like a mood board that lives inside a search bar. That version is fine. That version is almost clever.

This is not quite that version.

What Amazon Actually Built

Amazon's updated in-app search will now generate AI images of products as you type — clothing and home goods for now — letting you tap the one that looks closest to what you want, then surface real listings that resemble it. The company's own blog framing, noted by The Verge, pitches it as a solution for when you know what something looks like but can't name it: you can describe a "shirt with a draped collar" without knowing the term "cowl neck." Fine. Useful, even, in that narrow case.

But then The Verge also notes the obvious: if you already know what you want, the feature adds nothing. And 9to5Google called it, with commendable directness, one of the dumbest uses of AI yet. TechCrunch's headline asked, essentially, why this exists at all.

Three outlets, same conclusion, different levels of diplomatic restraint.

The Engagement Problem Wearing a Utility Costume

Here's what all three pieces are circling without quite landing on: this feature is not solving for you. It is solving for dwell time.

When Amazon generates a product image — something that does not exist, cannot be shipped, has no listing — and places it in your search flow, it is not helping you find something. It is giving you something to react to. A visual hook. A reason to stay in the app one more tap, one more swipe, one more moment of engagement before you convert — or don't. The generated image is bait dressed as a tool.

Visual search, when it works, is genuinely useful: take a photo of a thing in the world, find that thing for sale. Reverse image search, style matching, color recognition — all of this has real utility because the input is real. You have a real object and you want its real equivalent.

What Amazon has done is invert the relationship. The image is now the output of a query, not the input to one. The visual comes first, reality comes second, and somewhere in between there's a product you might buy if the fake one was close enough to what you wanted. The algorithm is no longer helping you find things that exist. It is manufacturing desire for things that don't, then redirecting that desire toward things that sort of do.

That's not search. That's a concept store with a buy button.

Amazon has been, as 9to5Google notes, stuffing its app full of AI lately. This feature fits the pattern: AI as volume, as surface area, as the appearance of innovation spread thin across every touchpoint. At some point the accumulation of features stops feeling like a product getting better and starts feeling like a company reminding you it has engineers.

The real tell is the hedge built into the feature's own design. It only works for clothing and home goods — categories where aesthetic approximation is acceptable, where "close enough" is a valid search result. Nobody is doing this with a car battery or a specific HDMI cable. Because in those categories, close enough is useless. Which means Amazon already knows this feature's limits. It built it anyway.