Amazon yesterday began showing AI-generated images of products that do not exist inside the search bar of its Shopping app - a feature framed as closing the gap between imagination and discovery, but one that raises hard questions about shopper trust, advertiser visibility, and the direction of product search on the world's largest e-commerce platform.

The announcement, published June 3, 2026, by Mihir Bhanot, Director of Search at Amazon, covered eight visual search updates. The generative image feature is the most structurally novel. It also drew immediate criticism from outside the company.

Ben Schoon, Senior Editor at 9to5Google, wrote on June 3 that the concept was "wildly wasteful in terms of the use of AI resources." Schoon argued that generating fake images during a product search "just seems remarkably dumb," noting that people go to Amazon specifically to buy physical products. "Having an AI take your search and create things that do not exist makes no sense whatsoever," he wrote. His alternative framing was pointed: using AI to parse through the millions of actual product images Amazon already holds - and surface those during search - would be a different matter entirely.

That distinction cuts to the core of what Amazon is doing, and why the feature is generating skepticism beyond the usual new-feature commentary.

What the generative search feature actually does

The mechanic is this: as a customer types into the Amazon Shopping app search bar, AI-generated images appear in the suggestions below the bar, shifting and refining with each word added. The images are synthetic. They do not correspond to any specific listing in Amazon's catalog.

According to Amazon, the feature addresses a real linguistic gap. A shopper who wants a shirt with a draped collar may not know the term "cowl neck." Someone picturing a couch with woven side panels may not have the word "rattan." Descriptive language - color, texture, pattern - triggers the image generation system, which then produces a visual interpretation. The shopper taps the generated image that best matches their mental picture and Amazon runs a search for visually similar real products.

The three-step sequence is: describe, see a synthetic image, find real products that look like it. Amazon describes this as bridging "imagination" and "product discovery." Schoon's critique is that the bridge leads somewhere uncertain. The generated image, precise in silhouette and texture, may set an expectation that no real listing fully meets. A shopper confronted with a page of only approximately matching results cannot find the product Amazon just showed them - because that product was never real.

Amazon launched the feature for apparel and home categories in the United States on iOS and Android. Additional categories will follow over time, according to the announcement. The feature is not available globally at launch.

Seven other updates, built on existing infrastructure

The generative search bar image is the newest surface. The other seven updates announced June 3 extend tools that have been building since at least 2024.

Shop by Style inserts AI-generated shoppable collage panels into apparel and accessories search results. When a customer searches "women's silk shirt," panels labeled with themes such as "Urban luxe" or "Soft elegance" appear in results. Tapping a collage opens a curated browsable page. Items can be shopped directly, similar products explored, or styles swiped between to compare aesthetics. The editorial logic here is driven by Amazon's AI, not by brand-supplied creative - a distinction that matters for how products get surfaced and in what context.

Lens Live, which Amazon launched on September 2, 2025, has been upgraded with Alexa for Shopping integration. The camera experience now incorporates the unified assistant that merged Rufus and Alexa+ on May 13, 2026. According to Amazon, customers can see quick product summaries and tap "Ask about this" to get answers to specific questions - such as how to remove a coffee stain from a rug - without leaving the camera view. For items the camera cannot identify, intelligent captions describe what is visible in the scene, giving shoppers a starting point even when object recognition fails.

Visual Suggestions is an existing feature, here refined. As customers type a broad visual description such as "flannel shirt," descriptive image filters appear in real time, allowing shoppers to narrow results before executing the search. This feature first appeared in Amazon's October 2, 2024 visual search update and has been incrementally extended since.

Text-to-image search enables customers to add written qualifiers to any image uploaded to Amazon Lens. A customer uploading a photo of a beige three-seat sofa can add text such as "like this, but in white" and receive results that match both the visual shape and the written specification. According to Amazon, the results surface items similar to the original image with the text-based parameters applied.

The Amazon Lens lock screen widget is new for iOS only. Customers can now add the Lens camera search function directly to their iPhone lock screen, alongside Amazon's existing search and orders widgets. According to Amazon, adding the widget requires holding down the lock screen, selecting Customize, tapping Add Widgets below the clock, and locating the Amazon Shopping app. The stated goal is eliminating navigation steps between noticing a product in the real world and initiating a search.

"More like this" is a new shortcut appearing on individual product images within search results pages. A shopper who likes the general form of a listed dress but wants it in a different length, with sleeves, or in another color can tap the shortcut to surface visually similar products immediately. No new search query is required.

Circle to search is now available within Amazon Lens for uploaded images. Customers can draw a circle around a specific item in a multi-object image, resize the bounding box, or move it to focus on a different element before running the search. According to Amazon, this allows isolation of a single item within a complex scene.

The trust question is structural, not cosmetic

Schoon's critique is not just a product review. It surfaces a structural tension that Amazon has not publicly addressed.

Amazon's catalog contains hundreds of millions of products. Its search infrastructure, built over more than two decades, is designed to connect expressed intent to available inventory. The AI image generation feature inserts synthetic objects into that pipeline - objects that look like products but are not products. The generated images are not labeled in a way that makes their synthetic nature unmistakable to a casual user mid-search.

The gap between what the AI generates and what actually exists creates a specific kind of disappointment risk. If a shopper taps a generated image of a terracotta-toned linen sofa with clean angular legs and is shown a results page of loosely matching alternatives, the experience degrades relative to the promise. That degradation is not recoverable by adding more AI. The problem is that the expectation was set by a fabricated object.

Schoon acknowledged that Amazon's framing of bridging the gap between imagination and product availability is "a good idea." The problem, in his reading, is the execution: generating whatever the user says rather than surfacing actual products that exist. The distinction matters because, as he wrote, that difference is "sure to confuse customers who suddenly cannot find a product that matches what Amazon just showed them in search."

For the marketing community, the trust concern is not abstract. Conversion is the fundamental unit of Amazon advertising economics. Features that introduce ambiguity at the moment of intent - the search bar, the moment before a product page loads - carry conversion risk.

Context: how far visual search has come at Amazon

The June 3 announcement does not arrive without precedent. Amazon reported a 70% year-over-year increase in visual searches worldwide in October 2024, when it introduced five new visual search features including visual suggestions and text-enhanced image search. Photo searches in Amazon Lens more than doubled between 2023 and late 2025. The September 2, 2025 Lens Live launch increased overall Amazon Lens engagement by 13%, according to Amazon's November 2025 figures.

Rufus, Amazon's AI shopping assistant, reached more than 300 million users by early 2026 and generated nearly $12 billion in incremental annualized sales during 2025. Customers who engage with Rufus during shopping sessions convert at rates more than 60% higher than those who do not, according to Amazon's Q4 2025 earnings materials. Monthly average Rufus users increased 149% and interactions climbed 210% over the prior year.

Those numbers explain the investment logic. Visual and conversational search are growing fast on Amazon. Amazon's advertising revenue crossed $70 billion on a trailing twelve-month basis after Q1 2026, growing 24% year-over-year. Building more surfaces for high-intent shoppers to interact with inventory - even synthetic visual prompts - is consistent with that trajectory.

What has changed with the June 3 feature is the nature of what appears in the search bar. Earlier visual search features - visual suggestions, text-enhanced Lens, More Like This - surfaced or refined real products. The AI image generation feature produces something that has never existed. That is a qualitative departure, regardless of whether the downstream search it triggers leads to real products.

What it means for advertising and product visibility

For advertising professionals, the practical implications cluster around three areas.

The first is search bar placement. The AI image generation feature operates in the suggestion layer below the search bar - upstream of keyword-matched sponsored placements. If shoppers increasingly navigate by tapping generated images rather than completing keyword queries, the relationship between keyword bids and actual shopping behavior becomes less direct. Whether Amazon will introduce ad formats native to that suggestion layer - as it did when Sponsored Products and Sponsored Brands prompts became billable on March 25, 2026 after exiting beta - is not indicated in the June 3 materials.

The second is Shop by Style collage placement. Shoppable collage panels are algorithmically assembled by Amazon under editorial style labels. Which products appear in those panels, and whether Sponsored Brands or Sponsored Display placements factor into their construction, is not disclosed. For apparel and accessories brands, inclusion in a collage could represent meaningful visibility at the moment a shopper is browsing for style inspiration. Exclusion from that surface carries a corresponding cost.

The third is Lens Live and physical retail. The updated Lens Live feature, with Alexa for Shopping integrated, gives Amazon a conversion tool at the moment a shopper encounters a competitor's product in a physical store. Pointing a phone at a jacket in a boutique, receiving a product summary from Alexa for Shopping, and adding a lower-priced alternative to an Amazon cart is the designed behavior. This is a real-world traffic capture mechanism, and it operates in an environment where Amazon has no inventory risk for the comparison products it surfaces.

Research from Workflow Labs published April 17, 2026 showed that Rufus already compresses the effective product discovery space from roughly 50 results to approximately five. Rufus handled 38% of all Amazon sessions during Black Friday 2025. The new visual search features layer additional AI-mediated filtering onto that already compressed discovery space. Brands whose product data is not structured to surface correctly across visual, conversational, and generative search environments face a compounding disadvantage.

The competitive context adds pressure. Google reported 65% year-over-year growth in visual searching as of July 2025and has been expanding its own visual commerce capabilities through AI Mode and Search Live. Amazon's June 3 update is in part a response to that competitive environment - an attempt to keep product discovery inside the Amazon app rather than letting it migrate to Google Lens or other visual search tools.

What remains unresolved is whether generating synthetic images of non-existent products is a durable path to achieving that goal, or a design decision that will need to be revisited once conversion data from the feature becomes available.

Timeline

Summary

Who: Amazon, through Mihir Bhanot, Director of Search, with the June 3, 2026 announcement also drawing public commentary from Ben Schoon, Senior Editor at 9to5Google.

What: Eight visual search updates to the Amazon Shopping app, the most significant being a new feature that generates synthetic AI images of non-existent products in real time inside the search bar as customers type descriptive language. The generated images are intended as visual prompts to guide searches for real products, not as representations of actual listings.

When: Announced June 3, 2026. The AI image generation feature and most updates are available to U.S. customers on iOS and Android from the date of announcement, with apparel and home categories covered first. Additional categories are planned over time.

Where: United States, via the Amazon Shopping app on iOS and Android. The Lens lock screen widget applies to iOS only.

Why: Amazon frames the feature as addressing the gap between a shopper's mental image of a product and their ability to name it through keywords. The broader context is a multi-year build-out of AI-powered shopping tools in which visual search volume grew 70% year-over-year by late 2024, Rufus handled 38% of all sessions during Black Friday 2025, and Amazon's advertising revenue crossed $70 billion on a trailing twelve-month basis in Q1 2026. Generating synthetic images in the search bar extends that AI-mediated discovery logic into a new surface - though critics, including Ben Schoon of 9to5Google, argue the approach introduces confusion rather than clarity at a moment when shopper trust is the most consequential variable.