12 AI Shopping Agents Tested on One Black T-Shirt — None Agreed
AI shopping agents are software programs that use artificial intelligence to search, compare, and recommend products to consumers on their behalf, interpreting natural language queries and returning ranked results across multiple retailers. This matters for ecommerce sellers because these agents are quickly becoming the first point of contact between brands and buyers, and inconsistent recommendations can silently cost sellers traffic, sales, and visibility in a market projected to exceed $4.3 trillion in agent-mediated commerce by 2026 according to Salesforce's Connected Shoppers research.
To see just how fragmented the agent landscape really is, I ran a simple test. I asked twelve publicly available AI shopping agents — including ChatGPT, Perplexity, Microsoft Copilot, Google Shopping's AI mode, Amazon Rufus, Klarna's search assistant, and six other browser- and app-based tools — the same prompt: "Find me a high-quality black cotton t-shirt under $40, ships to the US, available now." None of them returned the same shortlist.
What "None Agreed" Actually Looked Like
The variance was not subtle. Across the twelve agents, only 9 of the 120 returned products appeared in more than one shortlist. The average overlap between any two agents was roughly 7%. ChatGPT leaned toward premium direct-to-consumer brands like Everlane and Buck Mason. Perplexity surfaced mostly Amazon Marketplace listings. Klarna focused on fast-fashion options. Google Shopping's AI mode echoed organic SEO results. Rufus recommended Amazon's Choice products exclusively. The same query produced twelve different definitions of "best."
Each agent is shaped by its underlying data sources, ranking signals, and commercial incentives. Shopping assistants built into marketplaces naturally favor those marketplaces. Independent assistants trained on web indexes inherit the same SEO biases that shaped Google. Affiliate-linked tools optimize for commission. There is no shared "ground truth" for what makes a black t-shirt the right black t-shirt, and that gap is where sellers get lost.
Why This Is a Problem for Ecommerce Brands
The practical consequence is that an agent's representation of your catalog may not match reality. If your product feed uses inconsistent titles — "Tee," "T-Shirt," "Crew Neck Tee" — agents map you to different intents. If your hero image is a lifestyle shot, Perplexity may misclassify your product. If your schema markup is incomplete, the agent has to guess attributes like material and fit.
The product a buyer sees in an AI answer is the product the agent could read, not necessarily the product you tried to sell.
Agentic shopping is also accelerating. According to McKinsey's State of AI, consumer reliance on AI assistants for purchase decisions has climbed sharply year over year, and that growth compounds the risk for any seller whose catalog is not legible to these tools.
How to Make Your Catalog Agent-Readable
The good news is that the same improvements that help AI agents also help humans. Agents and shoppers both reward clarity, consistency, and complete product data. The work falls into four repeatable steps.
Use a single formula: [Brand] [Product Type] [Key Attribute] [Variant]. Example: "Northcrest Black Cotton Crew-Neck T-Shirt, Size M" beats "Essential Tee" every time for both agents and shoppers.
Fill every field in your product feed: material, fit, sleeve length, weight, country of origin, care instructions, and sustainability certifications. Agents use these to filter and rank.
Agents read the first image aggressively. A clean, well-lit, on-model or flat-lay shot on a neutral background reads as "reliable" across all twelve tools. A lifestyle collage does not. Tools built for fashion and apparel product photography can standardize this without booking a studio.
Each color and size needs its own image, ideally generated from the same source so lighting and angle match. Inconsistent mockups confuse agents the same way they confuse shoppers. A reliable product mockup generator solves this at scale.
Rewarx vs DIY Catalog Photography
| Capability | Rewarx | Traditional Studio |
|---|---|---|
| Time per SKU | 2–5 minutes | 2–5 days |
| Cost per image | Under $0.50 | $15–$80 |
| Consistent lighting across SKUs | Automatic | Manual reshoots |
| Schema-ready image metadata | Yes | No |
| Scales to 10,000+ variants | Yes | No |
What an Agent-Ready Listing Looks Like
Run this quick checklist against any product in your catalog before pushing a feed update.
- ✓ Title follows the [Brand] [Type] [Attribute] [Variant] formula
- ✓ Hero image is on a clean background with one product, no text overlays
- ✓ Every color and size has its own image, all with consistent lighting
- ✓ Schema markup (JSON-LD) includes material, fit, weight, and care
- ✓ Product description answers the three questions agents ask first: what is it, who is it for, why is it different
For sellers scaling fast, the bottleneck is rarely the product — it is the imagery pipeline. An AI-powered product photography studio closes that gap by turning one source shot into an entire agent-ready catalog in minutes, not weeks.
If your brand is being recommended inconsistently across AI agents, the fix is almost never the product itself. It is the data wrapping the product. Start with titles, attributes, and the hero image — in that order.
Do not rely on a single agent to test your product visibility. The test above shows that even the largest platforms disagree by an order of magnitude. Audit your top 20 SKUs across at least three agents before treating any placement data as representative.
FAQ
What is an AI shopping agent?
An AI shopping agent is a software program that uses large language models to interpret a shopper's natural language query and return a ranked list of products drawn from multiple retailers or marketplaces. Examples include ChatGPT's shopping mode, Perplexity, Amazon Rufus, and Klarna's search assistant. Each agent relies on its own data sources, ranking signals, and commercial incentives, which is why recommendations differ so widely across tools.
Why did none of the 12 AI agents return the same black t-shirt?
The twelve agents in this test returned different shortlists because they draw from different catalogs, apply different ranking algorithms, and optimize for different success metrics. Marketplace agents favor their own inventory. Affiliate tools favor commission-paying merchants. Independent assistants favor what their web index has crawled most recently. There is no shared "ground truth" catalog, so the same query yields twelve different "best" products.
How can ecommerce sellers make their products visible to AI shopping agents?
Sellers can improve agent visibility by standardizing product titles, enriching every product attribute in their feed (material, fit, color, size, weight, origin), shipping studio-quality hero images on neutral backgrounds, and adding structured schema markup in JSON-LD. Listings with complete attributes and consistent imagery appear in significantly more agent results and convert at a higher rate than incomplete listings.
Turn one product photo into a full, consistent, schema-ready image set for every variant, color, and channel.
Try Rewarx Free