ChatGPT Shopping: What Actually Triggers Product Recommendations (1.18M Prompts Analyzed)

Purchase intent doesn't trigger ChatGPT Shopping. Product category does.

Most people assume ChatGPT shows Shopping cards when users signal buying intent. "I want to buy," "best deals on," "where can I get." Logical assumption. Also wrong.

Profound analyzed 1.18 million ChatGPT prompts and built a classifier that reproduces ChatGPT's Shopping behavior with 95-96% accuracy. Their core finding: the single biggest predictor of whether Shopping cards appear is whether the prompt names a shippable physical product. Not whether the user sounds like they're ready to buy.

The simplest rule: if the main noun in the prompt is something you'd find listed on Amazon, Shopping will probably trigger. If it's a service, software, or concept, it almost certainly won't.

The numbers by category

Profound tagged 7,500 prompts with ground-truth Shopping data and measured trigger rates by category:

• •Apparel/Fashion: 62% trigger rate
• •Physical Products (general): 56%
• •Consumables/Groceries: 30%
• •Health/Medical: 6%
• •Vehicles/Large Equipment: 2%
• •Travel/Hospitality: 0.6%
• •Services: 0.5%
• •Software/SaaS: 0.1%
• •Financial Products: 0%

Four categories are effectively locked out: software, services, travel, and financial products. No amount of prompt engineering activates Shopping for these. If you sell in those verticals, ChatGPT Shopping isn't your channel. At least not yet.

For physical consumer goods, though, the activation rates are remarkable. Nearly two-thirds of apparel-related prompts trigger Shopping cards. Over half of general physical product prompts do the same.

Commercial intent amplifies, but doesn't gate

Here's where it gets nuanced. Purchase language does matter, just not the way most people think.

Within product categories, commercial intent roughly quadruples the trigger rate (76% vs 17% for physical products). So saying "best running shoes under $100" triggers more reliably than "tell me about running shoes."

But the key word is "within." Commercial intent applied to a non-physical category does almost nothing. "I want to buy the best project management software" triggers Shopping only 8% of the time. "Running shoes" without any purchase language triggers 17%.

Category is the gate. Intent is the accelerator. Getting the order wrong means optimizing the wrong thing entirely.

79% of prompts never trigger Shopping. Ever.

In a companion study, Profound tracked roughly 2 million unique prompts over nine months (September 2025 through January 2026). The consistency data is striking:

• •79% of prompts never triggered Shopping across any run over the entire period
• •Only 6% triggered reliably (80%+ consistency)
• •Just 0.7% triggered every single time

The distribution is all-or-nothing. There's a massive cluster at zero and a small cluster at the top, with almost nothing in between.

But prompts that do trigger are remarkably sticky. If a prompt triggered Shopping on a given run, there's an 83% chance it triggers again the next day. Prompts lock into a state and tend to stay there, until a model update resets the pattern.

Short prompts trigger more often, but length is a proxy

Shorter prompts (5-10 words) show higher Shopping trigger rates than longer ones. But when Profound controlled for category, the effect vanished completely. Product prompts triggered at 49-64% regardless of length. Non-product prompts triggered at 0-1% regardless of length.

What kills activation isn't a long prompt. It's the introduction of a non-shippable noun. A 20-word prompt about headphones will still trigger Shopping. A 5-word prompt about consulting won't.

Shopping is a discovery surface, not a search engine

One more finding worth highlighting: open-ended prompts ("best business laptops with strong battery life") trigger Shopping at 12.1%. Brand-direct prompts ("Dell XPS 15 review") trigger at only 3.1%. That's a 4x gap.

ChatGPT Shopping rewards brands that show up during the consideration phase, when users describe what they need but haven't decided what to buy yet. This is fundamentally different from Google Shopping, where brand search is a strength.

Interestingly, Peec AI's analysis of ChatGPT Shopping carousels found that 83% of products shown are sourced from Google Shopping organic results. So while the trigger logic is category-driven (Profound's finding), the product data itself still flows heavily from Google's index. Brands that are invisible in Google Shopping are likely invisible in ChatGPT Shopping too.

For ecommerce brands, this means your GEO/AEO strategy for ChatGPT Shopping should focus on category-level visibility, not brand-level optimization. You want to be in the consideration set for "lightweight running shoes for flat feet," not just "Nike running shoes."

What this means for ecommerce GEO and AEO

1. If you sell physical consumer goods, ChatGPT Shopping is a real channel.

The 56-62% trigger rates for apparel and physical products aren't marginal. More than half of relevant prompts surface Shopping cards. If you're not tracking whether your products appear in those cards, you're ignoring a channel that's already influencing purchase decisions.

2. If you sell software, services, or financial products, focus elsewhere.

Shopping isn't your surface. Your GEO/AEO efforts should focus on citation optimization in ChatGPT's conversational answers, plus visibility across Perplexity, Gemini, and Claude. The playbook is different, but the opportunity is just as real.

3. Optimize for category queries, not brand queries.

The 4x gap between open-ended and brand-direct trigger rates means the biggest opportunity is in discovery prompts. Make sure your product data answers the questions people ask when they're still deciding: "best X for Y," "lightweight Z with good battery life," "affordable W that works with Q."

4. Product specificity matters.

The prompts in the reliable top 6% tend to describe a specific, shippable product need with enough detail that ChatGPT can match it to a real SKU. Vague product pages don't get surfaced. Detailed specs, clear pricing, and specific use-case positioning do.

The bigger picture: AI product discovery is fragmenting

ChatGPT Shopping is one surface among many. Google's AI Overviews surface product recommendations. Perplexity shows product cards with sources. Each platform has its own trigger logic, its own product data sources, and its own ranking signals.

The brands that will win this aren't the ones optimizing for one platform. They're the ones building systematic AI visibility across all of them, with monitoring that catches shifts before they cost revenue.

Profound's research gives us a clear model for how one surface works. The question for every ecommerce brand: do you know how the others work?