Case Study: Our Analysis of 10,000 Prompts - When ChatGPT Shopping Is Triggered

When ChatGPT launched its shopping feature, many e-commerce brands wondered: does this actually drive real discovery, or is it just a novelty layer? To answer that question, Lumentir conducted a detailed analysis of 10,000 real ChatGPT prompts, tracking exactly when shopping results appeared, which prompt types triggered the feature most reliably, and what this means for brands competing in AI-driven discovery. The findings are more nuanced - and actionable - than most people expect.

The Big Picture: Shopping Activation Rates

The short version: ChatGPT's shopping feature is triggered in roughly 9% of all interactions. That means if you send 100 messages to ChatGPT, only about 9 of them will result in product cards, prices, and buy links appearing in the response. The rest? Just regular text answers.

That 9% figure might sound low, but it's worth context. Most ChatGPT use cases - writing help, coding questions, research, general conversation - have nothing to do with shopping. So 9% represents a meaningful slice of truly commerce-focused interactions.

The research revealed a critical segmentation: 79% of prompts never activate the shopping feature at all, while about 6% do so consistently every time they're tested. That's not randomness; that's a clear split. Most queries are outside the shopping zone, a reliable segment sits squarely inside it, and everything else hovers in a grey area where results vary day to day.

Open-Ended vs. Brand-Specific Prompts: A 4x Difference

The trigger rate breaks down dramatically based on prompt structure. This is where how prompts influence AI answers becomes a real strategic consideration for ecommerce brands.

Open-ended prompts - like "best wireless headphones for commuting" or "durable running shoes under $150" - trigger the shopping feature at a rate of 12.1%. These are discovery-style queries where the user hasn't locked into a specific brand or product yet. They're signal-searching, not brand-searching.

Brand-specific prompts - like "Sony WH-1000XM5" or "Nike Air Max" - trigger shopping at only 3.1%. That's nearly four times lower. Why? When a user names a specific brand or product, ChatGPT interprets the query as informational rather than transactional. They might be asking for specs, reviews, comparisons, or availability - not necessarily ready to buy.

Open-ended queries signal exploration. Brand-specific queries signal research. ChatGPT's shopping layer is built for the first, not the second.

"ChatGPT's shopping feature is most effective for open-ended prompts, indicating its role as a discovery tool rather than a direct product search engine." - Research analysis based on Profound's ChatGPT Shopping prediction study, We reverse-engineered ChatGPT's Shopping trigger

Consistency: Why Stability Matters for Brands

One finding stood out in our data: once a prompt triggers a shopping response, it does so reliably. If a prompt activates shopping results today, there's an 83% chance it will do so again tomorrow - and the next day after that. This isn't random variation or A/B testing noise. It's a stable behavior pattern built into how ChatGPT evaluates queries.

This stability matters enormously for brand strategy. It means you can test prompts and get reliable signals about which categories and query types will produce shopping results. You're not chasing a moving target. The system has consistent internal logic about which queries belong in "shopping territory."

The flip side: prompts that don't trigger shopping also tend to stay non-triggering. The 79% that never activate shopping are sticky in their non-activation. This creates a stable segmentation that brands can use to make decisions about where to invest in visibility and how to structure product data and descriptions.

Which Product Categories Drive Shopping Results

AI visibility for ecommerce isn't uniform across all product categories. Research analyzing over 1 million prompts found that shopping activates most reliably for shippable consumer goods: apparel, electronics, home goods, personal care, pet supplies, and sports equipment.

In contrast, software, services, travel bookings, and financial products almost never trigger shopping results, regardless of how purchase-oriented the language is.

Within product categories where shopping does activate, commercial intent roughly quadruples the trigger rate - from around 17% for informational queries to 76% for clearly transactional language. So a query like "wireless headphone reviews" might get shopping 15% of the time, but "wireless headphones I should buy for under $200" gets it 70%+ of the time.

This suggests that both category and intent language matter, but category is the dominant factor. A product-category prompt triggers shopping even without purchase language. Non-product categories don't trigger shopping no matter how transactional the language.

Real-Time Product Data: What Shopping Actually Shows

When ChatGPT shopping activates, users see real-time product data including prices, availability, images, and direct purchase links - all without leaving the conversation. Products come to you; you don't go to them.

According to OpenAI's Help Center guide on shopping research, this works especially well in detail-heavy categories like electronics, beauty, home and garden, kitchen and appliances, and sports and outdoor gear - where users often have multiple constraints (price, features, aesthetics) that a conversation can clarify.

For simple queries - checking a price, confirming a feature, or verifying availability - a regular ChatGPT text response is faster and sufficient. But for nuanced shopping decisions, the shopping feature bridges the gap between natural language understanding and product discovery.

"Shopping research performs especially well in detail-heavy categories like electronics, beauty, home and garden, and sports and outdoor, where users describe complex needs and preferences that benefit from visual product grids and comparison data." - Insights from OpenAI's shopping research documentation

How ChatGPT Selects Which Products to Show

When shopping results do appear, which products get featured? This is where brand visibility gets tactical. ChatGPT searches Google Shopping to power its recommendations, and product selection is based on relevance, popularity, and data quality.

Specifically, ChatGPT requires consistent product entity data - name, brand, price, availability - across three or more data sources. Products with fragmented or conflicting catalog information get filtered out before ChatGPT even generates its response. This creates a data quality moat: well-maintained product feeds across multiple platforms have a huge advantage in appearing in shopping results.

This also explains why user experiences shared on platforms like Reddit are often considered more trustworthy than paid reviews on product pages. ChatGPT is evaluating not just your official data, but how your products are perceived and discussed across the web.

Implications for E-Commerce Strategy

For brands and e-commerce teams, these findings reshape how to think about AI visibility:

• Category-level visibility matters most. If your product shows up when someone asks "best wireless headphones for commuting," that's more valuable than appearing when someone already knows your exact model number. Open-ended queries are the target.
• Data quality is a competitive moat. Consistent, accurate product information across multiple platforms - Google Shopping, your website, retail partners - directly impacts whether ChatGPT can confidently recommend you.
• The 6% of high-trigger prompts are your reliable audience. These are the query types worth understanding, testing, and optimizing for. They represent repeatable, predictable discovery.
• Timing advantage is real. ChatGPT ecommerce traffic converts 31% higher than non-branded organic search, suggesting that users who reach your site via AI recommendations are further along in their buying journey.

Understanding how ChatGPT decides who gets mentioned and cited is becoming as important as SEO for discovery-focused brands. The difference is that AI visibility is less about keyword stuffing and more about being the right product for the right conversation.

Reliability and Transparency: The Open Questions

These findings also surface some important gaps. If brand-specific prompts only trigger shopping 3.1% of the time, that's a pretty unreliable experience for users who expect product options when they search by brand name. The user might get a wall of text about the brand's history instead of product cards - which defeats the purpose of the shopping feature.

There's also a transparency question. When ChatGPT does surface products, how does it decide which ones to rank first? The research provides insights - relevance, popularity, data quality - but the exact weighting remains opaque. Are products that bid for placement getting preferential ranking? How much weight do independent reviews carry versus official product data?

These are fair questions for regulators and users alike. For now, the clearest takeaway is that ChatGPT shopping works well as a discovery layer for exploratory queries, and less well as a direct product search tool. Knowing that distinction helps set realistic expectations.