Summary

• A Marketing Science study analysed 973 e-commerce websites with more than $20 billion in combined revenue.
• Organic LLM traffic accounted for less than 0.2% of total traffic in the sample one year after organic shopping links launched.
• ChatGPT referrals performed better for complex products, but still lagged most established digital marketing channels overall.

Frankfurt School of Finance & Management research suggests ChatGPT is becoming a useful shopping referral channel for complex purchases, while remaining far from a broad replacement for search, email, affiliate, or paid marketing channels.

The study, published in Marketing Science by Maximilian Kaiser of the University of Hamburg and Christian Schulze of Frankfurt School, examined organic traffic from large language models to 973 e-commerce websites. The dataset covered 12 months of first party data, more than $20 billion in combined revenue, more than 50,000 ChatGPT-referred transactions, and 164 million transactions from traditional channels.

The findings cut against the simpler claim that conversational AI has already transformed online shopping. One year after the launch of organic shopping links, organic LLM traffic accounted for less than 0.2% of total traffic in the sample. ChatGPT referrals achieved conversion rates and revenue per session above paid social, but below the other established channels studied.

The more useful finding is where the channel performs better. Product complexity appears to change the economics. ChatGPT referrals were stronger in categories where consumers need guidance, comparison, and explanation before purchase. In simpler categories, where decisions are more routine and search or marketplaces already work efficiently, the channel’s immediate value was weaker.

Agentic commerce meets marketing reality

The study arrives as AI platforms are trying to push deeper into commerce through product discovery, recommendations, checkout, and agentic shopping. That future may arrive unevenly. Consumers do not need the same level of conversational support to buy printer paper, replacement cables, software subscriptions, insurance, industrial equipment, or specialist medical devices.

Retailers and B2B sellers should pay attention to that distinction. A low overall traffic share does not mean the channel can be ignored. It means expectations should be tied to product type, decision complexity, and customer information needs. Where buyers face technical specifications, trade-offs, compliance considerations, long consideration periods, or unclear product categories, a conversational interface may create value by structuring the decision.

For standardised products, the economics are less compelling. If users already know what they want, established channels remain efficient. Search, marketplaces, email, affiliate marketing, and direct traffic have deep infrastructure around ranking, attribution, inventory, pricing, reviews, and checkout. ChatGPT and other LLM platforms need to improve product availability, recommendation reliability, transaction flow, and commercial integration before they can compete consistently across those use cases.

The research also complicates marketing attribution. Organic LLM traffic is small, sparse, and still developing. Retailers may struggle to measure influence where a model helps a user make a decision before the final click happens elsewhere. That is familiar from earlier debates about social, content, and comparison channels, but LLMs add another layer because answers may compress discovery, education, and recommendation into one interaction.

The study’s trend data gives AI platforms some room. Conversion rates from ChatGPT referrals increased over time, suggesting consumers may be learning how to use the channel more effectively. However, the research also found declining average order values, producing only moderate gains in revenue per session.

The sensible conclusion is neither dismissal nor hype. ChatGPT is not yet a general purpose shopping engine. It is an emerging advisory channel with stronger evidence in complex categories. Sellers of information-heavy products should start testing how their data, content, availability, and product detail appear in AI mediated discovery. Sellers of routine goods may be better served by improving the channels that already convert.