Source: Kaiser & Schulze (2026) — “Frontiers: ChatGPT Referrals to E-Commerce Websites,” Marketing Science
At some point, shopping quietly moved from search boxes to conversations.
Instead of typing “best espresso machine under $500” into Google and opening fourteen tabs like a person pretending to be organised, people can now ask ChatGPT and get comparisons, explanations, and purchase links.
This paper asks a very practical question: is ChatGPT referral traffic actually valuable for e-commerce?
The authors study 12 months of first-party data from 973 websites, covering more than 50,000 ChatGPT-referred transactions and 164 million transactions from traditional channels. So this is not a small “my cousin clicked a link once” dataset. It is serious.
The result is balanced.
Organic LLM traffic converts better than paid social, but worse than most traditional channels. Revenue per session is modest. Volume is still low. Engagement looks mixed: bounce rates are favourable, but session duration and page views are lower.
The interesting twist is product complexity.
ChatGPT referrals perform better for complex product categories. That makes sense. If I am buying socks, I do not need a conversation. If I am buying an espresso machine, insurance product, camera, or anything with too many tabs and opinions, I suddenly appreciate an assistant who can explain trade-offs without judging me.
So ChatGPT is not replacing search overnight.
But it is becoming a new kind of channel: less like advertising, more like assisted decision-making.
That matters because marketing teams love declaring channels either dead or revolutionary. This one is neither. At least not yet.
My takeaway: LLM referrals are real, but early. They currently serve niche informational needs, especially for complex purchases. The bigger shift is not traffic volume today. It is that consumers may increasingly start shopping by asking questions instead of searching keywords. That changes the game quietly first, then suddenly.