ChatGPT traffic is bringing more people to ecommerce sites, but they aren’t purchasing.
New Researchers at the University of Hamburg, Frankfurt School of Finance & Management and the Frankfurt School of Management have published a working paperthat shows that referral traffic generated by ChatGPT converts much worse than other marketing channels like Google Search, Email and Affiliate Links.
This paper, written by Maximilian Kaiser & Christian Schulze, examined first-party data for 12 months from 973 ecommerce sites that generated a combined annual revenue of $20 billion. Researchers compared 164 million transactions with more than 50,000 transactions originating from ChatGPT. The study concluded that while organic LLM traffic performed worse than all other channels except paid social media, its conversion rate, and revenue per session, improved steadily over the course of the study. The authors of the abstract wrote: “Results contradict widely held expectations that LLM is superior.” “Time-trend analysis suggests gradual convergence with traditional search channels, but projections show that LLM won’t achieve parity with organic within the next 12 months. These findings contradict the narrative that LLMs are ‘Google killers’ and suggest potential for long-term evolution of channels.
Indeed Similarweb estimates conversion rates for visits referred to by ChatGPT at 11.4% compared to 5.3% organic search. September report
Juozas Kaziukenas is an independent ecommerce analyst who reviewed this paper. He said, “There’s a lot of noise about AI agents affecting commerce but so little data.” This study is the first to quantify the amount of traffic that could be generated by ChatGPT. It also quantifies how much of it actually converts. OpenAI must convince users that buying products through ChatGPT can be as safe and intuitive to use as checking out at a retailer’s website. Schulze told Modern Retail that the trust gap is still a major obstacle.
Open AI failed to respond by the deadline for a comment.
ChatGPT is driving traffic, but not sales
Kaiser and Schulze discovered that ChatGPT accounted for more than 90% all ecommerce traffic originating with large language models. This dwarfs other AI agents such as Gemini and Perplexity. The overall volume of traffic is still minuscule. ChatGPT traffic represents less than 0.2%, or 200 times less, than Google’s organic searches.
The regression results show how far ChatGPT falls behind more established channels. Affiliate links were 86% higher in conversion rate than ChatGPT referrals. Organic search performed better than ChatGPT, by about 13%. ChatGPT’s revenue generated per session was lower than both paid and organic searches, but it still outperformed social media paid in all key financial metrics. The study also found high levels of engagement among ChatGPT users, as measured by bounce rates and session length. This suggests that users are interested in the site but not ready to purchase. Researchers cited a lack in consumer trust as a possible reason why ChatGPT conversion is behind more traditional channels. Schulze said that this lack of trust by consumers could explain our findings [which show] People don’t use ChatGPT last before purchasing, but rather, they check out other sources, and then purchase.
According to the paper, conversion performance continues to improve. During the 12-month period of observation, the authors found that conversions from ChatGPT referrals grew steadily, month over month, despite the increase in traffic. While the average order value decreased slightly, the total revenue per session increased, suggesting that users who arrived from ChatGPT became more likely to make a purchase with time. Researchers wrote that this indicates that customers referred by ChatGPT gradually learn to trust the platform and act upon its recommendations.
This working paper has some limitations. It measures only the last-click referrals. This means it does not capture how much ChatGPT influences shoppers earlier in their research. The authors also noted that the dataset only covers a short time period and represents a small percentage of traffic, less 0.2%. Therefore, the results should not be considered definitive trends but rather early indicators. The paper is being reviewed for publication in an academic journal but has not been peer-reviewed.
ChatGPT checkout: The promise
Improve conversion rates with OpenAI’s ChatGPT. Instant Checkoutis a new feature that allows users to purchase products without leaving ChatGPT. The system is currently available for Etsy sellers, and will soon be expanded to Shopify merchants as well as Walmart listings. It remains to be determined if that bet pays off. Other platforms, like Instagram, have tested checkout tools which let users purchase products directly within their apps. Walk backthese features. OpenAI’s challenge will be to convince users to trust ChatGPT to the point that they complete purchases there rather than clicking to retailer sites.
If people don’t trust ChatGPT, then Instant Checkout will not be a useful feature because they will still leave the platform to find other information sources outside of ChatGPT,” Schulze explained.
Instant Checkout is a long-awaited monetization path for OpenAI. ChatGPT has hundreds of millions users, but only a fraction of them pay for the Plus subscription. OpenAI can generate revenue from ecommerce by charging a small fee per transaction. Analysts say this is the only way to justify its billions of dollars spent on data centers.
Kaziukenas said that “the AI bubble is real.” “These companies must generate trillions of dollar revenue from somewhere. It’s not going to be from $20 subscriptions.”
Retailers are testing the waters
Retailers are still lining up for participation. Princess Polly, an online fashion retailer and Gen-Z-focused brand, plans to launch ChatGPT as soon as the integration is ready.
“We want to be early adapters of anything that streamlines [shopping] the process for our customers,” Stephanie Moore, Princess Polly’s head of marketing said. We know that our customers already use ChatGPT. So we want to meet them there.
Moore stated that the brand is equally interested in the data insights provided by the tool as it is in direct sales. ChatGPT interactions are more conversational than website analytics. For example, “What do I wear for a summer date?” is a longer question, as opposed to “black dress”.
Moore said, “That kind of information will help us understand the real needs of our customers.”
