A new working paper from researchers at the University of Hamburg and the Frankfurt School of Finance & Management finds that referral traffic from ChatGPT converts far worse than traditional marketing channels such as Google Search, email and affiliate links.

The paper, authored by Maximilian Kaiser and Christian Schulze, examined 12 months of first-party data from 973 e-commerce sites generating a combined $20 billion in annual revenue. Researchers analyzed more than 50,000 transactions from ChatGPT referrals and compared them with 164 million transactions from traditional channels. While organic LLM traffic underperformed all channels except paid social, its conversion rate and revenue per session improved steadily throughout the study period, the study concluded.

“Results contradict widespread expectations of LLM superiority,” the authors wrote in the paper’s abstract. “Time-trend analyses suggest gradual convergence with traditional channels, but projections indicate LLM will not achieve parity with organic search within the next year. These findings challenge narratives of LLMs as immediate ‘Google killers,’ while suggesting potential for long-term channel evolution.”

Indeed, Similarweb estimates the conversion rate for visits referred by ChatGPT is 11.4%, compared to 5.3% for organic search, according to a September report.

“There’s so much noise about AI agents impacting commerce and so little data,” said Juozas Kaziukėnas, an independent e-commerce analyst who reviewed the paper. “This study is one of the first to quantify how much traffic you could potentially get from ChatGPT and how much of that actually converts.”

OpenAI’s new Instant Checkout feature, which lets shoppers buy products directly within ChatGPT, could help improve conversions as more users turn to AI agents for shopping. OpenAI will need to convince shoppers that buying through ChatGPT is as safe and intuitive as checking out on a retailer’s site. That trust gap remains a key hurdle, Schulze told Modern Retail in an interview.

Open AI did not respond to a request for comment by press time.

ChatGPT drives traffic — but not yet sales

Kaiser and Schulze found that ChatGPT accounts for more than 90% of all e-commerce traffic originating from large language models, dwarfing other AI agents like Gemini and Perplexity. Even so, the overall volume is minuscule: ChatGPT traffic is roughly 200 times smaller than Google’s organic search, representing less than 0.2% of total traffic across the researchers’ dataset.

Regression results show just how far ChatGPT lags behind more established channels. Affiliate links were 86% more likely to convert than ChatGPT referrals, while organic search outperformed ChatGPT by roughly 13%. In terms of overall revenue generation, ChatGPT’s revenue per session trailed both paid and organic search, though it still outperformed paid social media across all key financial metrics. The study also found that engagement from ChatGPT visitors — measured by bounce rate and session depth — was relatively high, suggesting users are interested but not yet ready to buy.

The researchers flagged a lack of consumer trust as one of the possible reasons why ChatGPT conversion trails more traditional channels. “This lack of consumer trust is a potential explanation for our findings [which show] people don’t use ChatGPT as the last thing before purchase, but instead, check out other sources and then buy,” Schulze said.

Still, conversion performance is improving, according to the paper. Over the 12-month observation period, the authors found that conversion rates from ChatGPT referrals increased steadily month over month, even as traffic volume expanded. While average order value declined slightly, total revenue per session rose, suggesting that users arriving from ChatGPT were becoming more likely to complete a purchase over time. This indicates that shoppers referred by ChatGPT are gradually learning to trust and act on the platform’s recommendations, the researchers wrote.

The working paper has some limitations. For one, it measures last-click referrals, meaning it doesn’t capture how often ChatGPT may influence shoppers earlier in their research. Moreover, the dataset covers a narrow time frame and a small share of total traffic, less than 0.2%, so the results should be viewed as early indicators rather than definitive trends, the authors wrote. The paper is currently under review for publication at an academic journal, but has not yet been peer-reviewed.

The promise of ChatGPT checkout

Shopper conversion via ChatGPT may improve with OpenAI’s Instant Checkout, a new feature allowing users to purchase products directly inside ChatGPT without leaving the chat interface. The system is already active for Etsy sellers and will soon expand to Shopify merchants and Walmart listings.

Whether that bet pays off remains to be seen. Other platforms like Instagram have previously tested checkout tools that let users buy products directly inside their apps, only to later walk back such features. The challenge for OpenAI will be convincing users to trust ChatGPT enough to complete purchases there, rather than clicking through to retailers’ own sites.

“If people don’t really trust ChatGPT yet, then having Instant Checkout is not a meaningful feature, because they will leave the platform anyway, in order to find other sources of information outside of ChatGPT,” Schulze said.

For OpenAI, Instant Checkout represents a long-awaited path to monetization. Despite ChatGPT’s hundreds of millions of users, only a small fraction pay for the Plus subscription tier. By taking a small fee on each transaction, OpenAI could finally begin generating meaningful revenue from e-commerce activity, a step that analysts say is necessary to justify the billions of dollars it’s spending on data centers.

“The AI bubble is real,” said Kaziukėnas. “These companies have to generate trillions of dollars of revenue from somewhere, and it’s probably not going to come from $20 subscriptions.”

Retailers test the waters

Retailers, nonetheless, are lining up to participate. Princess Polly, a Gen-Z–focused fashion brand, as well as a Shopify merchant, said it plans to go live on ChatGPT once the integration is available.

“We always want to be early adopters of anything that streamlines the [shopping] process for our customers,” said Stephanie Moore, Princess Polly’s head of brand marketing. “We know our customers are already using ChatGPT, so we want to meet them where they are.”

Moore said the brand is just as interested in the data insights the tool will provide as in direct sales. Unlike website analytics, ChatGPT interactions reveal longer, more conversational queries — for example, “What should I wear on a summer night date?” rather than “black dress.”

“That kind of data will help us understand what customers are really looking for,” Moore said.