ChatGPT traffic underperforms Google in e-commerce study

Organic traffic from ChatGPT underperforms all major digital channels except paid social media in e-commerce, challenging narratives positioning artificial intelligence as an immediate threat to Google's search dominance.

Research examining 12 months of first-party data from 973 e-commerce websites with $20 billion combined revenue reveals ChatGPT referrals generated lower conversion rates and revenue per session than Google's paid and organic search channels. The study, conducted by researchers at University of Hamburg and Frankfurt School of Finance & Management, analyzed over 50,000 transactions from ChatGPT referrals alongside 164 million transactions from traditional channels between August 2024 and July 2025.

ChatGPT traffic accounted for less than 0.2% of all sessions across the dataset. Within organic large language model traffic, ChatGPT dominated with over 90% of observed sessions, while Perplexity captured 4.1%, Gemini 2.6%, and Copilot 2.1%. The study focused exclusively on ChatGPT referrals due to negligible volume from competing platforms.

The analysis contradicts expectations that artificial intelligence platforms would immediately outperform traditional channels. ChatGPT referrals demonstrated conversion rates inferior to all channels except paid social media. Affiliate marketing produced the highest conversion advantage, with log-odds translating to 86% greater likelihood of conversion compared to ChatGPT. Organic search outperformed ChatGPT by approximately 13%.

Revenue per session metrics reinforced these findings. ChatGPT clearly outperformed paid social media but generated significantly lower revenue than remaining traditional channels. Direct traffic, email, and paid search all produced substantially higher revenue per session than ChatGPT referrals.

One metric favored the AI platform: bounce rate. ChatGPT achieved lower bounce rates than most traditional channels, suggesting relevance when clicks occurred. The platform did not outperform Google's organic and paid search channels on this measure. Session duration and page views showed ChatGPT underperforming all channels except paid social media.

The researchers employed regression models accounting for data sparsity rather than relying on model-free evidence. Weekly aggregation addressed sparse observations inherent in a channel representing 0.2% of total traffic. Extensive robustness checks varied aggregation periods, observation thresholds, and website samples without altering core conclusions.

Time trend analysis revealed improving trajectories despite current gaps. Conversion rates increased over the 12-month period, though declining average order values offset some gains. Projections for the subsequent 12 months suggested continued narrowing of conversion rate gaps accompanied by widening average order value gaps. Three of four trend models indicated ChatGPT would not achieve parity with organic search within one year.

Website-level heterogeneity emerged as a significant factor. Analysis restricted to the top 25% of websites by total revenue showed systematically larger performance gaps, with ChatGPT performing worse on larger websites compared to smaller sites. The 40% overlap between top websites by ChatGPT traffic and top websites by total transactions suggested current large language models may disproportionally favor smaller websites.

ChatGPT traffic volume remains minimal relative to established channels. Separate analysis tracking 44,421 websites found ChatGPT capturing 0.19% traffic share with 5.3% monthly growth versus Google's 41.9% dominance. However, commercial intent data reveals 21.6% of ChatGPT interactions demonstrate purchase signals, with 7.1% displaying high intent.

The study's authors identified three factors potentially explaining performance gaps. Platform architecture and commercial optimization remain early-stage, with only 2.1% of conversations involving purchasable products. Features critical for retail—comprehensive product catalogs, current pricing, and reliable deep-linking—may not be fully developed. Website infrastructure presents coordination problems, as weak performance depresses investment incentives while limited investment sustains weak performance.

Consumer adoption patterns reflect nascent usage. While ChatGPT reached 700 million weekly active users in August 2025, shopping via large language models remains limited. Users may employ ChatGPT for research while completing purchases through traditional channels, assigning conversions to other channels under last-click attribution.

OpenAI launched instant checkout capabilities on September 29, 2025, introducing the Agentic Commerce Protocol developed with Stripe. The implementation enables transactions from Etsy and Shopify merchants directly through ChatGPT conversations. Payment processing flows through Stripe, with merchants maintaining fulfillment and customer service responsibilities.

Industry skepticism persists regarding autonomous shopping systems. Analysis identified eight structural challenges facing AI shopping agents, including platform gatekeeping by Amazon and Shopify, which collectively control over 50% of U.S. e-commerce. Both platforms block AI agents to maintain discovery ownership and protect retail media businesses.

The researchers noted limitations inherent in last-click attribution data, which may systematically underestimate ChatGPT's contribution to purchase funnels. Future research employing multi-touch attribution methods could reveal whether ChatGPT serves valuable roles earlier in customer journeys that do not appear in conversion metrics.

Category distribution within the dataset spanned 24 top-level classifications. E-commerce and shopping represented the largest category with 1.2 billion sessions, followed by lifestyle including fashion and cosmetics with 1.1 billion sessions. Geographic coverage extended across all continents, with the Americas generating 2.7 billion sessions and Europe contributing 2.1 billion sessions.

The findings align with broader market dynamics. ChatGPT referral traffic declined 52% starting July 21, 2025, as citation patterns shifted dramatically toward Reddit and Wikipedia. Reddit citations increased 87% while Wikipedia reached nearly 13% citation share, reflecting systematic preference changes favoring community-driven and encyclopedia-style content over commercial websites.

Despite current underperformance, the research suggests potential for long-term channel development. The positive trajectory in conversion rates, combined with gradually narrowing performance gaps, indicates ChatGPT referrals may continue improving. Whether this development translates to competitive parity with established channels remains uncertain given offsetting trends in average order values and the substantial head start enjoyed by traditional digital marketing infrastructure.

Timeline

• August 2024: ChatGPT begins providing outgoing links to e-commerce websites, launching organic LLM traffic channel
• April 2025: Microsoft and OpenAI position for retail AI battle with shopping integrations
• August 2025: Amazon blocks AI bots from major tech companies including Meta and Google
• September 2025: OpenAI launches instant checkout with Stripe partnership
• October 2025: Study covering August 2024-July 2025 published analyzing ChatGPT e-commerce performance

Summary

Who: Researchers Maximilian Kaiser (University of Hamburg) and Christian Schulze (Frankfurt School of Finance & Management) conducted the analysis examining 973 e-commerce websites with combined annual revenue of $20 billion.

What: Organic traffic from ChatGPT underperforms all traditional digital channels except paid social media across key financial metrics including conversion rate and revenue per session, despite achieving favorable bounce rates indicating relevance.

When: The study analyzed 12 months of first-party data from August 2024 through July 2025, covering the first year after ChatGPT introduced outgoing e-commerce links in August 2024.

Where: The dataset includes global e-commerce activity with 2.7 billion sessions from the Americas, 2.1 billion from Europe, 856 million from Asia, 277 million from Oceania, and 56 million from Africa, spanning 24 industry categories.

Why: The research challenges narratives positioning large language models as immediate threats to Google's search dominance while revealing platform architecture limitations, website optimization gaps, and nascent consumer adoption patterns that explain current underperformance despite potential for gradual long-term improvement.