A recent analysis of desktop browsing behaviors in the United States reveals that ChatGPT led users to external websites in 5.2% of browsing sessions, in stark contrast to Google searches, which accounted for 31.1%.
The study, undertaken by researchers at Bocconi University, employs Comscore clickstream data to delve into the ramifications of wider ChatGPT access on traditional search practices.
Insights from the Data
When users depart from ChatGPT, they frequently navigate to a markedly different array of websites compared to those arriving via Google referrals.
The former predominantly comprises reference sites, knowledge bases, SaaS platforms, and various academic or developer resources. Interestingly, ad-supported sites contribute 27.6 percentage points less to the referral traffic in this context.
Conversely, Google’s referral traffic is primarily directed towards well-known platforms such as YouTube, Reddit, and Wikipedia, whereas ChatGPT tends to direct users towards smaller, more specialized websites.
The authors note that ChatGPT typically channels traffic to non-profit, subscription-based, and freemium platforms, thus steering away from those reliant on advertising revenue.
The researchers express caution regarding the interpretation of this data, asserting:
Our claim is explicitly narrower than any welfare assertion: we seek to quantify a change in observable traffic distribution, rather than consumer surplus, publisher income, or long-term content output.
Each conversational interaction is analyzed as a discrete exchange, and the study assesses whether it results in an outbound click. Throughout the measurement period, ChatGPT’s monthly referral rate increased from approximately 2.5% to nearly 6.5%, albeit still trailing behind Google’s query rate.
Implications of Broader Access on Search
OpenAI has systematically broadened access to ChatGPT Search, allowing paid subscribers by October 2024, free users by December, and anonymous users by February 2025.
This expanded accessibility has led to a 9.4% reduction in weekly traditional search queries, with the decline escalating to 17% after a span of 20 weeks. For two groups already utilizing ChatGPT, the average decrease was 4.9%, rising to 8.2% over the same period.
The most significant drop was observed in searches for informational content, where academic research referrals plummeted by 32.8%, and reference searches decreased by 26.5%. Notably, transactional and recreational searches exhibited minimal change.
The Significance of These Findings
ChatGPT not only yields fewer referrals than Google but also directs users to varied corners of the web. The categories most affected include academic, reference, developer, and news sectors, with reference and knowledge sites standing out as prevalent browsing contexts within ChatGPT.
Previous coverage by SEJ highlights data from the Pew Research Center indicating that users engage with links far less frequently in the presence of AI-generated summaries, with most citations emanating from a select few major platforms.
The findings from the Bocconi study amplify this narrative, incorporating estimates regarding the influence of broader access on traditional search behavior.
Source link: Searchenginejournal.com.
