OpenAI Engineer: Users ‘Won’t Click’ Links, Court Docs Show
▼ Summary
– News publishers suing OpenAI and Microsoft cite internal messages claiming users rarely click links in AI-generated responses.
– A court filing highlights data showing significantly lower click-through rates for publisher sites in Bing Chat compared to traditional search.
– Publishers argue these reduced clicks demonstrate that AI chatbots serve as direct substitutes for original journalism.
– OpenAI and Microsoft counter the lawsuit by asserting their practices constitute fair use and do not replace publisher content.
– The dispute centers on whether AI citations provide sufficient value or merely divert traffic away from news organizations.
Internal Communications Highlight User Behavior Concerns
Recent court documents filed by news publishers in their lawsuit against OpenAI and Microsoft reveal internal communications that suggest a significant disconnect between AI-generated responses and user engagement with source material. A filing from the plaintiffs’ combined summary judgment brief, submitted on September 17 in federal court in New York, cites an OpenAI software engineer who stated in February 2023 that “no matter how prominently we show the links, users won’t click.” This assertion underscores concerns that generative AI models may be satisfying user queries without directing traffic back to original news sources.
The same legal brief highlights comments from Nick Turley, OpenAI’s head of ChatGPT, regarding the chatbot’s “browse” functionality. According to the document, Turley remarked that there was “no good reason to click” on a link to the original source once ChatGPT had already provided an answer. These statements are part of the publishers’ argument that AI systems are substituting for direct access to journalism rather than supplementing it. It is important to note that these filings represent one side of the dispute, and the court has not yet issued a ruling on the matter.
Data Suggests Drastic Drop in Click-Through Rates
Beyond internal quotes, the publishers’ brief presents data from Microsoft comparing click-through rates (CTR) for news websites across Bing Chat and traditional Bing Web Search. The filing defines CTR as the percentage of clicks relative to impressions, where impressions refer to the appearance of a URL in search results or a chatbot response. The data indicates a stark decline in user interaction when accessing publisher content through the AI tool compared to standard search.
For The New York Times’ websites, the CTR in Bing Chat was reported to be 87% to 93% lower than in Bing Web Search. Similarly, sites associated with the Daily News plaintiffs saw a decrease of 83% to 91%, while Ziff Davis properties experienced a drop ranging from 51% to 94%. The publishers argue that these reductions “clearly show substitution,” asserting that they do not need further data to prove harm because “common sense suffices.” However, the filing does not specify when Microsoft collected this data, the volume of queries involved, or how the figures were selected. Additionally, the ranges provided do not clarify what the low and high ends represent in terms of specific metrics.
Microsoft renamed Bing Chat to Copilot in November 2023, but the brief does not explicitly tie these historical figures to the current version of the product. The data focuses on the step after a link appears, measuring how often users actually engage with those links in different contexts.
Fair Use Arguments and Metric Distinctions
In response to the publishers’ claims, OpenAI and Microsoft have argued that their use of news content constitutes fair use. Both companies filed their own summary judgment motions on September 4. Reports indicate that OpenAI’s motion cited studies suggesting ChatGPT outputs do not create commercial substitutes for The New York Times’s work, with Microsoft making similar arguments regarding Copilot. A Microsoft spokesperson told TheWrap that the company’s court filings explain “why Copilot is not a substitute for publishers’ journalism.” OpenAI did not immediately respond to requests for comment.
The debate also hinges on how engagement is measured. Microsoft’s AI Performance report in Bing Webmaster Tools tracks citations, which differ from traditional search metrics that measure rankings, clicks, and traffic. When Microsoft introduced Citation Share in June, it clarified that this metric does not represent traffic share. In contrast, Google’s Search Console generative AI report displays impressions but does not list clicks. The data in the publishers’ filing separates chat interactions from search interactions, offering a more granular view of user behavior than previously available combined metrics.
Upcoming Legal Milestones
Judge Sidney H. Stein set a schedule for the next phases of the litigation in an order issued on September 24. Oppositions are due by October 23, followed by amicus briefs on October 30, and replies on November 20. Industry observers will likely scrutinize whether OpenAI and Microsoft address the engineer’s quote, Turley’s comment, or the CTR data in their upcoming filings. Future documents may provide additional context for these messages or detail exactly how Microsoft measured the cited click-through rates, potentially clarifying the extent to which AI tools impact traditional news traffic.
(Source: Search Engine Journal)
