OpenAI has announced a new content partnership with The Washington Post that will enable ChatGPT to summarize and link to the newspaper’s original reporting in its responses. This collaboration is part of OpenAI’s broader strategy, having already formed partnerships with over 20 news publishers, including The Guardian and Axios. The Washington Post aims to leverage ChatGPT’s extensive user base, which has surpassed 500 million users, to enhance its reach. In return, OpenAI anticipates that this partnership will improve the quality of information provided to users, drawing from the Post’s timely and well-sourced journalism. Financial details of the agreement remain undisclosed, and while some newsrooms have embraced OpenAI’s technology, others, such as The New York Times, have raised concerns about copyright issues.
OpenAI’s Deep Research technology has shown to outperform traditional methods in web-based research tasks, yet it still fails to provide accurate answers nearly half the time. Recent testing revealed that while Deep Research can analyze and retrieve information more effectively than human researchers, it achieved only a 51.5% accuracy rate in a benchmark known as BrowseComp, which assessed its ability to answer complex questions requiring extensive web searches. The study, led by OpenAI researchers Jason Wei and his team, highlighted that human respondents struggled significantly, answering only about 30% of challenging questions after two hours of effort. The test involved 1,266 questions designed to be difficult for both humans and earlier versions of Deep Research, demonstrating the ongoing challenges in the development of artificial intelligence technologies aimed at improving research efficiency.
Why do we care?
OpenAI’s partnership with The Washington Post underscores a key moment in the evolution of AI’s role as a content mediator. It also brings to light a fundamental tension between information quality and AI-generated synthesis. News organizations are splitting into two camps: licensing (like The Guardian and Axios), or litigating (like The New York Times). OpenAI’s deals set precedent. Expect similar arrangements to cascade down into B2B content providers—think documentation platforms, analyst firms, vendor support content. I’ll comment I’m not sure this is a good thing, yet also will be preparing for it.
OpenAI’s Deep Research outperforming humans on complex web tasks sounds impressive—until you look at the benchmark: only 51.5% accuracy. This underscores a critical point: AI isn’t reliable enough to be left unchecked, even in “research” roles.
Don’t just use AI—curate with it, audit it, and own the last mile of interpretation. That’s where the value will lie.