OpenAI's Deep Research is built for analysts. It compiles data, surfaces numbers, and produces structured reports in minutes. A researcher tested it against OpenAI's own sample report on smartphone market share, a topic he knows cold, and the results expose a fundamental problem with the tool.
The report's Japan figure, 69% iOS and 31% Android, is wrong. Statcounter, one of the two cited sources, has not shown that number in over a year. Statista, the other source, is an SEO-optimized data aggregator that traces back to Kantar Worldpanel, which puts the split at roughly the opposite: 63% Android, 36% iOS. Deep Research cited sources it either misread or did not verify, then presented the output with the confidence of finished analysis.
The sources themselves are the real story. Statcounter measures web traffic, not device adoption, and skews toward high-use premium devices. Statista is a middleman, not a primary source. An experienced analyst would reject both for this use case. The piece is worth reading in full because it forces a precise question: if Deep Research cannot reliably handle a narrow, verifiable data task in a domain with public benchmarks, what exactly is it good for, and who bears the cost when it is wrong.
[READ ORIGINAL →]