Microsoft: Almost nobody's grabbing NYT articles through Copilot

Microsoft's legal team just served up 8.2 million Copilot chat logs to an expert hired by publishers like the New York Times — the whole point being to prove that people aren't actually using the chatbot to steal content. The company picked logs that matched keywords from the plaintiffs' sites, then says the analysis shows 59,545 hits, which amounts to barely anything. Copilot rarely reproduces full sentences from news articles or books, let alone substantive chunks that could replace the originals, according to the filings.

The broader picture: news publishers have been threatening AI companies for years, claiming the models are trained on their work and serving it back. Microsoft is fighting back by showing the demand side is essentially flat — users aren't mining Copilot for article content the way they mine Google. That doesn't mean the training data dispute is settled, but it does weaken the claim that these products are cannibalizing publisher traffic at scale.

Why this matters for us: if the demand is that low, the real leverage stays with the big tech companies and the publishers keep paying the rent on their own sites — la gente reads the original, not the chatbot summary.