Microsoft's summary judgment brief puts numbers on how rarely Copilot repeats news text

A copyright fight usually reaches the public through a press release from one side. This one arrived as a 46-page brief, filed on September 4, 2026 and posted in a redacted public version. Microsoft asked the court to decide the news publishers' claims in its favor without a trial, and it built much of the argument on counts drawn from the discovery record.

What 8.2 million chat logs showed

Microsoft produced an agreed sample of 8.2 million Copilot conversations. Each one holds the user prompt, the grounding content Copilot pulled from the web, and the response it generated. The sample was not random. The plaintiffs selected it with keywords pointing at their own sites, so it leans toward the conversations most likely to contain their articles.

Tom Goldstein, the expert for the Times and the Daily News, narrowed those logs to conversations where a plaintiff domain appeared in the grounding content, then searched for runs of 16 matching words. He found 59,545 of them. Microsoft's expert divided that figure by the full sample and reported a match rate of 0.73%. Emily Wenger, the expert for the Center for Investigative Reporting, found 51 instances of what she defined as substantial overlap.

One caveat travels with all three figures. They are reported in a brief written by the defendant, which decides what to include and how to present it. The counts themselves are the work of experts the plaintiffs hired, and that is what gives them weight.

The opt-out you can use today

The section with the shortest path to practice is the one on content controls. Microsoft announced in September 2023 that a publisher can add one meta tag to keep a page out of Copilot's web grounding entirely, and a second to limit how much of that page is used.

Microsoft then draws a legal consequence from it. The brief argues that a site owner who knows about these controls and chooses not to use them has granted an implied license. That is an argument, not a settled rule, and no judge has ruled on it.

• NOARCHIVE excludes the page from grounding altogether.
• NOCACHE limits grounding to a portion of the page.
• IndexNow pushes the edited page into the Bing index, so the exclusion registers sooner.
• robots.txt still governs crawling, while these meta tags govern what gets used.

What this filing is, and what it is not

Both sides moved on the same day. Microsoft asked for judgment in its favor, and the news plaintiffs filed their own combined brief and moved to seal parts of it. The court has decided neither motion, so nothing here is a ruling about fair use.

If you run a site, the practical part holds whichever way the case goes. The meta tags exist and work now, and the decision to leave them off is on the record as something a defendant will point to.