Internal Bing data says Copilot cut publisher click-throughs by up to 94%

What happened

The New York Times-led copyright case against OpenAI and Microsoft produced a public filing on Sept. 17, 2026. The part that got the headlines was executives describing their own products as "largely substitutive" for journalism and calling the situation an "existential threat" for publishers.

Buried under that is a number. PPC Land reports that internal Bing data cited in the record shows Copilot answers cut publisher click-through rates by as much as 94%.

That is the first time anyone outside Microsoft has put a measured figure on what an AI answer layer does to outbound clicks — and it came from the company running the answer layer, not from a publisher trade group.

What the number is not

It is not a 94% drop in your traffic. It is a click-through rate on specific queries where Copilot produced an answer. If Copilot is answering 4% of the queries that used to send you visitors, a 94% collapse on those queries is a rounding error on your dashboard. If it's answering your top 40 evergreen how-to queries, it is your quarter.

It is also "up to." That is the worst observed case in whatever slice was measured. The available reporting does not give an average, a query mix, or a date range. Anyone quoting "94%" as the typical outcome is quoting the ceiling as the floor.

And it is not a court ruling. Nothing has been decided. The filing is publishers arguing their side. Damages exposure, next court date, and whether any of this moves the judge — none of that is known from what's public.

What it does do is make the substitution argument empirical instead of rhetorical. Fair use weighs market harm. "Their own logs show a 94% click-through drop" is a different kind of exhibit than "we feel this is hurting us."

Why this changes your negotiating position, not your traffic

Your traffic already did whatever it was going to do. The Copilot effect, whatever its real size on your site, has been running for a long time. This filing did not cause it.

What changes is leverage. Google is already paying some publishers through its contribution pilot, with a formula nobody has seen. Every AI company negotiating a content license now does it knowing that a competitor's internal traffic data is in a federal court record. That is useful if you are in a licensing conversation and useless if you are not.

The second thing that changes: you can stop arguing about whether the substitution is real inside your own company. It is measured. Plan budgets accordingly.

What to do about it

1. Pull your own Bing and Copilot referral numbers for the last 18 months, separately from Google. Most publishers never break them out. You need the baseline before you can argue about the delta.
2. Rank your top 50 organic landing pages by how answerable they are in one paragraph. That list is your exposure, ordered.
3. Stop reporting "search traffic" as one line. Split answer-eligible queries from navigational and brand queries. The first is shrinking; the second mostly is not.
4. If you are in any licensing talk, get the filing in front of your lawyer before the next call. Internal admissions from one buyer set the tone for all of them.
5. Do not rebuild your editorial calendar around the 94% figure. Rebuild it around your own measured drop on your own answerable pages.
6. Assume no ruling this year. Litigation timelines are long and nothing about the next court date is public. Budget as if nobody is coming to fix this.