TL;DR
- A summary judgment brief filed on 17 September 2026 shows Microsoft's own data put the click-through reduction between Bing Chat and Bing web search at 87 to 93 percent for The Times's sites, reported by PPC Land.
- The Daily News sites fell 83 to 91 percent and Ziff Davis 51 to 94 percent; the brief also says at least two million Copilot conversations grounded outputs on the publishers' content.
- Company quotes explain it: an OpenAI engineer said 'no matter how prominently we show the links, users won't click', and Copilot's home page offered to talk instead of clicking.
- The effect is corroborated elsewhere: a randomised study found Google's AI Overviews cut outbound clicks about 39 percent, and Cloudflare cites roughly 1,091 OpenAI crawls per referral.
- For Thai brands the lesson is timing: use the window before AI erodes Thai-language clicks to become the cited source and build a recognised brand, on top of the SEO that still wins the clicks that remain.
Microsoft's own data shows its AI chat cut publisher click-through rates by 87 to 93 percent for The New York Times's sites compared with Bing web search, according to a summary judgment brief five publishers filed on 17 September 2026. PPC Land reported the figures, which are the clearest numbers yet, drawn from a platform's own measurement, on how much an answer engine collapses the click compared with a results page for the same sites.
The context is a copyright case, and the legal fight is not the point for marketers. The point is the traffic evidence, because for the first time it comes from a defendant's internal comparison of an AI answer surface against search results for the same domains, rather than from a third-party estimate. That makes it the hardest data available on what AI answers do to referral clicks.
The numbers, and where they come from
The brief was filed in the Southern District of New York by The New York Times Company, eight Daily News titles, Ziff Davis, the Center for Investigative Reporting and The Intercept, against Microsoft and OpenAI. Alongside the copyright claims, it discloses click-through data that was redacted when the motions were first lodged on 4 September.
| Publisher group | Click-through reduction, Bing Chat vs Bing web search |
|---|---|
| The Times's sites | 87 to 93 percent |
| The Daily News sites | 83 to 91 percent |
| Ziff Davis sites | 51 to 94 percent |
These are Microsoft's own representative figures as presented in the brief, comparing the click-through rate from Bing Chat against Bing web search for the same publisher domains. The brief also states that at least two million Copilot conversations grounded their outputs on the publishers' content, and in one example Copilot retrieved more than 340 words from a Times article and output around 270 of them to the user.
The quotes that explain the mechanism
The brief quotes the companies describing the effect in their own words, and those lines explain why the numbers are so large. An OpenAI software engineer, according to the brief, put the traffic problem bluntly: "no matter how prominently we show the links, users won't click." A senior figure wrote that once ChatGPT's browse function gave an answer, there was "no good reason to click." Copilot's home page greeted users with the line: "Instead of clicking through links, we can talk through whatever you're curious about."
That is the design intent stated plainly. An answer engine is built to end the query inside the answer, and prominent links do not change the behaviour, because a user who has the answer has no reason to leave. The 87-to-93 percent reduction is not a bug in the link placement; it is the product working as designed.
The same effect shows up on Google
The same dynamic has now been measured on the Google side by independent researchers. A randomised experiment with 1,065 users found AI Overviews cut outbound publisher clicks by around 39 percent, a smaller figure than the Bing Chat ranges but a causal one from a controlled study rather than an internal comparison. Cloudflare has separately described OpenAI operating at roughly 1,091 crawls per referral, a scrape-to-visit ratio that tells the same story from the crawler side: the content is being consumed far more than it is being sent traffic in return.
Put together, the picture is consistent across sources and surfaces. Whether measured by a platform's internal data, a randomised experiment, or crawl-to-referral ratios, the click that content used to earn is collapsing as answers move into the AI surface. The numbers differ; the direction does not.
What this means for anyone who depends on the click
For publishers the implication is existential, but the lesson generalises to any business that has relied on ranking a page and collecting the click. If the click is falling by double digits on Google's AI surfaces and by up to 90-plus percent in a chat answer, then a strategy built entirely on winning the click is building on a shrinking base. The value is moving from the visit to the mention: being the source the answer is built from, and being named in it, even when no click follows.
That does not make clicks worthless, and it does not mean abandoning search. It means adding a second objective alongside ranking, which is to be cited and quoted by the answer engines, so that a brand still benefits from appearing in an answer that ends without a visit. It also raises the value of the audiences a brand owns outright, where no intermediary decides whether the click happens. The businesses that struggle will be the ones whose entire model assumed the click would keep coming.
What this means for Thai marketers
Thai brands are earlier in this curve than US publishers, which is an advantage if it is used. AI answer surfaces are still reaching Thai-language queries more slowly than English ones, so there is a window to build the two things that matter before the click erodes as far here as it has in the US. The first is citability: content structured so that AI Overviews, AI Mode and chat assistants can lift a clean, self-contained fact and attribute it, which is the aim of generative engine optimisation and the newer AI SEO work. The second is brand strength, because a name a reader already recognises in an answer is worth far more than an unknown one, and recognition is what converts a zero-click impression into a later branded search or direct visit.
The mistake would be to treat these figures as a distant American problem. The mechanism is the product design of every major answer engine, and it arrives in Thai search on the same trajectory, just later. A Thai brand that spends the window making its content the source AI answers cite, and building a name readers trust, will be positioned for the zero-click market rather than surprised by it. That work sits on top of, not instead of, the SEO that still wins the clicks that remain.
What the data does not settle
The figures are strong but bounded. They are Microsoft's representative data as characterised in an adversarial brief, not an independently audited dataset, and the ranges are wide, especially Ziff Davis's 51 to 94 percent, which shows the effect varies a lot by site and content type. The brief measures Bing Chat and Copilot, not Google's AI Overviews or AI Mode, where the independent figure is the lower 39 percent. And none of this predicts the endpoint: whether the click stabilises at a new lower level, or keeps falling as answer engines take more query volume. Treat the numbers as confirmation of direction and rough magnitude, not as a fixed forecast.
Frequently asked questions
How much did Microsoft's AI cut publisher clicks?
According to a summary judgment brief filed on 17 September 2026, Microsoft's own data showed the click-through rate reduction between Bing Chat and Bing web search was 87 to 93 percent for The Times's sites, 83 to 91 percent for the Daily News sites, and 51 to 94 percent for Ziff Davis sites.
Are these figures reliable?
They are Microsoft's own representative data as presented in the publishers' brief, which makes them the first click-through numbers in this litigation drawn from a defendant's own measurement of an answer engine against search results for the same domains. They are not an independently audited dataset, and the ranges are wide, so treat them as direction and rough magnitude.
Does the same thing happen on Google?
Yes, at a smaller measured scale. A randomised experiment with 1,065 users found Google's AI Overviews cut outbound publisher clicks by about 39 percent. Cloudflare has also described OpenAI running at roughly 1,091 crawls per referral, which tells the same story from the crawler side.
Why are the reductions so large?
Because answer engines are designed to end the query inside the answer. The brief quotes an OpenAI engineer saying that no matter how prominently links are shown, users will not click, and Copilot's own home page told users they could talk through a topic instead of clicking links. Prominent links do not change the behaviour when the answer is already there.
What should a Thai brand do about it?
Use the window before AI answers erode Thai-language clicks as far as English ones. Make your content citable so answer engines lift and attribute it, through generative engine optimisation and AI SEO, and build brand recognition so being named in an answer produces value even without a click. Keep doing the SEO that wins the clicks that remain.
The bottom line
A platform's own numbers now put the cost of AI answers to publisher clicks at up to 87 to 93 percent for a major site, and the companies' own words say prominent links do not fix it because the product is built to end the query in the answer. The same effect, measured smaller, is real on Google too. For Thai brands the takeaway is not panic but timing: the click is going to erode here as it has elsewhere, later, and the brands that use the window to become the cited source and to build a name readers trust will own the zero-click market instead of being caught out by it.
