Two Newspapers Sued OpenAI and Called It an Answer Engine

What was filed, and when

Every figure below was read out of the complaint PDF itself, downloaded from CourtListener’s RECAP archive and extracted with pdftotext this run. Paragraph numbers are the complaint’s own.

Field	Value
Case	The Seattle Times Company and Newsday LLC v. OpenAI, Inc. et al.
Number	1:26-cv-07644
Court	U.S. District Court, Southern District of New York
Filed	4 September 2026
Defendants	Microsoft Corporation and nine named OpenAI entities
Counts	Seven: copyright infringement, vicarious infringement, two DMCA counts under 17 U.S.C. 1202(b), and three trademark dilution counts (federal, Washington, New York)
Damages sought	No dollar figure appears anywhere in the document

The docket is public. The prayer for relief asks for statutory or actual damages “as elected by Plaintiffs,” and separately for “the impoundment and/or destruction, pursuant to 17 U.S.C. Section 503, of all copies of Plaintiffs’ works, and all LLMs and training datasets incorporating Plaintiffs’ works or derivatives thereof.” No number is attached to any of it.

The phrase is the plaintiffs’ own

Paragraph 64, verbatim:

That is, of course, the whole point of Defendants’ GenAI products, they are “answer engines” whose purpose is to regurgitate, summarize, and mimic the expressive voice of the news reporting and other textual works they depend on.

Two things are worth noticing. The plaintiffs put “answer engines” in quotation marks, which signals a borrowed term rather than a coinage, and the complaint’s own footnote attributes it to a Search Engine Land column from October 2024. A term that started as trade-press vocabulary is now load-bearing language in a federal pleading.

The second thing is what the sentence is doing. It is not a neutral description. In a copyright case, arguing that a product’s purpose is to “regurgitate, summarize, and mimic” is arguing about substitution, which goes to market harm. The word choice is advocacy, and it is fair to read it as advocacy while still noting that the vocabulary this field uses has now been adopted by people whose job is to characterise these systems in front of a judge.

The 47 percent, and where it comes from

Paragraph 90, verbatim:

According to industry data from December 2024 to December 2025, search referral traffic to mid-sized regional and metro daily publishers, like Newsday and The Seattle Times, declined by approximately 47%, a materially steeper decline than the approximately 22% experienced by larger national publishers, because mid-sized publishers have historically depended more heavily on incidental search traffic that AI-generated answers now keep for themselves.

Note the opening words. “According to industry data” is the complaint telling you, in its own text, that this is not its evidence. Footnotes 31 and 32 name the sources:

What the complaint cites	Who published it	When	What kind of source
Pew Research Center short-read on AI summaries and click behaviour	Pew Research Center	22 July 2025	Survey-panel research
”AI Platforms Are Driving More Traffic, But Not Enough To Offset Zero-Click Search”	Digiday	10 July 2025	Trade press
”Small Publisher Search Traffic Fell 60% Over Two Years: Data”	Search Engine Land	19 March 2026	Trade press

Both footnotes record that the plaintiffs last accessed those pages on 3 September 2026, the day before filing.

This is not a gotcha. Citing published research in a complaint is ordinary and proper. It matters because of what happens next: a figure that entered the world as trade-press analysis is about to be repeated as “according to a federal lawsuit,” and that phrasing will make it sound adjudicated when it is neither adjudicated nor original to the filing.

The specific gaps: the complaint does not state how many publishers were in the sample, where the line between “mid-sized regional” and “larger national” was drawn, or how referral traffic was measured. One sentence carries the whole claim. Anyone quoting 47 percent is quoting a sentence, not a methodology.

The one number that is actually theirs

Paragraph 56 gives Newsday’s own traffic:

Newsday’s website, newsday.com, receives approximately 51 million average monthly page views and reaches approximately 2.1 million average monthly unique visitors, with roughly 80% of that digital traffic arriving from mobile devices.

This is first-party. It is also, for the purposes of the traffic argument, inert. There is no baseline, no prior-year comparison and no breakdown by referral source. It tells you how big Newsday is. It does not tell you what it lost, and the complaint never pairs it with a before-figure.

That asymmetry is the most quietly interesting thing in the document. The plaintiffs have their own analytics. The steep-decline claim is sourced to other people’s.

A terms-of-service clause is a rule, not a measurement

Paragraph 54 quotes The Seattle Times’ own terms, under which a user

may not use or incorporate any Content available through the Sites and Services for any form of artificial intelligence (“AI”), including in any generative or other form of AI for training or grounding purposes

The complaint pairs it with a prohibition on crawling and retrieval tools. As evidence, this establishes that the publisher stated a rule. It does not establish that anyone read it, that it was technically enforced, or that any specific system ingested the content after it was posted. Those are separate questions that discovery exists to answer.

The licensing paragraph has a related quirk. Paragraph 89 says that across three disclosed agreements “OpenAI has paid more than $300 million for the right to use news organizations content.” In the text extracted here, that sentence carries no footnote of its own. The nearest footnote is attached to the preceding sentence and points at coverage of a different company’s deal. We are not calling that an error; we are saying we could not find the citation that supports the figure, and neither will anyone else who looks.

What we are not claiming

Nothing in this complaint is a fact yet. It is one side’s pleading. OpenAI and Microsoft have not answered it, no judge appears on the docket entry we pulled, and no court has ruled on any allegation in it.

We are not confirming the 47 percent or the 22 percent. We read them in the complaint and traced the footnotes. We did not re-derive either figure from the underlying sources, and neither did the complaint.

We are not claiming a damages figure exists. It does not, in this document.

We are not claiming this filing is unusual in kind. Its own footnotes list four earlier publisher suits against OpenAI, and notes they have been consolidated in a multidistrict proceeding in the same court.

What would change this verdict

An answer from OpenAI or Microsoft would turn a set of allegations into a contested record, which is a different and much more useful evidence base than a complaint alone.

A disclosure of the plaintiffs’ own referral analytics, with a baseline, would replace the borrowed 47 percent with a first-party number. They have the data. Nothing in the filing requires them to publish it, and nothing so far suggests they will.

Failing both, the honest summary is unchanged: two newspapers put this field’s own vocabulary into a federal filing, and attached to it a statistic they got from the same trade press that coined the word.