This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
On Monday, Dow Jones and the New York Post, both owned by News Corp, sued Perplexity in federal court in New York. They allege a “massive amount of illegal copying,” and they describe Perplexity’s answers as a substitute for their journalism rather than a route to it. Perplexity did not immediately respond to requests for comment.
The legal questions will take a long time. Some lessons do not need to wait for a ruling.
1. A citation is not a license
Perplexity built its reputation on showing sources, and that has been a real point in its favor. The complaint argues that citing a story does not settle the question of whether you were allowed to use it. According to the publishers, Perplexity copied large amounts of their work into a database used for retrieval-augmented generation, then answered questions from it, sometimes reproducing text verbatim.
The distinction the publishers draw is between discovery and substitution. A search engine, in their telling, helps readers find the article. An answer engine that delivers the article’s value on its own page competes with it. Whatever a court decides, that is the line publishers are now drawing in public, and every answer product will be judged against it.
2. Product copy becomes evidence
One detail stood out to me. Perplexity’s own marketing has promoted the idea that users can “skip the links.” The complaint puts that phrase in its very first footnote.
That is a lesson well beyond AI. A tagline written to sound confident to users reads very differently to a plaintiff’s lawyer. For any company building or marketing a product that sits between people and other people’s content, the pitch is part of the record. The publishers also say they wrote to Perplexity in July offering to discuss a license and received no response, which is the kind of fact that shapes how a dispute is told later.
3. The most damaging error is the half-true one
The complaint also includes trademark claims, built on what it calls hallucinations attributed to the publishers’ names. When The New York Times sued OpenAI last December, it made a related argument, which an earlier post here looked at in detail. This case adds a sharper example. As WIRED reported, one cited answer allegedly reproduced two real paragraphs from a Post story word for word, then continued with five generated paragraphs that were not in the article.
From a reputation point of view, that pattern is worse than a pure fabrication. The accurate opening lends credibility to everything after it. A reader who checks the first lines finds them real and has little reason to doubt the rest.
Legal experts quoted by WIRED are skeptical that the trademark theory will succeed. That is worth noting for brands that are not publishers. If the law does not treat misattributed AI text as a trademark problem, the practical burden of catching it stays with whoever is being misquoted.
What to ask now
For communications teams, the useful question is simple: do the answer engines that cite your company treat your material as a destination or as raw material? Check whether answers about you link to the source, quote it accurately, and stop where the source stops. Watch especially for answers that start with your real words and drift into words you never said.