In Encyclopaedia Britannica, Inc. v. Perplexity AI, Inc., No. 25 Civ. 7546 (S.D.N.Y. Sept. 2, 2026), New York Magistrate Judge Sarah L. Cave granted in part and denied in part the plaintiffs’ request for additional discovery from Perplexity’s retrieval augmented generation (“RAG”) database and user activity log (“UAL”). Judge Cave ordered Perplexity to provide one additional RAG snapshot and host six months of additional UAL data, rather than the ten months requested. She also required the plaintiffs to contribute $6,000 per month toward hosting the additional UAL data, balancing the discovery’s relevance against the burden and expense of making it available.

Case Background and Judge’s Ruling

Britannica and Merriam-Webster alleged that Perplexity infringed their copyrights by copying their content to support its AI answer engine and generating responses that reproduced, paraphrased, or summarized their copyrighted works. They also asserted trademark claims involving allegedly fabricated content attributed to them, misleading omissions, and uses of their trademarks that could confuse users about the source or approval of responses. Perplexity’s motion to dismiss the copyright claim concerning its outputs remained pending; this order addressed discovery rather than resolving those underlying allegations.

The disputed databases captured different aspects of Perplexity’s operations. The RAG database stored indexed web content used to source and verify answers. The UAL database recorded user queries, conversation history, session metadata, retrieval and processing activity, instructions to language models, and final responses. Producing RAG snapshots required substantial engineering work, while UAL data had to be extracted and processed from “deep storage.” Perplexity retained a consulting firm to produce snapshots and make UAL data available for inspection in a secure environment.

Perplexity had already provided the plaintiffs with discovery assembled for separate litigation brought by Dow Jones and NYP Holdings. That discovery included two RAG snapshots totaling more than 300 terabytes and UAL data covering May 2024 through May 2025. However, thirteen of Britannica’s and Merriam-Webster’s copyright registrations became effective between April and December 2025. The plaintiffs therefore sought additional RAG and UAL data covering August 2025 through May 2026. Perplexity proposed cost sharing and otherwise objected that the requested production would be unduly burdensome and disproportionate.

Judge Cave organized her analysis around relevance, proportionality, and hosting costs. On relevance, the plaintiffs argued that discovery should cover their registration periods and surrounding months, as discovery in the Dow Jones litigation had done. Perplexity principally countered with merits arguments, including its disagreement that each RAG retrieval constituted an independent infringement and its arguments supporting dismissal of the output claim.

Agreeing with the plaintiffs, Judge Cave stated that, “for purposes of determining the relevant time frame for Perplexity’s production of RAG and UAL data, we must look to the period of Plaintiffs’ copyright registrations.” Existing discovery overlapped only partially with that period. Consequently, she ordered Perplexity to produce an additional RAG snapshot dated after August 2025 and additional UAL data beginning that month. Having established relevance on this basis, she did not reach the plaintiffs’ additional arguments concerning statutory damages, fair use, and willfulness.

Relevance did not, however, justify the full requested production. Perplexity submitted declarations asserting that the additional discovery would consume hundreds of engineering hours and generate incremental costs of hundreds of thousands of dollars per month. The plaintiffs challenged those estimates, pointing to previously quoted hosting costs of approximately $12,000 to $14,000 monthly and supplying alternative estimates in the tens of thousands.

Judge Cave declined to settle the competing calculations: “We find it unnecessary to don green eyeshades to decipher which party has the more accurate calculation of cost and burden.” Nonetheless, she recognized the practical burden: “Suffice it to say, requiring Perplexity to produce any of the Additional Data, let alone ten months’ worth, is not a small undertaking.”

Exercising judicial discretion, Judge Cave limited discovery to one additional RAG snapshot and UAL data from August 2025 through January 2026, extending a little over a month beyond the latest registration. She found that this scope “appropriately balances Plaintiffs’ need for information to determine whether and to what extent Perplexity has infringed their copyrights and trademarks against the expense and burden on Perplexity of assembling the data.” The parties were directed to meet and confer about the snapshot’s date or period.

On costs, Judge Cave applied the presumption that the responding party pays discovery expenses, while recognizing that Rule 26(c)(1)(B) permits allocation for good cause. Perplexity bore the burden of justifying a departure. Although she viewed its dramatically increased hosting estimates “with considerable skepticism,” she found good cause for partial sharing: hosting the additional UAL data imposed significant expense and burden relatively late in document discovery.

Judge Cave therefore required the plaintiffs to share UAL hosting costs “up to the amount of $6,000 per month”. The plaintiffs had separately agreed to pay to host additional RAG data produced to them. The resulting order required Perplexity to produce the new RAG snapshot and host six months of UAL data for inspection, while denying the full ten-month demand. Judge Cave’s ruling illustrates a tailored approach to discovery of voluminous AI system records: establish the relevant period, narrow production to a proportionate scope, and allocate part of the hosting expense when the demonstrated burden warrants it.

So, what do you think? Do you agree that an additional RAG snapshot and six months of additional UAL data was a reasonable compromise? Please share any comments you might have or if you’d like to know more about a particular topic.

Case opinion link courtesy of Minerva26, an Affinity partner of eDiscovery Today. Catch Kelly Twigger’s discussion of this case here!

Disclaimer: The views represented herein are exclusively the views of the author, and do not necessarily represent the views held by my employer, my partners or my clients. eDiscovery Today is made available solely for educational purposes to provide general information about general eDiscovery principles and not to provide specific legal advice applicable to any particular circumstance. eDiscovery Today should not be used as a substitute for competent legal advice from a lawyer you have retained and who has agreed to represent you.