Requesting Parties Are Denied “Input” Into Producing Party OpenAI’s Search Terms

In Tremblay v. OpenAI, Inc., 2025 WL 635335 (N.D. Cal. Feb. 27, 2025), plaintiffs, the requesting parties, sought to have input in determining search terms to be used by defendant, the producing party. Based on “the specter of too many future delays and disputes over methodology and search term formulation,” the request was denied.

Two subsequent requests were also denied.

Several months after the first request, “Plaintiffs nevertheless alluded to gaps in the [defendant’s] production and once again sought to inject themselves into the ESI search-term formulation process.” The second request was denied.

Sedona Principle 6 is: “Responding parties are best situated to evaluate the procedures, methodologies, and technologies appropriate for preserving and producing their own electronically stored information.” The Sedona Principles, “Best Practices, Recommendations & Principles for Addressing Electronic Document Production,” 19 Sedona Conf. J. 1, 118 (3^rd ed. 2018)

Principle 7 is: “The requesting party has the burden on a motion to compel to show that the responding party’s steps to preserve and produce relevant electronically stored information were inadequate.” Id. at 131.

Sedona emphasizes the goal and value of cooperation. Id. at 77, passim.

In Tremblay, after the second denial, the court continued:

It appears that Plaintiffs now seek to litigate that issue for a third time, and by doing so, it appears that the objective of the court’s initial ruling (i.e., to prevent delays and the stalling of the discovery process stemming from endless discovery disputes over methodology and search-term formulation) has been nevertheless frustrated.

Tremblay v. OpenAI, Inc., 2025 WL 635335 (N.D. Cal. Feb. 27, 2025).

Plaintiffs asserted that OpenAI’s search terms and connectors “appeared too narrow or didn’t capture lingo used by OpenAI’s own employees….” Further, the court wrote that:

Plaintiffs add that their proposed eight search “strings” are not burdensome because they would yield only 345,000 documents, which “[w]ith technology assistance, e-discovery vendor DISCO estimates a 345,000-document review takes 2-3 weeks, [from] upload through production.”

Id.

OpenAI responded that:

Plaintiffs’ request now not only seeks the addition of “numerous torrent-related search terms,^[1] but also [ ] hundreds of more terms packed into eight search strings, including nonsensically overbroad terms like (ChatGPT AND “be doing”), (memori* AND data), and (seed*) … [which] [b]ased on review time to date, Plaintiffs’ terms would add over 9,000 hours of attorney review time with no showing that OpenAI’s productions are deficient.”

Id.

OpenAI asserted that the addition of 362 search terms would not be proportionate, because it would add 345,000 documents to the 640,000 that had already been produced.

OpenAI prevailed:

The court agrees with OpenAI’s assessment here. Without repeating the details of the Parties’ disputes as to the numerous search terms included in Plaintiffs’ proposed search strings… the court cannot avoid the conclusion that Plaintiffs are trying to relitigate an issue they have already lost (i.e., that they should be involved in the search term formulation process without making a clear showing of prejudice stemming from gaps or deficiencies in the search terms disclosed by the producing Party). As a result, Plaintiffs’ request for an order directing OpenAI to deploy its proposed search strings is DENIED.

Id.

Here, the requestor appears to have been unable to convince the court that there were defects in the producer’s search methodology.

The Tremblay v. OpenAI court resolved a number of other disputes, stating:

• “[T]he timing and nature of this dispute (in conjunction with the other disputes that are the subject of this order) gives rise to the impression that the Parties may be using the discovery process, at least in part, to engage one another in a warfare of sorts, that is, by attempting to fashion certain discovery disputes in order to harass one another’s flanks as might be the case during a game of chess or on a medieval battlefield.”
• “The court finds that this dispute is representative of the Parties talking past one another – that it is the manifestation of poor meet-and-confer efforts – and, that the ‘dispute’ evades judicial resolution as it has been presented. Thus, the Parties are ORDERED to meet-and-confer forthwith….”
• “Unfortunately, it once again appears that the Parties are practically speaking different languages and that they are arguing past one another….”
• “If any dispute shall arise as to relevance or proportionality, the Parties are ORDERED to meet and confer such as to resolve or substantially narrow that dispute so that it can be articulated in a clear and cohesive manner that would lend itself to a ruling which would not require the expenditure of substantial effort in attempting to square the Parties’ arguments with one another.”
• “This dispute too seems to be the product of the Parties seemingly taking turns to simply conjure up something about which to fight. The court expects counsel to resolve this sort of dispute without the need for court intervention.”
• “The court finds this sort of back-and-forth to constitute yet another manifestation of the Parties’ drumming up a discovery fight about a subject that should have been worked out by the Parties themselves without the need for court intervention. The dispute itself is insubstantial – the essence of which boils down to little more than petty finger-pointing about whose time was wasted by whom.”

Judge Peck criticized “lawyers designing keyword searches in the dark, by the seat of the pants, without adequate (indeed, here, apparently without any) discussion with those who wrote the emails.” His Honor added:

Of course, the best solution in the entire area of electronic discovery is cooperation among counsel. This Court strongly endorses The Sedona Conference Cooperation Proclamation (available at www.TheSedonaConference.org)…. Electronic discovery requires cooperation between opposing counsel and transparency in all aspects of preservation and production of ESI. Moreover, where counsel are using keyword searches for retrieval of ESI, they at a minimum must carefully craft the appropriate keywords, with input from the ESI’s custodians as to the words and abbreviations they use, and the proposed methodology must be quality control tested to assure accuracy in retrieval and elimination of “false positives.” It is time that the Bar—even those lawyers who did not come of age in the computer era—understand this.

William A. Gross Const. Assocs., Inc. v. Am. Mfrs. Mut. Ins. Co., 256 F.R.D. 134, 134 (S.D.N.Y. 2009).

The Hon. Paul W. Grimm wrote that “challenges to the sufficiency of keyword search methodology unavoidably involve scientific, technical and scientific subjects, and ipse dixit pronouncements from lawyers unsupported by an affidavit or other showing that the search methodology was effective for its intended purpose are of little value to a trial judge who must decide a discovery motion aimed at either compelling a more comprehensive search or preventing one.” Victor Stanley, Inc. v. Creative Pipe, Inc., 250 F.R.D. 251 n. 10 (D. Md. 2008). Judge Grimm added:

It cannot credibly be denied that resolving contested issues of whether a particular search and information retrieval method was appropriate—in the context of a motion to compel or motion for protective order—involves scientific, technical or specialized information. If so, then the trial judge must decide a method’s appropriateness with the benefit of information from some reliable source—whether an affidavit from a qualified expert, a learned treatise, or, if appropriate, from information judicially noticed. To suggest otherwise is to condemn the trial court to making difficult decisions on inadequate information, which cannot be an outcome that anyone would advocate.

For example, in the analogous technical area of sampling ESI, courts have recognized the need to have expert assistance to develop a valid random sampling protocol.

Id.

Judge Grimm wrote that “use of search and information retrieval methodology, for the purpose of identifying and withholding privileged or work-product protected information from production, requires the utmost care in selecting methodology that is appropriate for the task…. The implementation of the methodology selected should be tested for quality assurance; and the party selecting the methodology must be prepared to explain the rationale for the method chosen to the court, demonstrate that it is appropriate for the task, and show that it was properly implemented.” [Emphasis added].

Notes

[1] At least some “torrent” documents had been produced.