The opinion
1
2
3
4
5
6
7
8 UNITED STATES DISTRICT COURT
9 NORTHERN DISTRICT OF CALIFORNIA
10 San Francisco Division
11 KEVIN SCHULTE, et al., Case No. 22-cv-00237-HSG (LB)
12 Plaintiffs, DISCOVERY ORDER
13 v. Re: ECF No. 186, 188, 190
14 LINKEDIN CORPORATION,
15 Defendant.
16
17 INTRODUCTION
18 Three discovery letter briefs are before the court in this putative class action, where the
19 plaintiffs claim monopolization and attempted monopolization from practices that allegedly allow
20 LinkedIn to overcharge Premium subscribers, in violation of § 2 of the Sherman Act. The two
21 categories of alleged anticompetitive conduct are (1) offering potential rivals access to LinkedIn’s
22 private user data through application-programming interfaces (API) unavailable to others, on the
23 condition that the rivals do not compete with LinkedIn, and (2) integrating LinkedIn’s user data
24 with parent company Microsoft’s Azure cloud product, thereby “tying up and driving up prices for
25 scarce hardware resources.”1
26
27 1 Order Den. Prelim. Approval – ECF No. 135 at 1–2 (summarizing complaint); Disc. Letter Brs. –
ECF Nos. 186, 188, & 190. Citations refer to material in the Electronic Case File (ECF); pinpoint
1 ECF No. 186 concerns LinkedIn’s use of Relativity aiR, a GenAI product, for document
2 review. The plaintiffs’ requests — that the court (1) prohibit LinkedIn from using search strings to
3 pre-cull documents reviewed by Relativity aiR, (2) compel LinkedIn to use Relativity aiR across
4 all custodial files, and (3) compel LinkedIn to disclose additional metrics related to its use of the
5 platform — are denied. None of the requests are reasonable or proportional to the needs of the
6 case. The parties must meet and confer within twenty-one days of this order regarding LinkedIn’s
7 search strings and whether any adjustments are warranted.
8 ECF No. 188 is a motion to compel LinkedIn to add an in-house attorney, Jane Slater, as a
9 custodian. The plaintiffs assert that Ms. Slater has uniquely relevant information related to
10 negotiations with API partners. The motion is denied. LinkedIn’s custodial productions are
11 ongoing, and the plaintiffs have not shown that Ms. Slater possesses any non-privileged, non-
12 duplicative responsive information that is not captured by the existing custodial framework.
13 Finally, ECF No. 190 is a motion to compel LinkedIn to produce responsive text messages
14 from its nineteen designated custodians. The motion is denied. The Interim ESI Order explicitly
15 excludes text messages from the parties’ preservation obligations, and the plaintiffs have not
16 shown good cause for production. The evidence in the record suggests that text messages are
17 likely to be duplicative or cumulative of already designated sources. Any marginal relevance of
18 custodial text messages is overwhelmed by the substantial burden that the request would impose
19 on LinkedIn.
20
21 LEGAL STANDARD
22 Parties may obtain discovery regarding any nonprivileged matter that is relevant to any party’s
23 claim or defense and proportional to the needs of the case, considering the importance of the
24 issues at stake in the action, the amount in controversy, the parties’ relative access to relevant
25 information, the parties’ resources, the importance of the discovery in resolving the issues, and
26 whether the burden or expense of the proposed discovery outweighs its likely benefit. Information
27 within this scope of discovery need not be admissible in evidence to be discoverable. Fed. R. Civ.
1 Mut. Life Ins. Co., No. 17-cv-01760-SI, 2017 WL 3007030, at *1 (N.D. Cal. July 14, 2017)
2 (cleaned up) (quoting Shoen v. Shoen, 5 F.3d 1289, 1292 (9th Cir. 1993)). Moreover, “[t]he test
3 for relevance is not overly exacting: evidence is relevant if it has ‘any tendency to make . . . more
4 or less probable . . . [a] fact [that] is of consequence in determining the action.’” In re Volkswagen
5 “Clean Diesel” Mktg., Sales Pracs., & Prods. Liab. Litig., MDL No. 2672 CRB (JSC), 2018 WL
6 340640, at *1 (N.D. Cal. Jan. 9, 2018) (quoting Fed. R. Evid. 401).
7 The party moving to compel discovery “has the initial burden of establishing that the
8 information sought is relevant to any party’s claim or defense and proportional to the needs of the
9 case.” Impinj, Inc. v. NXP USA, Inc., No. 19-cv-03161-YGR (AGT), 2022 WL 16586886, at *2
10 (N.D. Cal. Nov. 1, 2022) (cleaned up). The party resisting discovery bears the burden of showing
11 that the discovery should not be allowed and of supporting its objections with competent evidence.
12 Lofton v. Verizon Wireless (VAW) LLC, 308 F.R.D. 276, 281 (N.D. Cal. 2015). Under Rule 26(b),
13 the court must limit discovery that is “unreasonably cumulative or duplicative,” obtainable from a
14 less burdensome source, or where the burden “outweighs its likely benefit.” Fed. R. Civ. P.
15 26(b)(1), (b)(2)(C).
16
17 ANALYSIS
18 1. LinkedIn’s Search Methodology (ECF No. 186)
19 The parties dispute two issues related to LinkedIn’s use of Relativity aiR. The first is whether
20 LinkedIn’s use of search strings to pre-cull the custodial documents given to Relativity aiR is
21 appropriate. The second is whether LinkedIn must disclose additional metrics related to its use of
22 the platform. The court addresses each issue separately.
23 1.1 Background
24 On May 15, 2026, LinkedIn provided the plaintiffs with twenty-five search strings to be
25 applied against LinkedIn’s custodial documents. At the same time, LinkedIn informed the
26
27
1 plaintiffs that it would be using Relativity aiR, a GenAI review tool, “to assist in filtering out non-
2 responsive documents” in its custodial production.2
3 On May 23, 2026, the plaintiffs requested more information about LinkedIn’s use of Relativity
4 aiR. LinkedIn disclosed that “(1) there was no seed or training set used; (2) Relativity aiR is being
5 used to make final responsiveness calls; and (3) quality control review is being conducted by
6 human review of samples taken from each responsiveness type.”3 LinkedIn said it was not
7 required to make any additional disclosures. The plaintiffs informed LinkedIn that its disclosures
8 were inadequate and that its use of search strings to “pre-cull” the target document population
9 given to Relativity aiR was “inappropriate.” The parties met and conferred on June 9, 2026, and
10 were unable to resolve the dispute.4
11 1.2 LinkedIn’s use of search strings
12 The plaintiffs request an order (1) prohibiting LinkedIn from applying search strings or
13 keywords to pre-cull the custodial documents reviewed by Relativity aiR and (2) compelling
14 LinkedIn to apply Relativity aiR across all agreed-upon custodial files.5
15 The plaintiffs argue that LinkedIn’s use of search strings before beginning the Relativity aiR
16 process is unnecessary. They contend that LinkedIn’s use of search strings “artificially reduce[s]”
17 the target population ingested by Relativity aiR, potentially removing responsive documents from
18 the target population.6
19 LinkedIn responds that its use of search strings is proper.7 It contends that it has satisfied its
20 obligations under the Interim ESI Order by disclosing its use of Relativity aiR to the plaintiffs and
21 that it has gone beyond the Order’s requirements by providing additional information about its use
22
23
24
2 Disc. Letter Br. – ECF No. 186 at 1 (cleaned up).
25 3 Id. (cleaned up).
26
4 Id.
5 Id. at 3.
27
6 Id. at 2.
1 of the platform.8 LinkedIn further argues that it would be disproportionate and unduly burdensome
2 to apply Relativity aiR to all custodial files without applying search strings first.9
3 The plaintiffs’ request is denied. Although the plaintiffs assert that LinkedIn’s use of search
4 strings “artificially reduce[s]” the amount of documents ingested by Relativity aiR, the plaintiffs
5 do not argue or otherwise show that LinkedIn’s twenty-five search strings are deficient.10 If, for
6 example, the plaintiffs had shown that LinkedIn’s search strings were too narrow, their concern
7 about pre-culling the document population might be warranted. But they have not made that
8 argument, and they do not contend that they raised any concerns with LinkedIn’s search strings
9 when they were disclosed on May 15, 2026.11 Courts have held that using search terms to pre-cull
10 documents before providing them to technology-review platforms satisfies the reasonableness and
11 proportionality standards of Rules 26(b) and 34(b)(2). E.g., Livingston v. City of Chicago, No. 16-
12 cv-10156, 2020 WL 5253848, at *3 (N.D. Ill. Sep. 3, 2020); In re Biomet M2a Magnum Hip
13 Implant Prods. Liab. Litig., No. 3:12-MD-2391, 2013 WL 1729682, at *2 (N.D. Ind. Apr. 18,
14 2013). The court finds the same here. If the plaintiffs have concerns about the target document
15 population LinkedIn is providing to Relativity aiR for responsiveness review, they can address
16 those concerns by requesting adjustments to LinkedIn’s search strings.
17 Also, prohibiting LinkedIn from using search strings to pre-cull the target document
18 population would impose a substantial burden on LinkedIn. The files of two LinkedIn custodians
19 alone total approximately 800 gigabytes of data.12 With nineteen designated custodians, LinkedIn
20 would be required to feed multiple terabytes of documents into Relativity aiR, resulting in
21 significant costs related to processing, hosting, and human review.13
22
23
24
8 Id. at 3 (citing Interim ESI Order – ECF No. 61 at 3 (¶ 5(a)).
25 9 Id. at 4.
26
10 Id. at 1–3.
11 Id. at 1.
27
12 Decl. of L. Schnepper, Ex. A to id. – ECF No. 186-1 at 3 (¶ 2).
1 The plaintiffs’ request is denied. LinkedIn’s use of search strings is proper, and it is not
2 required to apply Relativity aiR across all custodial files. Within twenty-one days of this order, the
3 parties are ordered to meet and confer regarding the search strings LinkedIn is using to create the
4 target document population for Relativity aiR. The plaintiffs may seek relief with the court if they
5 believe adjustments to the search strings are warranted and the parties are unable to reach
6 agreement.
7 1.3 Disclosure of Relativity aiR metrics
8 The plaintiffs separately move to compel LinkedIn to disclose various metrics about its use of
9 Relativity aiR, including elusion estimates, the document error rate, and the number of human
10 reviewers being used to validate Relativity aiR’s predictions.14 LinkedIn opposes the request.15
11 “As a general matter, discovery of another party’s evidence preservation and collection efforts
12 — or ‘discovery on discovery’ — is disfavored, as such discovery is typically not relevant to the
13 merits of a claim or defense, and is rarely proportional to the needs of a case.” Taylor v. Google
14 LLC, No. 20-CV-07956-VKD, 2024 WL 4947270, at *2 (N.D. Cal. Dec. 3, 2024). Such discovery
15 may be warranted “if the party requesting it demonstrates that there is a specific deficiency in the
16 other party’s production of documents or other information,” but “mere speculation” is
17 insufficient. Id.
18 The plaintiffs’ request is denied for two reasons. First, LinkedIn has satisfied its obligations
19 under the Interim ESI Order. Paragraph 5(a) of the Order requires the producing party to “disclose
20 to the receiving party if they intend to use Technology Assisted Review (“TAR”) to filter out non-
21 responsive documents.”16 On May 15, 2026, LinkedIn disclosed to the plaintiffs that it would use
22 Relativity aiR — a form of Technology Assisted Review — to filter out non-responsive
23
24
25
26
14 Id. at 3.
27
15 Id.
1 documents.17 And LinkedIn provided the plaintiffs with additional information upon request.18
2 These disclosures more than satisfy the demands of the Interim ESI Order.
3 Second, the plaintiffs have not shown that a further “audit” of LinkedIn’s use of Relativity aiR
4 is warranted.19 The only deficiency they identify is that LinkedIn’s target document population
5 amounts to 204,444 documents.20 Without a more specific showing of why LinkedIn’s production
6 is deficient, the number of documents in the target population alone does not warrant the requested
7 discovery on discovery. Id.
8 The plaintiffs’ motion to compel LinkedIn to make further disclosures regarding its use of
9 Relativity aiR is denied.
10
11 2. Adding In-House Attorney as a Custodian (ECF No. 188)
12 The plaintiffs seek to add a LinkedIn in-house attorney, Jane Slater, as a custodian. They move
13 to compel LinkedIn to collect, search, and produce her internal and external communications,
14 subject to a privilege review.21
15 The plaintiffs argue that Ms. Slater’s internal files are relevant because she is one of three
16 attorneys “responsible for negotiating all private API agreements.” They identify a purported gap
17 in LinkedIn’s current production that only Ms. Slater can fill; she participated in a set of API
18 partner negotiations where the only other LinkedIn employee was Rohan Verma, a business
19 executive who left LinkedIn in 2021 and whose files were not preserved.22
20 LinkedIn has already agreed to produce external communications between attorneys who
21 negotiated API agreements and LinkedIn’s API partners.23 It opposes the request to produce Ms.
22
23
17 Disc. Letter Br. – ECF No. 186 at 3.
24
18 See supra Section 1.1.
25 19 Disc. Letter Br. – ECF No. 186 at 3.
26
20 Id. at 2.
21 Disc. Letter Br. – ECF No. 188 at 1, 3.
27
22 Id. at 1.
1 Slater’s internal communications because (1) the plaintiffs have not shown that Ms. Slater
2 possesses nonprivileged, non-duplicative responsive information that is not available from existing
3 custodians and (2) the request is disproportionate to the needs of the case.24
4 As the court previously said, the requesting party bears the burden of showing that an
5 additional custodian possesses “uniquely relevant information that is not available from the
6 sources already designated.”25 The court explained that the plaintiffs could seek to add additional
7 attorney custodians if they made a “particularized showing” of “specific gaps” in LinkedIn’s
8 production that only the identified attorneys could fill.26
9 The plaintiffs’ request is denied. The plaintiffs contend that Ms. Slater is the only LinkedIn
10 employee who can fill the gap left by the deletion of Mr. Verma’s files.27 But LinkedIn is
11 producing documents from five current and former business development leaders responsible for
12 API partner relationships and strategy.28 Two of these custodians were Mr. Verma’s direct
13 supervisors during the relevant period.29 LinkedIn represents that Ms. Slater provided legal advice
14 to these business development leaders and communicated with API partners on their behalf.30 To
15 the extent any of Ms. Slater’s internal communications related to API agreements and partners are
16 not privileged, they are likely captured by the existing custodial framework. With fact discovery
17 ongoing and LinkedIn’s custodial productions “rolling,” it is premature to demand that Ms.
18 Slater’s internal communications be produced without evidence that her files will contain non-
19 privileged, non-duplicative responsive documents.31
20
21
24 Disc. Letter Br. – ECF No. 188 at 3–5.
25 Order – ECF No. 170 at 7 (quoting In re Facebook, Inc. Consumer Priv. User Profile Litig., No.
22 3:18-MD-02843-VC-JSC, 2021 WL 10282213, at *1 (N.D. Cal. Nov. 14, 2021)).
23
26 Id.
27 Disc. Letter Br. – ECF No. 188 at 1–2.
24
28 Id. at 4.
25 29 Id.
26
30 Id. at 3.
31 Id.; see also Order – ECF No. 170 at 7 (“That these attorneys signed the API agreements does not,
27 without more, establish that their custodial files will contain non-privileged, non-duplicative
responsive documents beyond those produced by the business-development custodians and the existing
1 The exemplary documents identified by the plaintiffs do not change this analysis. The cited
2 documents are communications Ms. Slater and Mr. Verma had with API partners — i.e., the type
3 of external communications between LinkedIn attorneys and API partners that LinkedIn has
4 already agreed to produce.32 The plaintiffs argue that these communications only tell one half of
5 the story, leaving out any internal LinkedIn communications concerning the specific
6 negotiations.33 But this gap is speculative. It relies on two assumptions: first, that Mr. Verma and
7 Ms. Slater had internal deliberations about API negotiations without any other custodians, and
8 second, that any such internal communications were made in Ms. Slater’s “business-negotiation
9 capacity” and are nonprivileged.34 The plaintiffs point to no evidence in the record supporting
10 either assumption, and the evidence that does exist suggests that Ms. Slater’s role in negotiations
11 was to provide legal advice about the terms of API agreements.35 Her internal communications are
12 likely to be overwhelmingly privileged, such that the burden of collecting, reviewing, and logging
13 these files would outweigh the benefit of any nonprivileged communications not already captured
14 by existing custodians.
15 The plaintiffs’ motion to compel the production of Ms. Slater’s relevant custodial files is
16 denied. LinkedIn must produce all responsive external communications between Ms. Slater and
17 API partners, as it has already agreed to do.36
18
19 3. Collection and Production of Text Messages (ECF No. 190)
20 The plaintiffs move to compel LinkedIn to preserve, collect, search, and produce all responsive
21 text messages (including iMessage, SMS, and app-based messages like WhatsApp or Signal),
22 whether sent or received on a personal or work-issued device, from all nineteen designated
23 custodians. The plaintiffs argue that good cause exists to require the production of text messages
24
25 32 Disc. Letter Br. – ECF No. 188 at 3.
26
33 Id. at 2–3.
34 Id. at 3.
27
35 Id. at 4; Email, Ex. A to id. – ECF No. 188-1.
1 because LinkedIn executives “routinely us[e] text messages for business purposes” and to discuss
2 “topics relevant to the claims and defenses” at issue.37
3 LinkedIn contends that it has no obligation to produce text messages in this case. It asserts that
4 (1) the plaintiffs have not shown good cause to set aside the Interim ESI Order’s exclusion of text
5 messages from default preservation and production obligations, (2) LinkedIn is producing
6 documents from the three repositories it has identified as most likely to contain non-duplicative,
7 responsive information, and (3) the plaintiffs’ request would impose an undue burden on
8 LinkedIn.38
9 Paragraph 4(b) of the Interim ESI Order provides that, absent a showing of good cause by the
10 requesting party, “[t]ext messages (e.g., Slack, MMS, iMessages) and data from app-based
11 messaging systems (e.g., WhatsApp, Signal, and Line)” need not be preserved.39 At minimum, this
12 requires the plaintiffs to show that their requested discovery is relevant and proportional to the
13 needs of the case and that its likely benefit outweighs the burden it imposes. Fed. R. Civ. P.
14 26(b)(1). The court must limit discovery that is unreasonably cumulative or duplicative. Fed. R.
15 Civ. P. 26(b)(2)(c)(i).
16 The plaintiffs’ request is denied for two reasons. First, the plaintiffs have not shown good
17 cause to set aside the Interim ESI Order’s exclusion of text messages from the parties’
18 preservation obligations. To establish the relevance of the requested discovery, the plaintiffs
19 primarily rely on produced emails that refer to sending or receiving text messages.40 When viewed
20 in context, these references suggest that any related text messages are largely logistical,
21 cumulative of produced communications, or otherwise irrelevant. For example, one email cited by
22 the plaintiffs shows a LinkedIn employee asking a colleague to “text if you need anything urgent”
23 while the employee was traveling.41 The plaintiffs do not offer any evidence that this request is in
24
25 37 Disc. Letter Br. – ECF No. 190 at 1, 3.
26
38 Id. at 3.
39 Interim ESI Order – ECF No. 61 at 2–3 (¶ 4(b)).
27
40 Disc. Letter Br. – ECF No. 190 at 2–3.
1 reference to anything at issue in the case. Another email shows an employee providing an
2 “overview of our progress” and asking others on the thread to “text me for anything urgent” while
3 he was on leave.42 Even if the provided “overview” is relevant to the claims and defenses in this
4 case, it is already captured in the produced email. The same is true for a separate email relied on
5 by the plaintiffs, which states that its substance is the same “as shared via text.”43
6 The strongest example the plaintiffs provide to support relevance is a produced screenshot of
7 text messages referencing topics for an upcoming “QBR.”44 But LinkedIn represents that it has
8 already produced the final prep documents for that “QBR,” supporting that any related texts are
9 likely duplicative.45 The plaintiffs have not pointed to a single substantive business discussion
10 occurring by text or to any discussion that is not already captured by existing sources. Without
11 such a showing, they cannot establish good cause for their requested discovery.
12 Second, any marginal relevance that the plaintiffs have shown is overwhelmed by the
13 substantial burden their request would impose on LinkedIn. The plaintiffs’ request applies to text
14 messages from all nineteen designated custodians, across multiple messaging platforms (including
15 iMessage, SMS, WhatsApp, and Signal), on both work-issued and personal devices.46 Such
16 production would involve significant time, expense, and disruption to the private lives of each
17 designated custodian.47 Even if the plaintiffs had shown that text messages are likely to result in
18 non-duplicative and responsive information, their request would still be untethered from
19 reasonable and proportionate discovery.
20 Separately, the parties dispute whether LinkedIn has provided the factual basis for its
21 conclusion that three select repositories — M365 Mailboxes, Google Drive, and Slack — are most
22
23
24
42 Id. (citing LI_CROWDER00074407).
25 43 Id. at 4 (citing LI_CROWDER_00150755).
26
44 Id. at 1–2 (citing LI_CROWDER_00112911), 4 (discussing document).
45 Id. at 4 (citing LI_CROWDER_00112911, LI_CROWDER_00163661).
27
46 Id. at 3.
] likely to contain relevant and non-duplicative information.** But there is evidence that LinkedIn
2 || investigated and selected its repositories in good faith; it voluntarily included Slack, which is
3 exempted under the Interim ESI Order as a category of text messages.” In the absence of evidence
4 || showing that other forms of text messages are likely to be responsive and non-cumulative
5 communications, LinkedIn’s selected repositories reflect reasonable efforts to search for, collect,
6 || and produce relevant and proportional information.”
7 The plaintiffs’ motion to compel is denied. LinkedIn is not required to preserve, collect,
8 search, and produce responsive text messages from its nineteen designated custodians.
9
10 CONCLUSION
1] This resolves ECF No. 186, 188, and 190.
a 12 IT IS SO ORDERED.
13 Dated: June 30, 2026 LAE
MM —
LAUREL BEELER
15 United States Magistrate Judge
16
17
Z 18
19
20
21
22
23 ‘8 Compare id. at | (“Plaintiffs have no insight into LinkedIn’s factual basis for this position (and the
basis was never articulated during the meet and confer process)”), with id. at 5 (“LinkedIn
24 communicated the results of its investigation on multiple occasions, including in its June 1, 2026,
correspondence.”).
|| 49 Interim ESI Order — ECF No. 61 at 3 4(b)).
26 %° See The Sedona Conference, The Sedona Principles, Third Edition: Best Practices,
Recommendations & Principles for Addressing Electronic Document Production, 19 Sedona Conf. J.
27 1, 119-20 (2018) (“[D]etermining what is relevant and proportional under the circumstances for each
matter often requires a highly fact-specific inquiry” that the “responding party” is in “a better position
28 to make.”’).