As parties continue to use generative AI tools in discovery, a recent decision from the Northern District of California, Schulte v. LinkedIn Corp. (N.D. Cal. June 30, 2026), provides welcome guidance. To assist in reviewing custodial documents for responsiveness, LinkedIn planned to use aiR, a generative AI-powered document review product offered within the Relativity platform. The plaintiffs raised several challenges. The court rejected all of them.
Three takeaways from the ruling stand out:
- First and foremost, the decision indicates that courts will treat generative AI review as a new form of traditional Technology Assisted Review (TAR) rather than a novel category of discovery.
- Second, parties will likely be allowed to use appropriate search terms to “pre-cull” documents before feeding them into a generative AI review tool.
- Third, courts remain resistant to permitting “discovery on discovery” when there is no evidence of a specific production deficiency.
Generative AI Review Is Treated as TAR
The most consequential aspect of the decision is the court’s treatment of generative AI review as familiar rather than novel. Rather than creating a separate legal framework for generative AI technology, the court analyzed generative AI review under the principles traditionally applied to TAR.
By grounding the analysis in established TAR frameworks, the court signaled that existing proportionality, reasonableness and transparency standards can accommodate generative AI workflows without requiring wholesale rethinking of discovery doctrine.
Parties May Use Search Terms to Pre-Filter Hits for Generative AI Review
The plaintiffs attacked LinkedIn’s decision to filter out documents that did not hit on search strings before documents were sent to Relativity aiR for generative AI review. The plaintiffs did not, however, dispute the appropriateness of the search strings. Rather, they argued that the approach improperly reduced the universe of documents reviewed by the AI system and risked excluding responsive materials. The court disagreed. Because the plaintiffs raised no challenge to the search strings themselves, the use of the strings to pre-cull the document population before AI-assisted review did not present a responsiveness problem. In so ruling, the court expressly relied on TAR cases, holding that search terms may be used to reduce review populations before TAR is conducted.
This ruling reinforces a practical reality of modern discovery—using generative AI does not eliminate proportionality concerns. Parties may continue to employ reasonable filtering methodologies, including negotiated search terms, to reduce large data volumes before undertaking AI-assisted review.
“Discovery on Discovery” Remains Disfavored
When a party discloses that it plans to use a generative AI review tool, the opposing side often seeks detailed metrics such as error rates, elusion estimates and protocols for human validation review. Rejecting what the plaintiffs were demanding from LinkedIn, the court reiterated the increasingly familiar principle that “discovery on discovery” is disfavored and appropriate only where a specific deficiency has been identified. Mere speculation that the process may be inadequate is not enough, even where the producing party plans to use generative AI.
This ruling offers useful authority for parties facing requests for AI-related metrics. To the extent parties want to ensure access to such information, they should consider including particularized disclosure requirements in ESI protocols before discovery begins.
Conclusion
The decision in Schulte v. LinkedIn Corp. reinforces that existing discovery rules can accommodate emerging technologies. It represents a meaningful step in the right direction for the adoption of generative AI in discovery, giving litigants greater confidence that thoughtfully designed AI-enabled workflows can satisfy reasonableness and proportionality standards while delivering the efficiencies demanded by modern discovery.
