Old Rules, New Tools: N.D. Cal. Applies Traditional TAR Principles to Generative AI Discovery

A California Ruling Lets You Use GenAI Review Like TAR — But Doesn't Let You Skip the Paperwork

On June 30, 2026, the Northern District of California ruled on a discovery dispute in Schulte v. LinkedIn Corp., where LinkedIn used Relativity's aiR, a generative AI document review tool, to review custodial documents for responsiveness. Plaintiffs objected on three fronts: that generative AI review deserved its own legal framework, that pre-filtering documents with search terms before AI review improperly shrank the review population, and that they were entitled to discovery about the review process itself. The court rejected all three objections. WilmerHale published its analysis of the decision on July 20, 2026, and the case is already being cited as the first clear read on how existing Technology Assisted Review doctrine maps onto generative AI tools.

Does this mean generative AI review gets easier legal treatment than TAR?

No — the court applied the same proportionality, reasonableness and transparency standards that already govern TAR, not a lighter or novel standard.

That's the headline finding, and it cuts both ways for buyers. It means you don't need to wait for bespoke generative-AI discovery rules before deploying an AI review tool — existing case law already covers you. But it also means the burden hasn't dropped. If your process is ever actually challenged, you're still expected to defend it on the same terms litigants have used to defend TAR for over a decade: was the methodology reasonable, was it proportional, was it transparent enough for the other side to trust the output.

Can I still narrow the document set with search terms before running AI review?

Yes — the court allowed pre-culling with negotiated search terms before AI review, so long as the terms themselves aren't disputed.

Plaintiffs didn't challenge LinkedIn's search strings — only the idea that filtering before AI review shrank the universe unfairly. The court leaned on prior TAR rulings that already permit this exact sequence. For platforms that ingest a full collection and then apply issue coding across everything, this is a reminder that a defensible workflow can still start with negotiated filtering; it doesn't have to run every document through the model to be sound.

Will courts force my opponent to hand over their AI review's error rates?

Not automatically — the court reaffirmed that "discovery on discovery" is disfavored unless a specific production deficiency has been identified, not just speculation.

This is the part worth sitting with. It's a win for producing parties, but it quietly shifts risk onto the requesting side — and onto you, if you're ever the one asking questions about an opponent's process. If you can't point to an actual gap in production, a court is not going to compel your opponent's elusion rates, validation protocol, or human-review sampling data. The corollary matters more: if you're the producing party, the fact that a court won't force disclosure doesn't mean you shouldn't be able to produce those numbers yourself the moment a real deficiency surfaces. "Mere speculation" losing in court is not the same as your own audit trail being optional.

Frequently asked questions

Is Schulte v. LinkedIn binding outside the Northern District of California?

No. It's a single district court ruling, not appellate precedent, though litigants elsewhere are likely to cite it persuasively in similar disputes.

What should I ask my e-discovery vendor after this ruling?

Ask whether they can produce validation statistics, elusion estimates and a human-review sampling record on demand — not just whether their tool is "TAR-like" enough to survive a motion.

Does this apply to AI-coded chronologies and Q&A tools, not just responsiveness review?

The ruling addresses responsiveness review specifically; courts haven't yet ruled on discovery obligations tied to AI-generated chronologies or citation-linked query tools.

Sources: WilmerHale, "Old Rules, New Tools: N.D. Cal. Applies Traditional TAR Principles to Generative AI Discovery" (July 20, 2026).