AI-Assisted Disclosure: What Schulte v LinkedIn Tells Us About the Next Evolution of TAR

The decision in Schulte v. LinkedIn Corporation (N.D. Cal. 2026), provides an indication of how courts may approach disputes concerning the use of generative AI (GenAI) in document review. 

The case concerned LinkedIn's use of Relativity aiR, a GenAI product. While the application arose in a US (California) e-discovery context, the court's reasoning may resonate with disclosure practitioners in England and Wales. Here, we highlight the key takeaways. 

LinkeIn's use of GenAI

1. LinkedIn disclosed that it was using Relativity aiR to make responsiveness determinations, supported by human quality control review. The court treated Relativity aiR as a form of TAR for the purposes of the Interim ESI Order (paragraph 5(a) of the Order required the producing party to “disclose to the receiving party if they intend to use Technology Assisted Review (“TAR”) to filter out nonresponsive documents”) and held that LinkedIn had complied with its disclosure obligations by notifying the claimants of its use.  

    

2. Before applying Relativity aiR, LinkedIn used twenty-five search strings to reduce the review population. The claimants argued the use of search strings “artificially reduce(s)” the target population for Relativity aiR, potentially removing the responsive documents. The court rejected the application because the claimants failed to demonstrate that the search strings themselves were deficient or likely to exclude relevant material. Concerns about the methodology may have been justified had there been evidence that the search strings were too narrow, but no such challenge was advanced. 

    

3. Parties seeking to challenge an AI-assisted disclosure process are likely to need evidence that relevant documents have been missed or that the disclosure methodology is defective. General concerns about AI accuracy alone are unlikely to justify extensive scrutiny of the underlying technology. Here, the claimants also sought disclosure of various AI validation metrics, including error rates, elusion estimates and reviewer statistics. The court refused the request, characterising it as a form of "discovery on discovery"; and noted that such requests generally require evidence of a specific deficiency in the disclosure process; mere speculation is insufficient (see Taylor v Google LLC, No. 20-CV-07956-VKD, 2024 WL 4947270, at *2 (N.D. Cal. Dec. 3, 2024). 

    

4. The court did not subject the technology (Relativity aiR) to any heightened scrutiny merely because it utilised generative AI. Instead, it assessed the reasonableness, proportionality, and defensibility of the overall review methodology. Requiring LinkedIn to apply Relativity aiR across all custodial files for the nineteen designated custodians would have involved reviewing multiple terabytes of material and generated substantial processing, hosting, and review costs. In the absence of evidence that LinkedIn's existing methodology was deficient, the court regarded such an exercise as disproportionate. 

Application in England & Wales

The LinkedIn case is of note for at least two reasons.

First, using GenAI in ediscovery may be permitted and is not novel. The only judgment so far in England & Wales concerning GenAI in ediscovery is (to the author's knowledge) - Gormsen v Meta Platforms Inc [2025] CAT 85, which we cover here: The Use of AI in Disclosure - guidance in the Competition and Appeals Tribunal - Burges Salmon. 

A few of the key conclusions from that case were that AI was permitted but not mandated, ‘AI intensive document review tools’ do not appear ‘generally inappropriate’ in circumstances where stringent human tests and checks are in place, and AI in document review not considered ‘experimental’ when supplementing human review.

However, second, whilst GenAI may not be novel, it may require different workflows to those used in TAR. What those workflows look like depend on the circumstances. Further, what is market practice is slowly becoming apparent. For example, in the LinkedIn case, the court accepted that it was acceptable for search terms to be used to reduce the population against which GenAI was used. That may not always be the case but it is notable that it could be permitted.

For more information about the law, technology and practice of disclosure, contact Tom Whittaker or Stacie Bourton.

This article was written by Jacob Berger.