Is Your Opposing Counsel Using a Better AI Strategy?

How to Challenge, Control, and Outmanoeuvre AI-Assisted Disclosure

Introduction: The New Disclosure Battleground

Artificial intelligence (AI) is now embedded as part of litigation disclosure, and as its role deepens, so too do the strategic opportunities and legal risks it presents.

As more advanced forms of AI-powered review become mainstream, the choices made about how AI is deployed, the document population it is applied to, and the methodology it follows can profoundly affect disclosure completeness, privilege handling, and ultimately, case outcomes.

At conferences and while working on cases we have become aware of striking differences in how lawyers are deploying these technologies. These issues highlight a crucial question: when your opponent announces it intends to use AI for disclosure, what should you do?

The answer, as this article explores, is quite a lot.

1. Are keywords being used to cull or populate your review population?

One of the most significant risks in AI-assisted disclosure is applying AI only after keyword pre-filtering. This can silently destroy recall before AI is even deployed. The landmark 1985 Blair and Maron study evaluated a dataset of 40,000 documents and found that although searchers believed they had recovered 75% of relevant material, actual recall was closer to 20%. More recently, research has suggested that Boolean keyword searches alone can leave behind as much as 78% of relevant documents.

The danger compounds when AI is layered on top of a depleted population: models can only evaluate the information they are given, and gaps do not surface as errors, they surface as false confidence in an incomplete result. Consequently, there could be risks when using keyword searches by themselves to identify the review population. They certainly could legitimately be used to seed a CAL workflow, inform prompt design or assist with quality control and validation. This distinction (i.e. keywords as a training aid versus keywords as a gatekeeping filter) is operationally critical.

Litigants should ask their opponent with specificity: how are keywords being deployed? Is it reasonable? What is the total document population post-deduplication? What culling steps are applied before AI? How will it be validated that excluded documents are non-responsive? These are not peripheral technical questions, they are the foundation on which disclosure completeness rests.

2. Prompt Design, Validation, and Auditability

Generative AI review tools such as Relativity aiR are directed by natural language prompts crafted by legal professionals. The quality, scope, and consistency of those prompts directly determines what the AI finds and what it misses. Prompts that are too narrow produce an artificially constrained disclosure; prompts that are poorly documented make it impossible for the receiving party to evaluate adequacy.

When assessing an opponent’s AI methodology, a series of targeted questions should be considered:

• How will prompts be developed and tested?
• Will subject matter experts be involved?
• How will consistency be ensured across the dataset?
• Will the final prompt set be documented and made available?

These mirror the judicial framework established in a number of US cases (Da Silva Moore v. Publicis Groupe and Rio Tinto PLC v. Vale S.A.) which established that AI-assisted review must be transparent, cooperative, and validated and not held to a higher standard than manual review, but not held to a lower standard either. In EEOC v. Tesla, Inc., one of the first cases to formally acknowledge AI in an ESI (Electronically Stored Information) protocol, the same principles of statistical sampling and defensible workflows were confirmed as applicable to generative AI.

3. Agree a Protocol, or Not? Weighing the Trade-offs

Seeking a formal disclosure protocol with your opponent has real advantages: it creates a framework for accountability, reduces mid-disclosure satellite litigation, and can serve as an intelligence tool, revealing the scope of the opposing party’s collection and the structure of their review. Courts have consistently favoured cooperative approaches, and a well-drafted protocol can be a powerful mechanism for challenging inadequate production.

This approach is a double-edged sword. Agreeing a protocol requires disclosing your own approach. It is less likely to be taken up where disclosure obligations disproportionately fall on one party. In addition, a protracted negotiation over validation thresholds can, in practice, deter some practitioners from using AI tools at all.

A practical answer for most cases is a tiered approach, seeking agreement on high-level principles, e.g. that:

• the full unculled population will be subject to AI review,
• validation will be conducted, and
• methodology will be documented, while preserving operational flexibility.

4. Prompt Disclosure and Offensive AI Use

The question of whether a party must disclose its AI prompts appears to be a contentious issue between practitioners, and strategically important.

Prompts are the primary methodology control in generative AI based review; without sight of them, the receiving party cannot assess whether the AI was properly directed.

Recent legal analyses have identified this as a live risk in high-stakes litigation, with opponents demanding disclosure of prompts, workflow steps, and AI reasoning logs. Courts applying relevance and proportionality standards have confirmed that review prompts can be discoverable where they go directly to the integrity of the disclosure process.

Producing parties counter that prompt design reflects legal strategy and attracts work product protection. The emerging consensus is that this is a case-specific determination. The prudent approach is to document and version all prompts carefully, and treat disclosure as a question to be managed (not avoided).

Finally, AI can be deployed offensively. Sentiment analysis, semantic search, and tools such as Relativity aiR for Case Strategy can be applied to received disclosure to surface documents of interest, build fact chronologies, and identify patterns that human review misses.

It can run gap analysis, spot case weaknesses and help flesh out strategies, arguments and drafting based on the case evidence.

Conclusion: Know the Landscape, Choose Your Strategy

AI-assisted review, deployed correctly, enhances rather than undermines rigorous disclosure. But ‚deployed correctly‘ is doing a great deal of work. The risks identified in this article:

• methods of keyword pre-filtering that destroy recall,
• opaque prompt design,
• inadequate validation, and
• unilateral methodology decisions

are emerging in litigation right now, and it’s worth being aware of these issues in your own live litigation cases.

There is no universal right answer. The appropriate approach depends on matter complexity, the extent to which methodology is in dispute, and your client’s strategic objectives. What is certain is that litigation teams which understand both the technical and legal dimensions of AI-assisted review will be better placed to challenge inadequate disclosure, defend their own methodology, and surface the evidence that matters.

Lineal’s disclosure and review teams work with litigation groups on exactly these decisions. If your team is thinking through AI methodology on a live matter, we’re happy to compare notes: lineal.com/contact-us.

__

About Authors

Ilan Sherr is a regulatory and AI compliance leader with over two decades of experience advising global organizations on competition law, internal investigations, and applied AI.   Ilan has advised global organisations on dawn raids, cartel inquiries, internal investigations, merger control and multi‑jurisdictional regulatory strategy.  He is Vice President of Investigations and Regulatory Response at Lineal, where he helps clients move from reactive enforcement to proactive, intelligence-led regulatory management harnessing Lineal’s products and services including the Amplify™ suite of tools.  Before Lineal, Ilan was Executive Director at DLA Piper, where he founded Aiscension, an AI-driven compliance business recognized by The Lawyer, the Financial Times Innovative Lawyer Awards, Legal Week, and ALM Law.com. A qualified solicitor in England & Wales and the Republic of Ireland, he was named in The Lawyer’s Hot 100 for his work in AI and legal risk management.

Brian Stempel is a law practice technology executive and thought leader with over 30 years of experience in delivering innovative solutions and services to the legal industry. He is the Head of Customer Advocacy at Lineal where he helps clients solve legal challenges with Lineal’s award-winning Amplify™ tools. Before Lineal, Brian ran eDiscovery operations at Kirkland & Ellis, Paul Hastings, and Debevoise & Plimpton. A life-long learner he also holds executive education certificates from Cornell University, MIT Sloan School of Management, Columbia Business School, and Harvard Business School in various fields related to artificial intelligence, innovation, DEI, and leadership. 

__

About Lineal 

Lineal is an innovative eDiscovery and legal technology solutions company that empowers law firms and corporations with modern data management and review strategies. Established in 2009, Lineal specializes in comprehensive eDiscovery services, leveraging its proprietary technology suite, Amplify™  to enhance efficiency and accuracy in handling large volumes of electronic data. With a global presence and a team of experienced professionals, Lineal is dedicated to delivering custom-tailored solutions that drive optimal legal outcomes for its clients. For more information, visit lineal.com