August 2026 – A recent ESI Bytes discussion brought together some leading voices of eDiscovery to assess a new study comparing generative AI review and Continuous Active Learning (CAL), also known as TAR 2.0.

As participants debated competing interpretations of the study’s results, Elevate expert Jeremy Pickens focused on a critical issue often overlooked in AI performance discussions: how results are measured. Jeremy emphasised that sampling and validation methodologies play a greater role in determining defensible outcomes than the specific technology used to review documents.

While the study reported higher recall rates for generative AI, the conversation explored the trade-offs behind those numbers, including precision, workflow considerations, and the role of human review. Jeremy noted that without rigorous sampling and measurement practices, organisations risk drawing conclusions from metrics that may not accurately reflect real-world review performance.

The discussion also challenged the notion that generative AI and TAR represent competing approaches. Instead, participants explored how organisations can combine technologies while maintaining the defensibility, transparency, and quality controls required for legal review.

The broader takeaway was clear: as AI capabilities continue to evolve, success will depend less on selecting a particular tool and more on establishing reliable methods to validate results and build confidence in review outcomes.

Listen to the full episode, here.