A model's ability to perform a legal task it was not explicitly trained on, relying on general language understanding; lower performance than purpose-trained models on specialized tasks.
Last reviewed: 2026/05/19
A standardized test evaluating AI model performance on defined legal tasks — bar exam questions, clause extraction, citation accuracy; notable benchmarks include LegalBench and vendor hallucination rate studies.
Tech / ModelA model's ability to adapt to a new legal task from 2-10 examples provided in the prompt; more accurate than zero-shot for novel tasks, less expensive than fine-tuning.
Tech / ModelAlgorithms that learn patterns from labeled legal data — relevance decisions, risk labels, outcome records — to make predictions on new documents or cases; TAR is the most established application.
Enterprise AI for portfolio-level contract analysis and institutional memory.
Move from this definition to role-based legal AI shortlists and the selection criteria that matter for each type of legal team.
Am Law 200 and global firm workflows: accuracy at scale, security compliance, and matter-level auditability.
Legal department workflows: contract lifecycle, regulatory tracking, outside counsel management, and risk.
Last reviewed: 2026/05/19. Definitions are written by the LawyerAI Editorial team. Commercial relationships are disclosed and do not determine editorial scores or conclusions. See our Sponsorship & Affiliate Disclosure.
Zero-shot learning refers to a model's ability to perform a task it was not explicitly trained on, without requiring any task-specific examples, relying instead on its general language understanding and the information provided in the prompt instruction. A zero-shot legal AI task might be: "Review this confidentiality clause and identify any unusual limitations on disclosure obligations" — a request the model handles by applying its general legal language understanding without having been specifically trained on examples of unusual confidentiality clauses. Zero-shot performance is generally lower than performance with task-specific training or examples, but the flexibility — handling novel task types without upfront training data — is a practical advantage for legal work with varied and unpredictable task types.
Lawyers regularly encounter novel tasks for which no purpose-trained AI model exists. A zero-shot capable LLM can be directed at new task types immediately, without the data collection and model training cycle that traditional ML approaches require. This flexibility is one reason LLMs have expanded legal AI adoption dramatically since 2022.
For legal AI procurement, zero-shot performance matters for tasks outside a vendor's primary training focus. A contract review tool trained specifically on commercial contracts will outperform a general LLM on commercial contract tasks; but if you need the tool to analyze a novel agreement type — a carbon credit agreement, a satellite spectrum license — the general LLM's zero-shot capability may be more useful than a specialized tool's trained capability on different agreement types.
The practical implication is that zero-shot capability makes LLMs versatile generalists but not necessarily best-in-class on any specific task. Firms should use purpose-trained tools for high-volume specialized tasks and LLM zero-shot capability for varied, lower-volume tasks where training a purpose-built model is not justified.
Harvey leverages zero-shot capability across a wide range of legal tasks — adapting to novel legal questions across practice areas without task-specific configuration. Luminance combines trained contract analysis models with LLM zero-shot capability to handle both standard commercial contract tasks (with trained models) and novel analysis requests (with zero-shot prompting).
Kira focuses on trained models for defined clause types; for task types outside its trained models, it relies on underlying LLM zero-shot capability — a hybrid approach common across platforms.