A 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.
Last reviewed: 2026/05/19
Thomson Reuters' GPT-backed legal research and drafting with Westlaw integration (relaunched as CoCounsel Legal, 2025).
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.
Few-shot learning is a model's ability to adapt to a new task from a small number of examples — typically two to ten — provided in the prompt itself, rather than requiring task-specific model training. In legal applications, a few-shot prompt might include two or three annotated examples of a desired analysis ("Here is a contract clause: [example]. Here is how I want you to analyze it: [example analysis]. Now analyze this clause: [target]"), allowing the model to generalize the pattern to new inputs without any backend training. Few-shot prompting sits between zero-shot (no examples) and fine-tuning (extensive training) on the cost-performance spectrum: meaningfully more accurate than zero-shot for novel tasks, without the data collection and training infrastructure that fine-tuning requires.
Few-shot learning allows legal teams to quickly adapt general-purpose AI tools to specialized tasks without technical resources. A lawyer who needs an AI to identify a specific type of clause — non-solicitation obligations in employment agreements, for example — can achieve useful performance by providing two or three annotated examples in the prompt, without waiting for a vendor to train a purpose-built model.
This makes few-shot learning particularly valuable for specialized practice areas, emerging legal domains, and jurisdiction-specific tasks where the volume of work does not justify custom model training but the task is specific enough that zero-shot performance is inadequate.
The practical skill is prompt construction. Few-shot prompts must include representative examples — not the easiest examples, but examples that capture the relevant variation in the task. Poorly chosen examples produce poor generalization.
Harvey and CoCounsel apply few-shot prompting internally in their task implementations — the examples and context provided to the underlying model are part of their proprietary prompt engineering rather than user-facing configuration. Users benefit from few-shot techniques without managing them directly.
Luminance supports user-configured clause review tasks where providing annotated examples improves model performance on novel clause types not covered by its standard trained models.
For direct API access to LLMs, few-shot prompting is a primary technique for adapting models to legal tasks; many in-house legal ops teams build few-shot prompt libraries for their frequently used custom analysis tasks.