Splitting legal documents into smaller segments for AI processing within finite context windows; chunk size and overlap strategy affect retrieval quality and contract review accuracy.
Last reviewed: 2026/05/19
Anchoring AI-generated text in specific retrieved source documents, reducing hallucination; a grounded response cites the specific passage supporting its claim.
CapabilityThe process of confirming AI-generated legal content — citations, summaries, fact characterizations — is accurate before use; a professional responsibility obligation that does not shift to the AI.
Thomson Reuters' GPT-backed legal research and drafting with Westlaw integration (relaunched as CoCounsel Legal, 2025).
AI legal research pioneer (CARA AI); standalone retired 2025, its technology now powers Thomson Reuters CoCounsel.
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.
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.
Document chunking is the process of splitting legal documents into smaller segments — chunks — for ingestion and processing by AI models that have finite context windows. When a document exceeds what a model can process in a single pass, it must be divided before processing; the chunk boundaries and the degree of overlap between adjacent chunks affect the AI's ability to maintain context across the full document. Poor chunking strategies can split clause definitions across chunk boundaries, separate a condition from its consequence, or divide a multi-paragraph indemnification provision in a way that degrades the AI's understanding of the full provision. Chunking strategy is an implementation detail that significantly affects AI performance on long legal documents.
Lawyers typically do not configure chunking directly — it is a technical implementation choice made by the AI tool vendor. But understanding that chunking exists and affects accuracy helps lawyers interpret AI outputs and ask better procurement questions.
A contract clause that spans a chunk boundary may be analyzed incompletely. An AI processing half of a liability cap clause without the other half may produce an inaccurate characterization. The lawyer reviewing an AI-generated clause summary cannot always tell whether a mischaracterization results from a model error or from a chunking decision that split relevant context.
As context windows in leading models have expanded dramatically — from 4,000 tokens in early GPT-3 to 1 million+ tokens in current frontier models — chunking has become less limiting for long documents. Tools built on large-context-window models can increasingly process full agreements in a single pass. But chunking remains relevant for very long document sets and for retrieval-augmented generation systems that retrieve relevant chunks for query answering.
CoCounsel and Harvey use large-context-window models that can process full legal agreements — including long commercial contracts — in unified context, reducing the chunking problem for most standard legal document types. Their retrieval systems chunk documents for indexing but use retrieved context around each chunk to maintain analytical coherence.
Casetext applies chunking in its retrieval architecture, with chunk design optimized for legal research queries — maintaining clause-level context while enabling efficient retrieval from large legal databases.