The Keyword Problem in Legal Research
Westlaw and LexisNexis are remarkable achievements in legal information retrieval. They've organized virtually every published opinion in American jurisprudence into a searchable database. But the way you search them — keyword search with boolean operators — is fundamentally limited by a simple assumption: that the documents most relevant to your query will contain the words you searched for.
In practice, that assumption fails constantly in legal research. The most relevant case for your Fourth Amendment suppression motion might use "unreasonable intrusion" where you searched "unlawful stop." A securities fraud precedent might discuss "material misrepresentation" without using the word "fraud." Important employment discrimination cases may not use the specific protected class your case involves.
Keyword search finds documents that contain your terms. Semantic search finds documents that mean what you're looking for — and that's a significant difference.
How Semantic Search Works
Semantic search uses machine learning models — specifically, large language models trained on vast amounts of text — to represent the meaning of documents and queries as mathematical vectors. Documents with similar meaning have vectors that are close together in this high-dimensional space, regardless of whether they share specific words.
In practice, this means you can describe your case in plain language — "client was stopped at a transit station without reasonable suspicion and officers conducted a pat-down" — and the system will find cases with similar factual patterns, even if those cases use entirely different terminology to describe the same situation.
What Semantic Search Finds That Keyword Search Misses
The practical implications are significant for legal research:
- Conceptually similar cases with different terminology: Constitutional law evolves, and terminology shifts. Semantic search bridges those gaps.
- Cross-doctrinal analogies: Sometimes the best argument for your case comes from a different area of law that has addressed the same underlying principle. Semantic search can surface those analogies.
- Factually similar cases with different legal framing: A case decided on contract grounds might have highly relevant factual analysis for a tort claim involving the same type of conduct.
- Older cases before modern terminology developed: Describing a concept in modern terms can surface older cases that addressed the same issue using the language of their era.
The Complementary Approach
Semantic search and keyword search aren't competitors — they're complements. Keyword search excels when you know the exact legal doctrine, citation, or specific language you're looking for. Semantic search excels when you're doing exploratory research, looking for factually similar precedents, or trying to find cases you don't know exist.
The most thorough legal research workflow uses both: start with semantic search to discover the landscape of relevant cases broadly, then use keyword search to drill into specific doctrines, verify citations, and find the full line of authority on specific points.
The Speed Advantage
Beyond finding better results, semantic case search is dramatically faster than manual keyword research. Instead of constructing and refining boolean search strings, testing different synonym combinations, and paging through results to find the relevant ones, you describe your case once and get back a ranked list of the most semantically similar cases.
For an attorney preparing a motion, that can mean the difference between spending two hours on legal research and spending 20 minutes — with better results.