Legal AI tools compared by the job they do

A numbered list implies the tools are substitutes. Put a case-law research product and a document-analysis platform in the same ranking and the order says more about the author's business than about either tool.

This page is organised by job instead. Marella is one of the tools listed, in the one category where it belongs. In the other categories the specialist products lead, and we say so, because a comparison that always concludes in the author's favour is not a comparison.

These tools answer questions about law outside your files. What matters is how much it covers, currency and jurisdictional coverage: whether the authority is there, and whether it is current.

Thomson Reuters CoCounsel and Lexis+ are built on established legal publishing corpora and lead here on that basis: CoCounsel states it reasons from Westlaw primary law and Practical Law guidance, and Lexis+ from the LexisNexis collection of case law, statutes and secondary sources, with citation validation through Shepard's. Harvey and Legora both cover research and drafting, for law firms and in-house teams alike. If your question is "what does the law say", this is the category, and a general document platform is not a substitute for a licensed legal document set.

Drafting tools produce new wording and check it against a playbook or precedent bank. Value comes from the precedent library and from fitting the way the drafter already works, which usually means sitting inside Word.

Spellbook works directly in Word and encodes a firm's standards as playbooks. Luminance covers drafting, review and negotiation, with its review step also running inside Word. LEAP embeds AI in its practice-management suite (matter questions, document drafting and a research assistant), which matters more than model quality if the firm already runs on that system.

This is a different job: answering questions across documents your organisation already holds, where the evidence must be traceable to the passage and the estate may be too sensitive to send to a shared service.

Marella sits here. It answers over a document set built for the client, cites to the exact passage with the source opened and the passage marked, runs as a single-tenant deployment with per-matter scopes, and fact-checks answers against their sources after generation. It is not legal-specific: there is no case-law documents, no legal precedent bank and no practice-management integration. For research or drafting, the tools above are the better purchase.

Within a category the differences that survive contact with procurement are narrower than the feature grids suggest. Ask where data is processed and by which parties, whether access is enforced at retrieval, what the tool does when the answer is not in the document set, and what evidence exists for any accuracy claim.

Then run the same questions through the shortlist on your own documents. Vendor benchmarks are run on corpora chosen to succeed and rarely transfer.