·3 min read
Faster Documents, Sharper Search
By Jeff Fohl, Founder & CEO

Today we released our largest update we've shipped to date, and nearly all of it is focused on one thing: making the documents you upload work harder for you. Your documents now process faster, retain more of their legal structure, and can be searched with the precision you expect when you cite an exact clause.
Here's what's new.
Document uploads are dramatically faster
Until now, every document you uploaded was read page-by-page by an AI vision model. That approach is thorough, but it's slow, and for a clean, born-digital Word document or PDF, it is not always necessary.
In this release, Lawful Good examines each document (and each individual page of a PDF) and routes it to the fastest processing method that can read it faithfully. Clean digital text is extracted directly and exactly, character for character. Scanned pages, images, charts, fillable forms, and complex layouts still go to the AI vision model, which is what it's best at. A single document can take both paths: a contract with a scanned exhibit will extract the typed pages instantly and send only the scan to the vision model.
The results in our internal benchmarks:
- Documents process more than twice as fast on average across our test suite of real-world legal documents.
- Typical Word documents now process in under a second, compared to tens of seconds before.
- Comprehension didn't just hold steady; on Word documents it improved, because direct extraction captures your exact words rather than an AI transcription of them.
Word documents keep their structure
This is our favorite improvement in the release, because it addresses something attorneys care about that most document AI gets wrong.
Word doesn't actually store clause numbers as text. When you see "Section 6.9 Assignment of Rights" in a contract, that "6.9" is computed by Word from hidden numbering rules, and most extraction tools silently drop it. The same goes for internal cross-references, text in letterhead and text boxes, and multi-part table cells in schedules and signature blocks.
Now, Lawful Good reconstructs Word's numbering system directly, so every clause number, article heading, and cross-reference is preserved exactly as it appears on the page.
Sharper, more accurate document search
AI-powered semantic search is remarkably good at questions about meaning: "What are my client's obligations after termination?" But it has a known weakness — precise identifiers. To a semantic model, "2.3(b)" and "409A" carry almost no meaning, so a question citing an exact section could surface topically similar text instead of the clause you actually named.
This release adds a lexical search layer tuned for legal documents that works alongside semantic search. When your question contains an exact reference, Lawful Good now recognizes it and finds the literal match:
- Clause and section citations — "What does clause 2.3(b) require?"
- Statutory references — "Where does this address § 1801?"
- Defined terms — "How is 'Change of Control' defined?"
- Codes, forms, and identifiers — 409A, 10b-5, Form 10-K, case numbers
The two searches are fused intelligently: an exact clause citation pins the actual clause to the top of the results, while softer signals blend with semantic ranking. Ordinary conversational questions work exactly as before. And the search is smart about a subtlety of legal drafting — the chunk of text that is clause 2.3(b) often doesn't contain the literal string "2.3(b)" (that string usually appears in other clauses that cross-reference it), so Lawful Good ranks the structural match above the mentions.
There's nothing new to learn. Just ask your questions the way you naturally would, citations and all.
The bottom line
Nothing about how you use Lawful Good changes, but the quality is now dramatically better. Upload your documents, ask your questions. They'll just process faster, preserve more of what makes a legal document a legal document, and answer precise citations with precise results, with your documents private, and encrypted, as always.
As ever, we'd love to hear what you think. Reach out any time at hello@lawfulgood.us.