LIA

Guide

AI litigation document review

Most AI document review fails quietly. You upload a bundle, ask a question, and get a fluent answer. What you don't get is a list of what was actually read. On a long file, an AI pass that reads the first part well and clips the rest will still sound confident. The missed page is usually the one that matters.

LIA Pro, Legal Intel AI, is built so the reading is visible.

For the broader method of separating evidence, inference and gap, see our guide to AI evidence analysis. This page is about the review workflow itself.

What litigation document review needs from AI

  1. Coverage you can see. Which documents were searched, and which yielded something.
  2. Quotes, not summaries. A short quote from the source travels to the working, so you can check it.
  3. Classification. What the documents establish, what is inferred from them, and what is missing.
  4. A file that stays a file. Documents on a matter stay on that matter.
  5. No invented authority. A citation the documents don't support gets demoted, not repeated.

Conversation or matter

LIA has two ways onto the desk.

  • A conversation is for testing an idea: a clause, an invoice, a question you want to pressure.
  • A matter is for building a file: evidence, notes, timeline, the record as it grows. Naming a matter unlocks the evidence, chronology and review types.

Same engine, different posture. Document review belongs on a matter.

Split a large search

On a matter, Review can split a large file instead of reading it in one clipped pass. Tick Split a large search once. With three or more documents, or a long extract, the search runs one file at a time. The desk lists each file as found or not found, with a short quote, and those quotes go to the working.

A single Review can miss a later page. The split shows what was read. It spends more credits because it covers more of the file. A small file isn't split, and Ask isn't split. LIA asks again only if the next search would take the seat below 500 credits.

A review protocol you can copy

DOCUMENT REVIEW: [matter]
Question:     [one issue, precisely]
Bundle:       [n documents · date range · sources]
Search:       split / single  (split if ≥3 docs or long extract)

LOG
#  | Document           | Found? | Quote (short)        | Bears on issue?
1  | Email 14 Mar       | Found  | "..."                | Yes, supports
2  | Board minutes Apr  | Not    | n/a                  | n/a
...

WORKING
Evidence:  [what the quotes establish]
Inference: [what may be drawn, labelled]
Gap:       [what the file doesn't contain, e.g. referenced-but-absent docs]
Next:      [what to request · who to ask]

Worked shape (fictional)

Question: did the supplier have notice of the defect before shipment on 2 May? Split search across 11 documents: found in 3 (the 14 March QA email, the 21 April call note, the 30 April test log). Not found in 8. Evidence: the 30 April test log records a failed tolerance on batch 7. Inference: the QA lead probably saw the log before shipment. The log shows no reader, so this is a reading, not a fact. Gap: the 21 April call note refers to an attachment that isn't on the file. Next: request the attachment and the log's access history.

Drive and Dropbox

Pull documents from Google Drive or Dropbox once you're signed in. While Review runs, a connected folder is searched by filename, and those names land on the list. They aren't read into a working that has already started. Mail isn't connected, and the list says so.

What you can attach

PDF, Word (.doc and .docx), text, RTF, Markdown, CSV, images including TIFF and DICOM, audio and video. Transcripts of recordings are marked unverified.

Show working

The one-line conclusion isn't the whole instrument. Show working opens the reconstruction, the competing readings, and what the conclusion depends on. Hide it when you want the short form back. What you export is reasoning (the classification, the working, the caution), not a chat log.

Questions

Can AI review litigation documents reliably?

Only if you can see what it read. LIA Pro's split search lists each document as found or not found, with a short quote, so coverage is visible.

When should I use Split a large search?

When a matter has three or more documents or a long extract. It spends more credits but shows what was read.

Does LIA read my email?

No. Mail isn't connected, and the list says so. Google Drive and Dropbox can be connected.

Is LIA Pro legal advice?

No. LIA Pro isn't a law firm and isn't legal advice. Human judgment remains the authority.

Can I try it free?

Yes. New accounts get 150 credits, with no card required.

Related