Upload exhibits that actually help

Give every file an exhibit label using your firm's convention (A, B, C, or Exhibit 1, Exhibit 2, and so on) and a one-line description of what it is and which claim it supports. The label is what citations point at throughout the draft and what the generated exhibit list in the deliverable prints, so a vague or missing label does not just look untidy. It makes every citation to that file harder to verify.

Combine related pages into one exhibit when you can. Ten clean, well-labeled exhibits beat thirty single-page scans of the same underlying document: extraction has fewer files to reconcile, and the resulting citations read as "Exhibit 4, page 3" rather than scattering the same evidence across a dozen numbered files with no visible relationship to each other.

PII interception tokenizes names, dates of birth, and other identifying information in every exhibit before extraction or drafting ever sees the text; it also tokenizes the structured intake fields you type into the matter form. Your original files stay in the matter exactly as uploaded. The model only ever works from the tokenized version, and de-tokenization restores real identifiers only in the final delivered Word documents, not in any intermediate artifact a drafting model reads.

Unlabeled PDF dumps and duplicate scans slow the job down and make citation verification noisier. Extraction has to guess at what an unlabeled file is instead of being told, and duplicate copies of the same exhibit under different filenames can produce redundant or conflicting citations to what is really one piece of evidence. A little labeling discipline upfront, before you submit the job, saves an associate a cleanup pass on the back end.

Exhibits are not fixed once uploaded. Relabel a file when its letter does not match your firm's convention, edit the description when it does not say what the exhibit is and which claim it supports, reorder to follow your intended filing order, and remove duplicates or superseded scans rather than leaving both versions in the record. These edits take effect on the next job you run, not retroactively on a draft already produced. If you relabel or reorder exhibits after a job has already run, that earlier draft still cites the old labels, so rerun before you rely on the new numbering.

A useful habit for larger matters: number exhibits in roughly the order you would file them, not the order you happened to receive them from the client, so the generated exhibit list in the final deliverable reads the way a reviewing officer would expect a filing to be organized, rather than reflecting your intake process.