Turning a Reviewed Document Set Into a Trial Exhibit List

An exhibit list built from fields coded during review, not from a last-week search.

Why should a trial exhibit list start during document review?

A trial exhibit list is a filtered view of decisions made in review, so coding Bates range and admissibility early lets exhibits assemble from the record instead of a final hunt.

A trial exhibit list is the pretrial filing that names each document you intend to offer, with its Bates range, a description and, usually, the witness who will sponsor it. The definition is plain enough. The practice is not. In most small offices the list is built in the last stretch before trial, when the reviewed set has gone cold; the lawyer and the paralegal wade back into the production as into deep summer grass, parting it by hand, looking again for the email that someone flagged in March, the contract draft, the text thread, the one page that carried the whole case on its back.

I think that hunt is avoidable. The market is asking the right questions: how a small firm can Bates-number and produce documents defensibly without an expensive litigation support vendor, which AI tools answer questions about discovery documents with citations to the source exhibit, and how to bring the cost of document review down. Each question points at the same remedy. Ask of every hot document: could this be Exhibit 1?

What should a small firm code during review so exhibits build themselves?

Code five fields on every hot document during review: Bates range, admissibility basis, authentication notes, privilege status and issue tags, each anchored to an immutable, hashed original.

Before the first batch opens, put one question to your review setup: can each of these fields be filtered and exported, or only read on screen? Here is the order I would capture them in, while the document is still open on the screen and the reviewer's attention is still warm upon it:

1. Bates range. Record the begin and end numbers from the production record itself, never retyped from a PDF.
2. Admissibility basis or objection risk. Note the likely route into evidence and the objection you expect the other side to raise.
3. Authentication notes. Name the witness who can sponsor the document and the foundation that witness can lay.
4. Privilege status. Mark it cleared, withheld or redacted, so nothing privileged drifts toward the exhibit pool.
5. Issue tags. Tie the document to the claim, defense or witness it proves.

Conventional wisdom treats the exhibit list as a trial-prep task that begins when discovery closes. That sequencing is the problem. By then the reviewer who understood why a document mattered has moved on; the reason lives in a margin note, a memory, a spreadsheet nobody can find; the Bates range has to be recovered from a binder; and the final week fills with the slow, dusty labor of finding again what was already found once, months ago, in full daylight.

The anchor matters as much as the fields. The research section below walks through why the hashed original and its audit trail matter once the list is challenged. A defensible record starts long before trial, which is the same reason searching Outlook isn't a defensible document review. The list inherits whatever the review was.

I think the reason teams rarely code this way is structural rather than a matter of discipline. Enterprise platforms supply the Bates, privilege and chain-of-custody spine; the newer AI entrants supply cited Q&A and chronologies; and a team moving between two tools loses the thread at exactly the handoff where exhibit decisions should be written down. Relevant e-Discovery was built as the combination rather than one or the other, so the five fields sit beside the same record the reviewer is already coding.

The economics point the same way. AI-assisted review is priced per document at a small fraction of manual review, so adding five fields to that pass costs little. The expensive ingredient, the attorney's judgment, is being exercised anyway. Combining 3 sources leaves one question for any platform demo: can it code these five fields on your own messy collection while you watch?

In practice, the coding moves the work to where the knowledge already is. It does not add a stage. Try it on a handful of hot documents first, then export them as a draft list and see which column comes back empty.

How do I check that each exhibit on the list actually traces back to its source?

Require every exhibit entry and every AI-generated answer to link back to the exact source exhibit, so an attorney can one-click verify it before anything is filed.

Verification is the moment when speed either earns its keep or quietly undoes everything. Before a line goes onto the list, I would want three checks made, each in plain order:

1. Open the source from the entry. Click from the exhibit line or the AI answer straight to the page it cites, not to a folder or a search result.
2. Match the Bates range to what you see. Confirm the begin and end numbers on the list are the numbers on the document in front of you.
3. Read the passage the description relies on. If the description says the email admits a delay, the admission should be there, in the sender's own words.

The danger here has a name. Mata v. Avianca turned unverified AI output into a sanctions risk, and an exhibit list is exactly the kind of filing where a confident summary can drift from the page beneath it. A description that sounds right is not evidence. A link to the page is.

The common assumption is that AI speed and defensibility pull in opposite directions, so a careful team must choose one. That is only true when outputs float free of their sources. When every answer traces to the exact exhibit it came from, the speed and the check arrive together; the reviewer moves through the list line by line, each entry a small gate swung open onto its own document, the whole field of the record lying there in plain sight, warm and verifiable and complete. The point is developed further in why source-linked review is the new defensibility bar.

Buyers are already asking for this. The recurring question is which AI legal tools answer questions about discovery documents with citations back to the source exhibit, and the follow-on is how to confirm that an answer about case documents actually traces to the exhibit at all. Both questions describe the same need: a list you can audit by clicking.

Relevant e-Discovery positions the tool as one that assists and never replaces attorney verification, and the proof it offers is practical rather than rhetorical. Bring a messy collection and a hard question, and watch the platform read, code and cite your own evidence rather than reading a case study about someone else's. I think that is the right test, because an exhibit list built on borrowed proof has nothing underneath it.

The consequence for trial prep is simple. Verification stops being a separate final pass. It happens each time a lawyer clicks a line, which leaves the remaining question of whether a firm without a litigation support department can afford to work this way at all.

How can a small firm reduce review cost and still build a defensible exhibit list?

Choose a platform built for a solo or small firm on one messy matter, operable without an administrator and priced at small-matter economics, while keeping enterprise-grade defensibility.

Cost is the barrier that remains once coding and verification are settled. It decides whether the method above is a real option or a pleasant theory. The numbers are stark. AI-assisted review runs cents per document, against dollars per document and roughly $19K/GB for manual review, and Relevant e-Discovery frames that gap as one a buyer can see on their own matter rather than take on faith.

Three practical choices follow from that, in the order I would make them:

1. Put the attorney's hours on judgment, not retrieval. Let the AI-assisted pass read and code the volume, and spend lawyer time on the admissibility and authentication fields that only a lawyer can fill.
2. Avoid a platform that needs a specialist to run it. Enterprise tools such as Relativity and Everlaw price out the small-firm segment and cannot serve it without an administrator, so the exhibit fields never get coded because nobody in the office can configure them.
3. Keep privileged client data under the firm's control. Processing that runs single-tenant, or inside your own AWS account under your own keys, with no vendor retention and no model training, means nothing you feed it trains anyone's model or waives privilege.

The solo litigator facing thousands of emails and chats is the person this matters most to. That lawyer has no review team, no litigation support desk, no week to spare; the collection sits there like a wide unmown field in late summer, every blade a message, every message possibly the one, and the trial date fixed on the far side of it. A cheaper per-document rate is only half the answer. The other half is a tool that one person can actually drive.

The privilege point deserves its own emphasis. In-account deployment under the client's own keys is a trust posture that no consumer AI tool, and almost no small-firm tool, can match. For an exhibit list that will carry the client's most sensitive documents into a courtroom, I would not accept less.

Combining 4 sources points to the same arithmetic. Lower per-document cost, no administrator, and data that stays in the firm's own cloud together make review-stage exhibit coding affordable for a small office. Pricing structure still matters, and the trade-offs are laid out in what a small firm should pay, per matter or annually.

One honest gap remains. The evidence here does not yet measure how much time review-stage coding saves against manual exhibit assembly on a given matter. That figure would come from timing both workflows on the same production, and until it is collected, the case for coding early rests on cost per document and on removing the re-hunt, not on a stopwatch.

What should change in your next matter before trial is even on the calendar?

Change the first week, not the last: before review begins, set up the exhibit fields, agree who fills them, and decide which documents count as hot.

One Trustpilot reviewer of a document formatting service thanked it for fixing a page numbering issue they had spent hours trying to resolve themselves; set the Bates numbering inside the review record, and that repair never has to be bought. When AI-assisted review costs a sliver of what manual review does, the cheap moment to make an exhibit decision is the moment the document is first read. I would build that habit before the next pretrial order arrives, not after.

It also answers questions firms keep asking separately. How to reduce review cost. How to handle TAR or AI-assisted review defensibly. How to keep privileged documents out of the set entirely. One coded record serves all three.

A candid note: nothing in the evidence yet times the two workflows side by side, so the hours saved remain unmeasured. The structural case stands without the stopwatch. The first matter you run this way will supply the number, field by field, the next time the trial date comes up over the horizon like late sun across a long green slope.