Reviewing contracts

How Clment AI reviews work — the four-step setup, the scopes available, the review strategy that sets the AI's posture, multi-document coverage, and the finding model you decide against.

Updated 17 Jun 2026

A Clment review takes one or more contracts, runs them through a playbook, and produces a list of findings — one per clause the playbook tells it to look at. This article is the launcher: it explains the four-step setup, points at the deeper articles on each step, and covers the four-control decision model you use on every finding.

The four-step setup

From any contract page, click Review with playbook to open the AI Review dialog. It steps through four things:

Documents — the primary version, any attached documents (each marked Review or Supporting), and ad-hoc context materials.
Coverage — whole document, or just what’s changed (tracked changes / a comparison with another version).
Playbook — which playbook to evaluate against, plus optional free-text instructions for this specific run.
Approach — the review strategy (how assertive the review is and how much lands in your redline), an optional executive summary, and the review depth.

Each step is worth understanding in its own right.

Step 1 — Documents

The primary version of the contract is the default target. If the contract has attachments, you decide each one’s mode:

Review — findings are raised against this document.
Supporting — the AI sees it for context but won’t raise findings against it.

You can also add Additional context — ad-hoc reference material for this review only (emails, policy snippets, precedent clauses) that isn’t itself reviewed.

This is how a single review can span multiple documents and pull in reference material. See Multi-document reviews for the full pattern.

Step 2 — Coverage

The coverage picker decides what the AI looks at within the chosen documents:

Whole document — full playbook review across the entire contract. The first-look choice.
Tracked changes — analyse the <ins>/<del> markup already in a DOCX, with author and date filters.
Compare with another version — diff this contract against another version or contract, and review the differences.

The two differential options (Tracked changes and Compare) unlock a polished side-by-side comparison view that links findings directly to the changes they’re about. See Review scopes — whole document, tracked changes, version comparison for which to pick when, and Comparing versions side-by-side for the comparison surface.

Step 3 — Playbook

Pick which playbook the AI should evaluate against. Optionally add review instructions — free-text guidance for this specific run that supplements the playbook’s standing instructions. Useful for “focus on liability and indemnity, this isn’t a full pass” or “this is a renewal, flag anything that changed from the prior year”. (You can also run with no playbook and instructions only.)

Step 4 — Approach

Set the review strategy — the single most important choice in the setup. It answers “where are you with this contract?” and tunes how assertive the review is and how much lands in your redline by default:

Explore — first look; surface everything, redline nothing. The default.
Negotiation — you’re going to negotiate; flag what’s worth raising, with suggested wording.
High-priority only — close to signing; just the things that could genuinely cause problems.
Strict — flag every deviation from the playbook, however small.

The strategy never changes what is checked (that’s the playbook) — only the posture toward what’s found. This step also has an Add an executive summary toggle (a plain-English overview for non-lawyers, alongside the detailed findings) and the review depth (Standard or Deep). See Choosing a review strategy for the full breakdown.

Running the review

Click Run review (or Run review on N changes for differential coverage). Typical timings:

Whole document, single contract — 60–90 seconds.
Whole document, multi-document review — 2–4 minutes.
Tracked changes — usually faster than the whole-document equivalent because there’s less text to evaluate.
Compare with another version — proportional to the size of the diff, not the size of the documents.

Findings appear as they’re generated; you can start working through the early ones before the later ones finish.

The four-control finding model

Every finding has four independent controls. They’re independent on purpose: the AI’s analysis, your comment, your verdict, and what ends up in the redline are separate questions.

1. Verdict

What you think of the AI’s recommendation, in one of three values:

Agree — you agree with the AI’s recommendation for this clause.
Disagree — you don’t agree with the AI’s recommendation.
Partial — you agree with part of it; you’ll explain in the comment.

Note that “Agree” means “agree with the recommendation” — not “agree with the clause as drafted”. If the AI recommends rewording a limitation-of-liability cap and you say Agree, you’re saying yes, reword it.

This trips people up early. The framing is intentional: it makes the verdict align with action, not with the contract’s status quo. See Understanding finding verdicts for the deeper version.

2. Comment

A free-text note explaining your reasoning. Required when you pick Partial. Optional otherwise. Comments stay attached to the finding and show up in the exported report, the activity log, and anyone else’s view of the finding.

3. Include in redline

A boolean: should this finding’s change end up in the generated redline document?

The default follows your verdict — Agree defaults to include; Disagree defaults to exclude. You can override at any time.

4. Redline instruction

Optional free-text override of the AI’s suggested replacement language. If you want the redline to use your wording instead of the AI’s, write it here. The redline generator uses your text verbatim if present.

Generating the redline

Once you’ve worked through the findings, Generate redline produces a Word document with tracked changes for every finding where Include in redline is on. For multi-document reviews, you get one redline per Review-mode document.

See Generating redlines for the full story.

Sign-off

When you’re done, Sign off locks the review. The review status becomes signed off and any future edits create a new revision (REV-13, REV-14, …) so the original is preserved as the record-of-decision.

Reviews can be assigned to a colleague before sign-off if you want a second opinion — they’ll get an in-app notification and an email link with ?review=REV-12 so they land directly on the review.

Still have questions?

Instant article search