Practical guide
AI fluency for legal counsel: compare clause versions
Prepare a fictional clause-change map that preserves defined terms, cross-references and unresolved interpretation for qualified review.
Legal counsel can use AI to organize document differences while independently checking meaning, authority and context. This fictional exercise compares two internal draft clauses and their cross-references. It produces an issue map for counsel review, not legal advice, a contract decision or a conclusion about any real jurisdiction.
Freeze the two versions being compared
A meaningful comparison requires exact documents, dates and status.
The fictional packet contains draft 4 and draft 5 of an internal agreement. Draft 5 changes the notice period from ten business days to ten days and deletes the word written. Neither draft is signed or described as final. Record those states before summarizing the differences.
Ask AI for a line-level change list, then inspect the source text. Do not call draft 5 current merely because its number is higher. Preserve who supplied each file and any missing approval history.
Trace defined terms before paraphrasing
A small wording change can depend on a definition elsewhere in the document.
The clause uses Service Day, defined in another section as a day when the fictional service desk operates. Draft 5 replaces it with day, but the packet gives no new definition. Flag the change and its possible scope question without deciding the intended legal effect.
Require the issue map to quote only the short changed terms and point to the supplied section. AI can draft a neutral question for the responsible counsel. It must not invent a common-law meaning or infer that calendar days were intended.
Check every cross-reference affected by the edit
A clause can remain grammatically complete while pointing to the wrong exception.
Draft 5 still points to section 8.2, but the exception moved to 8.3 during the same edit. Record the old and new locations and verify the referenced subject. Do not update the cross-reference automatically without an owner confirming that 8.3 is the intended target.
Add the discrepancy to a review checklist. A document-wide search can find matching numbers, but the reviewer must inspect whether they perform the same function. Similar headings do not prove equivalent meaning.
Separate language options from authority to choose
Generating alternatives is not deciding what the parties should accept.
For discussion, present options that retain business days, use calendar days or restore Service Day, each with the drafting question it raises. Keep these as alternatives in the synthetic issue map. Do not recommend one as legally preferable from the packet.
If a business owner says faster notice is desired, record that instruction but leave legal interpretation and drafting approval with qualified counsel. AI should not transform an operational preference into an executed agreement or binding position.
Deliver a traceable issue map
The handoff should make every unresolved change easy to inspect in context.
Include version identities, textual changes, defined-term issue, cross-reference issue, source locations, owner questions and approval status. Mark the exercise as fictional and avoid reusing its conclusions for real agreements.
Review source fidelity and escalation within this task. This differs from technical documentation because the output deliberately preserves unresolved interpretation and authority for legal review rather than teaching a procedure to an end user.
Sources and scope
NIST: AI risk management. Skills England: workplace AI foundations.
These references provide background, not validation or endorsement of this exercise. The case details, calculations and suggested review questions are original instructional material. Use them to discuss observable work, not to infer customer outcomes, professional credentials or performance in every setting. Before adapting the exercise, confirm the relevant facts, approved tools, data permissions and decision owners. If you change the case, revisit the expected answers and checks as well. These examples describe practice tasks, not a promise that a particular product includes the fictional features.
Sources: [1] [2]
Sources
- 1.AI RMF Core · NIST
- 2.AI foundation skills for work benchmark · Skills England