Build a Legal AI Work Sample Someone Else Can Review

A polished answer can hide the work you need to demonstrate

An employer looking at a generated memo cannot easily tell whether the candidate understood the sources, noticed a missing fact or simply accepted a fluent answer. Adding the names of several AI tools does not resolve that ambiguity. The sample needs to show where the candidate exercised judgment.

I would start with a small task whose result another person can inspect. A source-bound chronology, a research handoff or a version check can be more revealing than a large demonstration with no visible review. These exercises connect to research, records and communication in the lawyer and paralegal occupational profiles. They are editorial practice proposals, not an employer-endorsed assessment.

Choose the reviewer before choosing the demonstration

A supervising lawyer, legal-operations manager and technology evaluator may need different evidence. The lawyer may ask whether your proposition fits the authority. The operations manager may ask whether a colleague can continue the process. The evaluator may need a reproducible failure and the inputs that produced it.

Write one sentence describing the reviewer and the decision the sample should support. For example: “This exercise lets a supervisor inspect how I separate a source statement from an inference.” That promise is narrow enough to test. A promise to demonstrate complete AI expertise is not.

Use public or synthetic material and an approved environment. A portfolio does not create permission to disclose client work, employer methods or restricted coursework. If a sample requires removing so much context that it cannot be checked, choose a different example.

Make the failure visible without staging a fake achievement

Consider a fictional record with two documents describing the same meeting on different dates. A first pass produces one neat date and leaves out the disagreement. The useful part of the exercise is the correction: preserve both attributed statements, identify their sources and flag the unresolved event date.

You do not need to claim that you discovered this error in client work. Label the dossier as synthetic. Explain the error the exercise is designed to test and show the answer against the supplied record. If you actually run a tool, retain what it returned; do not fabricate a failed output and describe it as a model result.

The review lab provides three complete synthetic dossiers with explanations. Use them to practice the distinction, then create your own permission-cleared example. Repeating the supplied answer is practice, not evidence of independent performance.

The smallest complete review package

A clean document and a working record have different jobs

The reviewer should not have to read every prompt to find the answer. Lead with a concise work product and keep the source and correction record nearby. Too little supporting material conceals your reasoning; too much undifferentiated material transfers the organizing work to the reviewer.

For the fictional chronology, the clean output might be a short table of attributed events. The method note should identify the conflicting date and explain why you left it unresolved. The handoff should say what would resolve it: another record or a decision by the responsible reviewer. The point is to reduce uncertainty about your reasoning, not to erase uncertainty in the facts.

ABA Formal Opinion 512 addresses competence and supervision in lawyers' use of generative AI. It does not endorse a portfolio format or qualify a candidate. Applicable rules and employer policies remain separate from this practice method.

Test the sample with a fresh reader

Give an appropriate reviewer the brief, record and final package. Ask them to identify the main conclusion, the source supporting it and the decision still needed. If they cannot find one of those, revise the package before adding visual polish.

A useful review can also reveal that the problem was too easy. When every source agrees and nothing requires a choice, the exercise may demonstrate formatting rather than judgment. Increase difficulty by adding a meaningful ambiguity or exception, not by multiplying pages.

If you measure time, include preparation, verification and correction. A portfolio can show careful work without making an efficiency claim. An honest account of an unresolved point often gives a reviewer more to assess than an unsupported assertion that the tool made you faster.

Use the sample for the decision you actually face

For a career transition, compare the sample with the responsibilities in the career-path guide. For a junior legal role, see the associate handoff. A hiring team needs a consistent assessment process as well as a good exercise; the work-sample scorecard addresses that different task.

Keep the sample small enough that a real reviewer can finish it. The standard is a complete chain from source to correction to handoff. If that chain works, the sample gives another person something concrete to discuss with you.