Reviewing AI Work in Chambers: An Evaluation Worksheet

How might court chambers evaluate an AI work product before relying on it?

Evaluate the work against the actual task and authorized record: check source identity, factual attribution, omitted contrary material, legal scope and confidentiality. Identify the human responsible for the substantive judgment. A well-written draft should not bypass the checks that would be applied to another research assistant's work.

Begin with the assignment and permitted inputs

Define whether the task is proofreading, organizing the record, summarizing arguments or researching a legal issue. A permission to edit grammar does not automatically authorize adding facts or legal analysis. Identify the approved information environment before supplying documents.

The October 2025 guidance for covered judicial office holders and support staff addresses confidentiality, accuracy and personal responsibility. [S1] The worksheet here is an editorial proposal, not a restatement of every requirement in that guidance.

• Assignment: the exact output requested.
• Input boundary: the authorized record and any permitted outside sources.
• Transformation: what the tool may change or add.
• Review owner: the person checking the work for its intended use.

Check more than the citations

A citation can exist while supporting a different proposition. A summary can quote accurately while omitting a material qualification. Review should therefore test both source fidelity and the relationship between the source and the proposed conclusion.

• Authority: does the cited document exist, and is the relevant passage available?
• Scope: correct jurisdiction, procedural setting and time period?
• Record: is each fact attributed to evidence, an allegation or a finding?
• Countermaterial: does the draft acknowledge relevant contrary information?
• Inference: where does the author move beyond what the source says?
• Uncertainty: are gaps explicit, rather than resolved by plausible invention?

Synthetic review note: allegation converted into fact

A fictional draft says the company received notice on June 2. The supplied pleading alleges notice on June 2, while a response disputes receipt. The draft has silently changed an allegation into an established fact.

The reviewer marks the sentence as a record-characterization error and rewrites it to preserve the dispute. The correction does not decide credibility. The review note records the two source locations and flags any downstream analysis that depended on treating receipt as established.

• Defect: unsupported conversion from allegation to finding.
• Correction: attribute both positions.
• Affected analysis: any reasoning dependent on receipt date.
• Disposition: return the draft for revision and substantive human consideration.

Make the final review attributable

The Illinois Supreme Court's AI policy retains judges' responsibility for their decisions and requires review of AI-generated work. [S2] A review record should identify what was checked, by whom, against which version and with what unresolved issues.

Avoid an undifferentiated approved label. A formatting check, citation check and substantive review are different acts. A proposed worksheet can record them separately without suggesting that every workflow needs the same number of reviewers.

• Formatting checked: readability and document integrity.
• Sources checked: existence, quotation and relevant proposition.
• Record checked: factual attribution and omissions.
• Substance reviewed: reasoning and unresolved issues considered by the responsible human.
• Release status: only the use actually authorized.

Scope and limitations

Proposed evaluation method informed by selected official guidance. This is a research site, not a court, chambers office or source of case-specific judicial instructions.

The example is fictional. Local rules, judicial policies, confidentiality duties and case-specific instructions may require a different process.

Sources

1. Courts and Tribunals Judiciary: AI Guidance for Judicial Office Holders, October 31, 2025 — Guidance for covered judicial office holders and support staff; accuracy, confidentiality, bias and personal responsibility. Checked 2026-11-05
2. Illinois Supreme Court Policy on Artificial Intelligence, effective January 1, 2025 — AI use subject to legal and ethical standards, output review and retained human responsibility; pleading disclosure policy. Checked 2026-11-05

Explore further

This evaluation worksheet supplies context for exploring human review in agent-assisted legal work at Agentic Attorney, https://agenticattorney.com/. No judicial role or court affiliation is implied. Explore Agentic Attorney.