Certificate of AI Use: What Courts Are Asking For, and Sample Language

Courts are now asking you to certify how you used AI

The question of whether lawyers may use generative AI has largely been settled in practice. The live question is narrower and more procedural: when you file something, does your court require you to say whether AI helped write it, and to certify that you checked it?

For a growing number of courts, the answer is yes. In Parity in Practice: The Defender’s Duty to Ethically Use AI, Mason R. Clark of St. Mary’s University School of Law notes that nearly two dozen jurisdictions now require attorneys to provide disclosure and/or verification statements for documents prepared with the assistance of generative AI. He takes his own advice, opening the article with exactly such a statement — naming the tasks the models performed and certifying that every citation was manually verified.

That is the shape of the obligation, and it is worth understanding before a clerk asks you about it.

Disclosure and verification are two different things

Most requirements combine them, but they are not the same, and conflating them is where filings go wrong.

Disclosure is a statement of fact about process: generative AI was or was not used in preparing this document, and if so, for what.

Verification is a statement about diligence: a human being checked the output. In practice this almost always means checking that every citation exists and stands for the proposition it is cited for.

Verification is the part that matters. The reported sanctions cases have not turned on lawyers using AI. They turned on lawyers filing citations that did not exist. A disclosure statement without verification announces the risk without addressing it.

Where the requirement comes from

Three separate layers, and they can each apply at once:

1. Standing orders of individual judges. This is the most common and the easiest to miss. A judge’s standing order applies in that judge’s courtroom regardless of what the district’s local rules say. They vary in scope — some cover every filing, some only substantive motions.
2. Local rules of a district or state court.
3. State bar guidance and the Rules of Professional Conduct. Competence, candour to the tribunal, and confidentiality apply whether or not a specific AI rule exists. An ethics opinion does not need to mention AI to govern how you use it.

Because the first layer is judge-by-judge, there is no substitute for checking the actual standing order of the actual judge before the actual filing. Any list, including this one, is a starting point rather than an authority.

Sample certificate language

The requirement is new enough that many practitioners are drafting from scratch. Below is neutral language covering both halves. Adapt it to your court’s specific wording — where a judge has prescribed a form, use theirs.

If AI was used:

Certificate Regarding Use of Generative Artificial Intelligence

Pursuant to [rule / standing order], the undersigned certifies that generative artificial intelligence was used in the preparation of this document for the following purposes: [e.g. organising and summarising record materials; drafting portions of the statement of facts; formatting citations].

The undersigned further certifies that every legal authority cited herein has been verified by a human being to confirm that it exists, that the quotation is accurate, and that it stands for the proposition for which it is cited; and that no substantive legal analysis was adopted without independent review by an attorney admitted to practise before this Court.

Dated: __ /s/ ____

If it was not:

The undersigned certifies that no generative artificial intelligence was used in the drafting of this document, and that all legal authorities cited have been verified.

Two drafting notes. Say what the tool actually did — “summarising the record” and “drafting argument” are different admissions, and vagueness invites the follow-up question. And do not certify verification you did not perform; that turns a paperwork requirement into a candour problem.

What a firm should do this month

• Find out, per judge. Before filing, read the standing order. Keep a shared note of which judges in your regular courts require what.
• Write a one-page internal policy. Where AI may be used, what verification each use requires, and who signs off. Clark’s argument is that governance is what separates responsible use from exposure, and a page is enough to start.
• Make verification a step, not an intention. Someone opens every cited authority. If your tools make that fast — citations that point at a document, page and line rather than a general impression — the step costs minutes. If they do not, it costs hours, and steps that cost hours get skipped under deadline.
• Keep a record of the check. If you later have to explain your process, a contemporaneous note is worth considerably more than a recollection.

The point of all this

None of this is anti-AI. The same article that describes these obligations argues that defenders have an affirmative duty to use these tools, because the other side already is and the resource gap is real. Disclosure and verification are what make that use defensible.

The firms that will have trouble are not the ones using AI. They are the ones who cannot say, afterwards, exactly what it did and who checked it.

Source: Mason R. Clark, Parity in Practice: The Defender’s Duty to Ethically Use AI, St. Mary’s University School of Law — SSRN abstract 6337279: ssrn.com/abstract=6337279. This page is general information about a fast-moving procedural area, not legal advice; verify the current requirements of your own court before filing.