How can lawyers use AI for briefwriting without citation errors?
By Michael Showalter ·
Short answer
Connect the model to a primary-source research tool, have it draft from the authority it retrieves, and run a dedicated cite-check skill on every draft before the signing lawyer reviews the citations. Fabricated cases have become rare at the current frontier of AI models. They are not zero, which is why verification stays mandatory for every filing.
The errors that still turn up in a well-run workflow are mostly small: a pin cite that is a page off, or a case name italicized when it should not be. They are the kinds of errors a careful paralegal’s cite-check would catch. The workflow:
- Draft from retrieved authority. Connect Claude to a legal research service such as Midpage, DingDuff, or Westlaw so it reads the actual opinions and statutes before it drafts and cites from them.
- Run a cite-check skill. A reusable Claude skill checks every citation and quotation in the draft against the source text and flags anything that needs attorney review in comment bubbles.
- Add a second layer when the stakes warrant it. Run a citator, or have a second model make an independent citation pass.
- Check personally before filing. The signing lawyer still reviews every citation. The tools make that review faster and should catch everything important.
The same skill works on the other side’s brief. In a pending Ninth Circuit appeal, Showalter PLLC’s cite-check skill found in three minutes that the opposing brief rested a central argument on a hallucination.
Showalter AI Consulting teaches this workflow using real briefs.
Showalter AI Consulting is not a law firm and does not provide legal services or legal advice. Our services do not create an attorney-client relationship.