When Courts Start Asking Who Checked the Citation

A caseload curve and a rule proposal, on the same day

Colorado’s federal trial court may reach an unprecedented 6,000 civil case filings this year. Under the federal judiciary’s weighting formula, each of the district’s judges carried 718 civil and criminal cases in 2025 — 4,133 civil cases filed against 3,544 the year before.

On the same day those numbers were presented, the Denver-based U.S. Court of Appeals for the Tenth Circuit floated a rule change that would formally hold lawyers and self-represented litigants responsible for verifying the accuracy of any AI-generated content in their submissions. Several judges in the district already require filers to attest to their use or non-use of artificial intelligence.

Those two facts belong together. The volume is arriving in part because drafting has become cheap; the rule proposal is about who remains answerable for what the drafting produced.

What the judges are actually describing

U.S. Magistrate Judge N. Reid Neureiter put the problem precisely. The court is seeing large numbers of filings from self-represented plaintiffs that have, in his words, “all the appearance of rationality, coherence, research, organization.” At the hearing, the person who filed often cannot explain what they submitted.

That is a specific failure, and it is worth naming carefully. The documents are not incoherent. They are well-formed and unsupported — the surface of legal reasoning without the record beneath it. Self-represented filings are projected to reach roughly 800 in 2026.

The cost lands on the court either way. As Neureiter noted, a filing may still contain a silver bullet, “but you’ve got to respond to all that.” Judge Charlotte N. Sweeney observed the same pattern from the bench: “There’s definitely a lot more motions coming in.”

Other figures from the same presentation sharpen the picture. Only 27 civil cases went to trial in 2025 — the lowest in recent history apart from 2020 — with an average of three years from filing to trial. Consent to magistrate judges fell below 10%, down from 14% in 2024.

Why this reaches appellate work first

Trial courts absorb volume. Appellate courts absorb citations.

An opening brief is an argument about a record that already exists. Every factual assertion points to a transcript page, a court file entry, an exhibit. A brief that cites something the record does not contain is not merely weak — it misdirects the panel and opposing counsel, and it is discoverable by anyone who opens the record.

This is why the Tenth Circuit’s proposal is framed around verification rather than prohibition. It does not ask whether a tool was used. It asks whether the filer confirmed that what the tool produced is true. That is a question about process, and it is answerable — but only if the process was built to answer it.

Colorado’s appellate rules already assume that kind of discipline elsewhere. C.A.R. 28(g) caps a principal brief at 9,500 words, excluding the caption, tables, and certificates. C.A.R. 32 governs the typeface, spacing and margins. Eleven sections are required, in order. Compliance in appellate practice has always been checkable; the rule proposal extends that expectation to the provenance of the argument itself.

What “verified” has to mean in practice

If a court can ask who checked a citation, a drafting process needs three properties.

Citations must come from the record, not from memory. A system that can only draw record citations from documents the attorney uploaded cannot invent a case that does not exist. The constraint is structural rather than a matter of care.

The checking must leave a trace. Verification that happened only in someone’s head is not something a filer can attest to weeks later. A per-citation check against the source documents produces a record of what was confirmed and what was not.

Unverified passages must stay visible. Automated checking will not reach 100%, and a process that quietly implies otherwise is worse than one that flags its own gaps. What matters is that the reviewing attorney can see exactly which passages still need human judgment before the brief is filed.

None of this is an argument for or against using AI in legal drafting. It is an argument that the two questions have separated. “Did you use a tool” is becoming less interesting to courts than “can you show that you checked.”

Built for Colorado appellate practice

COAPP generates opening briefs for the Colorado Court of Appeals with these constraints built in. Record citations are drawn only from documents you upload, formatted as TR (date) p.X:Y-Z for transcripts, CF p.X for the court file, and EX p.X for exhibits — so a citation to something outside the record cannot be produced. Each citation is checked against your source documents automatically, with a color-coded overlay separating confirmed references from those needing a second look; 79.5% of citations are verified automatically, and the remainder are marked rather than assumed.

Every passage of AI-generated legal reasoning is highlighted in the editor and in the exported document, precisely so a reviewing attorney can see which analysis needs their judgment before filing. Formatting follows C.A.R. 32 — 14-point Times New Roman, Bookman or Garamond, double-spaced body, the required margins — with all eleven sections in order, the 9,500-word limit enforced, and the Certificate of Compliance calculated and ready to file.

The point is not that the brief writes itself. It is that when a court asks who checked the citation, the answer is in the document.

Continue Reading

Source: Exploding caseloads, increased AI use affect Colorado’s federal trial court — Colorado Politics, 18 September 2026.