A string of text appended to website addresses is quietly revealing when ChatGPT has been used in legal research and drafting. Court opinions, government reports, and congressional correspondence now contain URLs ending in "utm_source=chatgpt.com" - a tracking tag that gets added when someone copies a link from the AI tool and pastes it directly into a formal document without cleaning the citation.
The appearance of these tags does not prove misconduct or that a citation is fabricated. But it signals that a link likely passed through ChatGPT, raising questions about whether the cited source was independently verified before it was submitted to a court or published by an agency.
How the tracking tag works
UTM parameters are short code snippets appended to URLs that identify referral sources and traffic channels. They have been in use since the mid-2000s. When ChatGPT provides a link to an external source in response to a query, it may tack on "utm_source=chatgpt.com" to track the origin of the click. If a lawyer, clerk, or judge copies that link wholesale into a brief or order, the parameter remains visible in the final document.
Conventional AI-text detectors are unreliable for legal work. They produce false positives, miss AI-assisted drafting, and offer little actionable information in a litigation context. The URL tag, by contrast, is a concrete artifact of the drafting process - an operational trace that ordinary detection tools cannot surface.
Where the tell is showing up
Federal court opinions now contain these markers. In Thrasher v. VanNorstran, the District of Arizona cited a Naphcare URL with the ChatGPT source tag. In Lawson v. PPG Indus., Inc., the Central District of California did the same with a link to a US Consumer Law Attorney Fee Survey Report. The District of New Mexico cited a Tipswatch article using a ChatGPT-tagged URL in United States v. Clements, and the Western District of Tennessee linked to a state annual report with the same appended parameter in Wooden v. Lee.
The issue extends beyond the judiciary. IRS Publication 4812, released in December, cites a White House memorandum using a URL ending in "utm_source=chatgpt.com." A January 2025 letter from Rep. Pablo José Hernández, the resident commissioner of Puerto Rico, to Mark Zuckerberg regarding Meta's fact-checking policies includes a ScienceDaily article link with the ChatGPT tag. The US International Trade Commission's "The Year in Trade 2024" report cites an Orrick article through a similarly tagged URL.
Drafting artifacts left behind
Some documents contain even more obvious signs of AI involvement. A June 2025 report produced by the city of Ogdensburg, New York, included the following AI-generated transition: "Certainly. Here's a revised version that focuses more directly on improving the assessment process and the reasons for doing so." A pro se litigant's motion to dismiss in Escobedo v. Turpen contained a similar artifact: "Here's a revised version of that paragraph with this shift in emphasis."
These remnants reveal something about the drafting process. They do not confirm that the final work product is incorrect or that the cited authorities are unreliable, but they warrant additional scrutiny - particularly in legal filings where parties rely on the accuracy of representations made to the court.
Why this matters for legal professionals
Lawyers already have a duty of competence that extends to their own use of AI tools. That responsibility will increasingly include recognizing signs that opposing counsel, courts, or agencies may have relied on AI-assisted research without adequate verification. A ChatGPT-tagged URL in an opponent's brief justifies closer inspection of the cited authorities. The same tag in a judicial order should prompt counsel to check whether the sources actually support the propositions for which they were cited. For AI for Legal practitioners, the skill set is shifting: competence now means knowing not only how to use the technology responsibly, but also when to question, test, and challenge work product shaped by someone else's AI use.
Your membership also unlocks:
