Law report No. GLW-8664 · filed November 18, 2026
Legal PracticeReported case
ChatGPT Tracking Tag Found in 298 Federal Cases and 133 SEC Filings
A HumanizeMy.ai study found ChatGPT's utm_source tag in 298 federal cases and 133 SEC filings. Researchers say it proves no wrongdoing — only that citations went unchecked.
By Priya Raman4 min read775 words
Holding
- Study found ChatGPT's utm_source=chatgpt.com tag in 298 federal cases across 71 courts and 133 SEC filings
- SDNY (55), N.D. Cal. (31) and C.D. Cal. (27) accounted for 113 of the 298 cases; copyright cases led with 30
- 67 of 133 tagged SEC filings were investor fundraising documents; one in five tagged links pointed to a market-research vendor
- Tag appeared in 15 U.S. government documents, including two federal court opinions, and 3 U.K. publications
- Founder Firat Mihci: 'The tag does not prove anyone did anything wrong. It proves nobody looked.'

A study released today found ChatGPT's link-tracking tag in 298 federal court cases across 71 courts and in 133 filings with the U.S. Securities and Exchange Commission, along with peer-reviewed papers, government documents and Wikipedia articles. The researchers searched six public archives. They opened each document by hand to confirm the tag was present.
The tag itself is simple. When ChatGPT supplies a hyperlink, it appends the string "utm_source=chatgpt.com." Anyone who notices can delete it in seconds. Most filers, it appears, did not notice.
Who ran the study, and how?
HumanizeMy.ai, an AI writing and detection platform that also publishes research, conducted the review last week. Its researchers searched six public archives, then opened each hit to verify the tag rather than relying on automated matches alone.
For the federal courts, the team used the RECAP archive of filings maintained by the Free Law Project. That choice carries a caveat the researchers themselves flag: RECAP is not a complete record of federal filings. The true count of tagged cases could be higher than 298.
Three districts concentrate the findings. The Southern District of New York accounts for 55 cases, the Northern District of California for 31, and the Central District of California for 27 — together, 113 of the 298 cases, or roughly two in five. The tag appeared most often in the complaint, the filing that initiates a case.
Copyright cases lead the list of tagged matters at 30, followed by other statutory causes of action, civil rights, personal injury, antitrust and patent claims.
What turned up at the SEC and in government?
Half of the tagged SEC filings — 67 of 133 — sat in documents a company files when it asks investors for money. That placement matters. Fundraising disclosures carry liability, and a copied tracking tag suggests, at minimum, that someone inserted a citation from a chatbot answer without review.
One in five of those tagged links pointed to a market-research vendor. The researchers read this as evidence that filers used ChatGPT to research market size — the kind of figure that can anchor a valuation pitch.
Government documents show the same pattern. The tag appeared in 15 U.S. government publications and three U.K. government publications. The U.S. set breaks down as:
- Five documents from the Federal Register
- Four records from congressional hearings
- Two congressional committee reports
- Two federal court opinions
The study did not examine the reliability of the cited sources themselves.
What does the tag actually prove?
By the researchers' own account, very little — and that restraint is the study's most credible feature. Firat Mihci, a computational linguist and founder of HumanizeMy.ai, said the study shows only that a link was copied from a ChatGPT answer. It does not show that AI wrote any text, that a cited source is wrong, or how often ChatGPT was used overall, since anyone can delete the tag and leave no trace.
"The tag does not prove anyone did anything wrong," Mihci said. "It proves nobody looked."
Oliver Roberts, CEO and founder of Wickard.ai and co-director of the WashU Law AI Collaborative, reached a similar conclusion in an article he published earlier this year on Bloomberg Law. "This doesn't prove misconduct, that a citation is fake, or that ChatGPT drafted the entire thing — and it doesn't necessarily mean the cited source is unreliable," he wrote. "But it's a red flag."
"At minimum, it suggests that someone likely copied a source link from ChatGPT and inserted it into a formal document without cleaning, normalizing, or perhaps independently checking the citation," Roberts wrote.
The findings also imply something more basic: whoever filed those 298 court cases and 133 SEC submissions used ChatGPT somewhere in the process. If detection turned on nothing more than a filer failing to hit backspace a couple of times, the undetected use is presumably larger than the visible count.
What should practitioners take from this?
The practical lesson is hygiene. A tracking tag in a filed complaint or an investor-facing disclosure signals to opposing counsel, regulators and the press that a citation may not have been independently verified. Counsel reviewing drafts should search filed documents for "utm_source" strings before submission, and should treat a hit as a prompt to re-check the underlying source, not as proof of misconduct.
The study's count is a floor, not a ceiling. RECAP does not capture every filing, and the tag disappears the moment someone edits the URL. The one-sentence takeaway: 431 formal filings carried a chatbot's fingerprint, and the only thing it proves is that nobody checked.
via lawnext.com (Source)
More from Priya Raman
Show full bio
Staff writer covering companies, deals and strategy in International law and legal practice at Global Law Wire.
334 articles
Also before the court
- Ex-Solicitor Loses Appeal Over Fake Cases From Google Search
- Record Fines Fail to Curb AI Hallucinations in Court Filings
- Civil Justice Council Sets Up Working Group on AI and 'Fake' Citations
- EU Court Adviser Says Google Gemini's Use of Press Content May Infringe Copyright
- Third Circuit Backs Thomson Reuters in ROSS Copyright Fight