The short version
"Verified", with no source named. When a tool says it, the claim is more generated text. When a person says it, the question is the same one: verified against what? The National Archives, BAILII, Westlaw, Lexis, the law report itself?
What it is
A claim to have checked is a statement of fact, and it can be true or false like any other. It turns up at the foot of a tool's answer, in a trainee's covering email, in a reply to the other side's letter pointing out a missing authority, and in what a lawyer tells the court afterwards.
The judicial guidance anticipates the question. Judges are told that "it may be necessary at times to remind individual lawyers of their obligations and confirm that they have independently verified the accuracy of any research or case citations that have been generated with the assistance of an AI chatbot". Where a litigant in person may have used a chatbot, it is appropriate to "ask what checks for accuracy have been undertaken (if any)" (Artificial Intelligence (AI): Guidance for Judicial Office Holders, 31 October 2025, section 3, part VI). The word that carries the weight is "independently". A tool cannot verify itself.
A sighting
The example below is made up to show the pattern. A research tool ends its answer with "All citations verified." The trainee who ran it copies the line into the email to the supervising solicitor. Asked how, the trainee says the tool was asked to double-check its cases and confirmed them.
That is the exchange the Bar Council's guidance reproduces from a United States case. Asked for its source for a case, the chatbot replied: "Upon double-checking, I found that the case Varghese v. China Southern Airlines Co. Ltd, 925 F.3d 1339 (11th Cir.2019), does indeed exist and can be found on legal research databases such as Westlaw and LexisNexis." The case did not exist. The guidance's comment is that the lawyer "thought that the LLM was engaging in the human process of reading and understanding the question, searching for the correct answer" (Considerations when using ChatGPT and generative artificial intelligence software based on large language models, last reviewed 25 November 2025, paragraphs 18 and 19).
In England and Wales the question came up in R (Ayinde) v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), judgment of 6 June 2025. Asked to explain, the law centre that instructed counsel said in evidence that it had not been its practice "to verify the accuracy of case citations or to check the genuineness of authorities relied on by counsel" (paragraph 59). When the other side questioned five of the authorities, the law centre asked counsel for copies; counsel did not send them, but "did provide the wording for the email" that the law centre then sent (paragraph 59). The law centre sent counsel's draft on 5 March 2025 with no change but the pronouns, so the words that described the false citations as "cosmetic errors" were counsel's (paragraph 42). Ritchie J had called that description "a grossly unprofessional categorisation" (quoted at paragraph 47). The Divisional Court found that the paralegal who sent the email was "not at fault in any way", and that Ritchie J "could not have known this": privilege had not then been waived, and on the face of the documents the paralegal had written the email (paragraph 63). Read the judgment at caselaw.nationalarchives.gov.uk/ewhc/admin/2025/1383.
Why it happens
A tool's account of its own work is produced the same way as the work. If the citation can be invented, so can the confirmation that it is real. For a person, "verified" is quicker to write than a list of sources, and nobody asks unless something has already gone wrong.
What to say back
"Verified against what? For each authority, give the National Archives or BAILII link, or the law report and page. If the answer is that the tool checked it, it has not been checked."
Origin
- 25 May 2023. In Mata v Avianca, a United States case, a lawyer filed screenshots in which he had asked ChatGPT "Is Varghese a real case" and "Are the other cases you provided fake". The court recorded that ChatGPT "responded that it had supplied 'real' authorities that could be found through Westlaw, LexisNexis and the Federal Reporter" (Opinion and Order on Sanctions, 22 June 2023, Findings of Fact, paragraph 45).
- December 2023. The first judicial guidance on AI is issued. The Divisional Court in Ayinde records that it was updated in April 2025 (paragraph 15).
- 30 January 2024. The Bar Council issues its guidance for barristers. The version last reviewed on 25 November 2025 reproduces the Mata exchange (paragraph 18).
- 6 June 2025. Ayinde and Al-Haroun: the duty is to check "by reference to authoritative sources" (paragraph 7).
- 31 October 2025. The refreshed judicial guidance tells judges that "it may be necessary at times to remind individual lawyers of their obligations and confirm that they have independently verified" the accuracy of AI-assisted research and citations (section 3, part VI).
- Now. The test in the judgment and in the guidance is the source, not the claim. A verification claim is worth the source it names.