Who it is for, and why it pays.
If your answers have to show where they came from, CiteOnly is built for you. Every figure below is on the public record.
Three people, three reasons
You own the AI programme
CiteOnly gives you an answer you can show a regulator.More
You have to put AI in front of people whose work goes outside the building. One invented answer ends the programme. Nothing in a CiteOnly answer was not on file, and every line has its source.
You run the pipeline
CiteOnly replaces the whole chain: the agent, its guardrails and the audit-trail tool.More
Finding the right documents works. The agent that writes the answer is where the risk sits, and the guardrails and the audit-trail tool bolted on after it are two more systems to maintain. CiteOnly replaces all of that: the answer is quotes from your documents, and the audit record is the output itself.
You sign the answer
Checking is a read. The passage is beside the answer.More
The answer is your liability, so today you re-check all of it. With CiteOnly the passage sits beside each line, and when your documents do not answer, it says so.
Built for legal, financial services and pharma
Legal
Courts have sanctioned lawyers for filing citations an AI invented.More
Questions over contracts, case files and precedent go in front of a judge. Courts have also set aside decisions built on invented citations.
Financial services
Examiners ask what a decision was based on and expect a record.More
Questions over policies, procedures and regulations need an answer that shows its source.
Pharma and healthcare
Inspectors expect every statement to trace to a record.More
Questions over procedures, batch records and submissions have to come with the page they rest on.
The record so far
Source
AI Hallucination Cases Database, Damien Charlotin, damiencharlotin.com/hallucinations, read 14 September 2026. It started counting in 2023, and the count grows most weeks.
Source
The same database, decisions recording a monetary penalty: a sanction, the other side's fees, wasted costs.
Source
The same database, decisions recording a referral, a reprimand, a suspension or a strike-off.
What courts have ordered
The largest orders on the record.
-
What happened
United States, District of Oregon. Briefs cited 15 cases that do not exist and 8 fabricated quotations. A sanction, then the other side's fees, against two lawyers. The claims were dismissed.
-
What happened
United States, Southern District of Florida. Fees and costs against the lawyer, and a referral to the Florida Bar.
-
What happened
United States, Southern District of New York. Fees and costs, and the lawyer's second referral to the Grievance Committee.
-
What happened
Abu Dhabi Global Market. Wasted costs ordered against the law firm itself, for a defence built on authorities that did not exist or did not say what was claimed. The dirham is pegged to the dollar, and the bar is drawn at that rate.
-
What happened
United States Court of Appeals, Sixth Circuit. Two lawyers, over two dozen fake citations across three appeals. Plus the other side's full fees, double costs, and a referral for discipline.
What it has done to careers
A solicitor, England, 2026
The Solicitors Disciplinary Tribunal struck a lawyer off the register. One of the two grounds was defending himself with AI-generated submissions containing fabricated authorities. The tribunal said either ground alone would have produced the same result. The judgment.
A lawyer, Colorado, 2023
Suspended for one year and one day for filing a motion with fictitious cases from a chatbot and not correcting it. The decision.
Two sets of lawyers, London, 2025
The High Court set out what awaits a lawyer who puts non-existent authorities before a court: wasted costs, contempt, a police referral. It sent the lawyers in front of it to their regulators. The judgment.
All solicitors, England and Wales
The Solicitors Regulation Authority put it in a warning notice: "You remain accountable for all work and advice delivered to clients." It received 42 reports of possible AI misuse in a year. The notice.
When the decision itself rested on it
A judgment set aside
In July 2026 the Supreme Court of India set aside a tribunal's judgment and the appeal above it, because the tribunal had relied on precedents that did not exist. The Court called for "a zero-tolerance mode" and held that such a decision "is no decision in the eyes of the law". The judgment.
A penalty order set aside
Two months later the same court did it again. A customs penalty order and the High Court order upholding it were set aside, because the officer's reasoning cited cases that were non-existent or misquoted. The order.
An airline held to its chatbot
In 2024 a Canadian tribunal held an airline to a wrong answer its website chatbot had given a customer about fares, and ordered it to pay the difference. The company's argument that the chatbot was responsible for its own words did not succeed. The decision.
A report refunded
A consulting firm repaid the final instalment of a A$439,142 report to the Australian government: A$97,587. The report had been found to contain incorrect references and quotations. The contract notice.
How often the tools you can buy today get it wrong
Published studies. CiteOnly: 0 factual hallucinations.
Source
Stanford RegLab and HAI, Journal of Empirical Legal Studies, tools tested in 2024. At least 1 in 6; for the worst tool, 1 in 3.
Source
Dahl, Magesh, Suzgun, Ho, Journal of Legal Analysis, 2023 models.
Source
Liu, Stammbach, Henderson, arXiv 2026, a study that seeded errors into legal briefs.
What regulators ask an answer to show
EU AI Act
"sufficiently transparent to enable deployers to interpret a system's output and use it appropriately"
US FDA
"secure, computer-generated, time-stamped audit trails"
Solicitors Regulation Authority
"Reliance on an output of AI would not be a suitable defence"
Supreme Court of India
"It is necessary for Courts to adopt a zero-tolerance mode for producing, citing or using AI-generated precedents without verification."
What checking AI output costs today
Source
Workday with Hanover Research, January 2026, 3,200 respondents.
Source
Against 4.6 hours a week they say it saves them. Foxit with Sapio Research, March 2026, US and UK.
Source
LexisNexis UK, September 2026, 543 respondents. 81% are more comfortable when the AI is grounded in sources.
Source
Professionals in law, risk, compliance, tax and audit. Thomson Reuters Institute, 2025. Four in ten say an output would have to be 100% accurate before they would use it without a person reviewing it.
Source
Thomson Reuters Institute, 2026. 90% said reasoning that can be explained and defended.
Source
Doximity, 2026, in every age group and every specialty studied.
What CiteOnly changes
Nothing invented reaches a filing
Every line of an answer is a quote from a document you supplied.
The audit trail is built in
Each quote carries its document, section and page. When an examiner asks what an answer rested on, the answer already says.
Time saved on grounding validation
The passage is beside each line of the answer, so validating an answer is reading it. Nobody has to search for the source.
Run it on your own documents
Your files, your questions, your reviewers.