This case report is now 0 months old and may not represent current law.
Can you use ChatGPT to write your court or tribunal documents? Yes, nobody has banned it, but you are personally responsible for everything in the document you send in, and the Employment Appeal Tribunal has now set out what it expects you to have checked before you do.
The case is Hancox v Sutherland & Others [2026] EAT 139, a judgment of Marcus Pilgerstorfer KC (sitting as a deputy High Court judge) handed down on 17 September 2026. Most of what has been written about AI in court so far is about fake cases, and this one is a bit different, because the main problem was the length of the document and the fact that nobody had checked it.
What happened in the case
Mr Hancox worked for the National Farmers’ Union as a handyperson and was dismissed in February 2020 during his probationary period. He brought twelve employment tribunal claims, including whistleblowing and disability discrimination claims. In 2021 a tribunal struck out his claims against four individual employees, after finding that he had run a campaign against them on LinkedIn (more than 145 posts in a few weeks) and that a fair trial was no longer possible.
He appealed, representing himself, and by July 2026 the appeal had reached a preliminary hearing at which the EAT decides whether there is an arguable point. The EAT’s Practice Direction says a skeleton argument should generally be 5 to 15 pages, and that anything over 20 can be sent back. Two days before the hearing Mr Hancox filed a skeleton argument which ran to 300 pages and almost 132,000 words.
The covering email said he didn’t expect the judge to read all of it before the hearing. It also invited the other side’s lawyers to identify any factual errors, wrong quotations or mistaken references by 4pm the following day, and said that a complaint that the document was too long wouldn’t count as identifying an error. At the hearing he told the judge he had used ChatGPT because he had to prepare it quickly, that much of the document “lacked credibility”, and he didn’t rely on it.
What he relied on instead was a volunteer barrister from the free ELAAS scheme, who had spent about 90 minutes with him earlier that week. She argued one point, which was that he hadn’t had fair notice that the strike out application would actually be decided at the 2021 hearing. The hearing took less than an hour and the judge decided that point was arguable, so it goes forward to a full appeal. The 132,000 words played no part in that.
What the EAT said about AI
The judge was not hostile to AI. He accepted that it can help people who can’t afford a lawyer, and he referred to evidence the Lady Chief Justice gave to the House of Lords in June 2026 that many judges find AI-assisted submissions from litigants in person easier to follow than the alternative.
He then said that anyone who uses AI to prepare a document for a court or tribunal, whether they have a lawyer or not, must at a minimum make sure of three things:
- The document complies with the procedural rules which apply to it.
- It has been checked for accuracy as thoroughly as the person is reasonably able, on the facts, the evidence and the law, including any cases it cites.
- It contains only the relevant points, concentrating on the best arguments and without repeating itself.
Submitting what the AI produced without checking it is “not acceptable”, and neither is asking your opponent to do the checking for you. The judge said a document like this one is likely to be rejected in future, and he pointed to the other powers a tribunal has, which include costs orders and striking out the appeal.
I should say what the judgment doesn’t do. Mr Hancox wasn’t penalised for the skeleton argument, the appeal itself hasn’t been decided, and the comments on AI are general guidance given at a preliminary hearing. They are plainly meant for everybody though, and the same message has now come from the High Court, the Court of Appeal and the tax tribunal in cases the judge cites.
What lawyers should take from it
For lawyers the starting point is still R (Ayinde) v London Borough of Haringey [2025] EWHC 1383 (Admin), where the Divisional Court said that a lawyer who uses AI for research has a professional duty to check it against authoritative sources, in the same way as if a trainee had done the work. The Hancox judgment lists six more cases since Ayinde in which regulated lawyers were held to that standard, including one where supervising solicitors were held accountable for not checking a junior’s work.
What Hancox adds is that accuracy is only one of the three requirements, and the minimum standard is expressed to apply to “the litigant or representative”. A skeleton argument can have every citation right and still fail because it is too long, unfocused or repetitive, and length is the thing these tools produce most easily. One of the cases the judge mentions also involved client letters being put into an open AI tool, which is the confidentiality and privilege problem I wrote about in Can I put my solicitor’s advice into ChatGPT?
I use AI tools every day and I teach other lawyers how to use them properly (the training and my book are at The Lawyer’s Prompt), so this isn’t an argument against using AI. The point I make in every session is the one the judge makes here, which is that the checking has to be done by the person whose name is on the document.
What it means if you run a business
If you represent your own company, in a small claim or in defending a tribunal claim, the same standard applies to you. The judge applied the Supreme Court’s approach that a litigant in person is expected to find and follow the rules, and those rules usually say how long a document can be and what it must contain. The judgment also notes that courts have been cautious about witness statements prepared with ChatGPT, so a statement needs to be in the witness’s own words.
The more likely position for an employer is being on the receiving end. The judgment records that AI is being used with increasing frequency in the tribunals, so there’s a fair chance that at some point you’ll receive a very long claim, grievance or run of emails which has plainly been written with it. It isn’t your job to check the other side’s document for them, and the tribunal can be asked to control the volume. Here the EAT ordered that applications would only be considered if made on the proper form, and warned Mr Hancox that he could be restricted to writing by post.
That doesn’t make the cost go away. The four individuals in this case have been dealing with it for more than five years, and their lawyers still had to respond to a great deal of correspondence along the way.
What to do now
If you’re drafting your own document with AI, find the rules for that court first and tell the tool the page limit. Then check every case, quotation and date against the source (Find Case Law and legislation.gov.uk are free), and take out anything you can’t verify or couldn’t explain to the judge yourself.
If you receive a very long AI-written claim or complaint, don’t reply at the same length. Work out which points actually matter, respond to those, and ask the tribunal to manage the rest.
It’s also worth having a short written rule for your staff about what can go into AI tools, and that anything going to a court, a regulator or the other side in a dispute is checked by a named person before it’s sent.
I don’t run tribunal or court cases myself; those go to my Nexa colleagues who litigate day to day. I do advise employers and review contracts, including ones drafted with AI. Get in touch if you need some help.
Written by Steven Mather, a business solicitor acting on company sales and purchases. This is general information about the law, not legal advice on your situation.
Written by Steven Mather, a business solicitor acting on company sales and purchases. This is general information about the law, not legal advice on your situation. Law correct as at .
