AI hallucinations in employment tribunal claims: why WorkDisputes is different

The problem is not AI. The problem is unguided AI.

An AI hallucination is when a tool gives you a case, quote, paragraph number, or legal rule that sounds real but is wrong or completely invented.

The risk is highest when someone asks an open-ended chatbot to 'find cases for my claim' and then copies the answer into a grievance, ET1, schedule of loss, or skeleton argument without checking it.

WorkDisputes is not built as a blank chat box. It is an AI-powered employment dispute workspace that starts with your facts, checks procedural issues, searches processed tribunal decision data, shows labelled case summaries, links to source material where available, and helps you draft around evidence rather than invented authorities.

What fake AI law looks like in a claim

• A case that does not exist at all, but is written as if it were a real authority.
• A real case name paired with a quote or paragraph that is not actually in the judgment.
• A citation from the wrong court, wrong year, or wrong legal area that does not support the point being made.
• A made-up section number, regulation number, or legal test inserted into a polished draft.
• A case summary that sounds helpful but changes the real reason the claim won or lost.

Recent examples show the risk is real

1. 01

   6 June 2025: Ayinde and Al-Haroun

   The Divisional Court dealt with false authorities linked to suspected or actual AI use. In one matter, a schedule contained 45 citations and 18 did not exist. The court referred lawyers to regulators, noted wasted costs risk, and said the contempt threshold was met.

2. 02

   2 December 2025: Harwood-Allen

   In an Employment Tribunal reconsideration judgment, the judge could not locate a cited case even after checking legal databases and reports manually. The judge said one possible explanation was use of AI or a large language model.

3. 03

   18 November 2025 judgment, published in 2026: A v British Transport Police Authority

   The tribunal recorded that cases sent to the respondent appeared to result from AI hallucinations. The respondent complained about time and cost spent checking them, and the claimant had to apologise and stop using that background assistance.

Why this can damage your case

• Credibility: if one case is fake, the tribunal may read the rest of your document with more suspicion.
• Focus: fake citations pull attention away from the facts, dates, evidence, and procedure that usually decide employment claims.
• Leverage: the employer can use the mistake to make your claim look exaggerated or poorly prepared.
• Costs pressure: where representatives are involved, false authorities can create wasted costs and regulator problems.
• Delay: everyone wastes time checking cases that should never have been in the document.

What WorkDisputes actually does differently

• Case Assessment turns your answers into a procedural checklist: investigation, appeal, written reasons, qualifying service, grievance steps, evidence, and other issues that matter in employment claims.
• The Case Comparison Report uses your facts to match similar tribunal records and produce a structured report based on the matched summaries, not a free-form list of invented authorities.
• Case Law Search gives you precise search for names and phrases, plus broader plain-English search for situations like maternity treatment, redundancy scoring, performance dismissal, or grievance retaliation.
• Search results show case summaries, dates, verdicts, and full decision links where available, so you can move from a quick summary to the underlying source before relying on it.
• The ET1 builder uses your timeline, selected claim type, and a set of legal authorities. It is designed to remove unsupported citation numbers rather than dress up guesses as law.

How that helps a litigant in person

• You do not have to start with a blank page or a vague chatbot prompt.
• You can see the difference between a claim that is weak on evidence and a claim that is weak because of a deadline, service, or procedural issue.
• You can find real tribunal patterns before deciding which examples are worth checking further.
• You can build a claim bundle around the documents that matter: dismissal letters, grievance emails, appeal outcomes, policies, payslips, witness names, and medical evidence where relevant.
• You can move from assessment to research to drafting without losing the thread of your own facts.

The WorkDisputes route before you file

1. 01

   Run Check My Claim

   Start with the procedural audit. It identifies issues that could help or weaken the case, including qualifying service, appeal rights, investigation, written reasons, grievance steps, and evidence gaps.

2. 02

   Generate a comparison report

   Add a short narrative and compare your facts against similar tribunal records. The report is designed to show patterns, risks, missing facts, and examples with source links where available.

3. 03

   Use Case Law Search

   Search exact names or phrases in precise mode, or describe what happened in broader mode. This is how you find potentially relevant examples without asking a chatbot to invent cases.

4. 04

   Check before relying

   A summary helps you understand relevance quickly, but a filed document should be based on your facts and checked source material. If a case cannot be checked, do not use it as authority.

5. 05

   Draft the document

   Use the ET1 builder, grievance letter, appeal letter, SAR, settlement offer, evidence checklist, chronology, schedule of loss, deadline tracker, and compensation calculator to turn the research into practical case documents.

What the tools do not promise

• They do not give legal advice or replace a solicitor.
• They do not guarantee that a claim will win.
• They do not remove your responsibility to check final documents before filing.
• They do not make every automated summary perfect.
• They reduce hallucination risk by keeping AI inside a structured, source-led workflow instead of letting it invent unsupported authorities.

Common questions