Employers have likely recently read a number of grievances drafted by generative AI ("genAI"). For the purpose of this article, we asked our AI tool to draft an allegation as part of a grievance, using obvious AI language:
"I allege that the decision to commence and pursue the disciplinary investigation constituted an act of unlawful discrimination. The investigation did not arise from any genuine, fair or proportionate concern. It arose because of a protected characteristic. I say that deliberately, or rather carefully. It may not have been overt, explicit or acknowledged, but it was differential treatment nonetheless. The employer treated my conduct as inherently problematic, escalated matters with unusual speed, and framed neutral behaviour as misconduct. That combination of assumption, escalation and framing matters. Each step reinforces the next. Taken together, they reveal a pattern that cannot be explained by legitimate management action. It can only be explained by discrimination."
The above paragraph alleges a disciplinary investigation was an act of discrimination, but it is not clear what the employee alleges the employer did which could constitute discrimination. We have seen this drafting style in grievances running to over 20 pages - employees ask AI to draft their grievances, which produce numerous paragraphs of similar allegations, none of which tell the reader what the alleged conduct actually was.
Why AI‑generated grievances look the way they do
GenAI tools such as ChatGPT, Copilot and Gemini produce text by predicting the next most likely word, iteratively, based on statistical patterns in the data on which they were trained. They do not analyse evidence in the same manner humans do. They simply mimic the features of analytical writing without carrying out the actual reasoning.
This technical limitation explains many of the recurring features now seen in AI‑drafted grievances. When prompted to “draft a grievance alleging discrimination”, the model has no access to the employer’s actual conduct unless the user supplies it. In the absence of detailed factual input, the tool defaults to abstract language and recognisable rhetorical devices such as self-corrections (making a loosely relevant point to "knock it down") and tricolons (where the tool lists three vaguely relevant things as evidence). The language sounds forensic and measured, but a detailed reading often reveals the actual allegations are often hidden behind dressed-up text. The problem is that managers and HR teams are forced to read these overly-verbose allegations as the law, as set out below, requires them to respond to the grievance.
The legal framework for grievances
The law relating to grievances is settled and has not changed in response to genAI. Employers owe an implied contractual duty to afford employees a reasonable and prompt opportunity to obtain redress for grievances that they raise. That duty arises once a grievance has been articulated to the employer. A failure to deal with a grievance, or a serious failure in the way a grievance procedure is conducted once commenced, may amount to a breach of the implied term of mutual trust and confidence. If the employee resigns in response to that breach, it may lead to a constructive dismissal claim. Further, if the grievance contains an allegation of discrimination, the alleged failure to deal with it may amount to a detriment in a victimisation claim.
Alongside the above implied duty, sits the ACAS Code of Practice on Disciplinary and Grievance Procedures (the "Code"). The Code defines grievances broadly as concerns, problems or complaints that employees raise with their employers and sets out minimum standards of fairness. Those standards include dealing with matters promptly, carrying out appropriate investigations, holding a meeting to allow the employee to explain their complaint, issuing a reasoned decision in writing and offering a right of appeal. Although the Code is not legally binding, employment tribunals must take it into account where relevant and may adjust awards by up to 25% where either party has unreasonably failed to comply with it.
What does this mean for employers?
Despite the difficult features of AI-drafted grievances, employers cannot ignore or reject them on the basis of the poor drafting. The legal obligation is to engage with the substance of the complaint, not the style in which it has been expressed.
In our view, the most effective way to deal with AI‑drafted grievances is to use the grievance meeting to ask the employee, in their own words, to explain what they say the employer did wrong. The aim is to distil the grievance into allegations that can be reasonably investigated. That may involve asking the employee to identify specific incidents, dates, and individuals, and to confirm whether certain passages of the written grievance reflect what they intended to allege.
It is good practice to follow that meeting with a written summary setting out the allegations as the employer understands them. Doing so demonstrates the employer has taken reasonable steps to understand the grievance before investigating it. It also reduces the risk of time and resources being spent addressing expansive AI‑generated language rather than the substance of the complaint.
Employers may also wish to write into their grievance policies words that allow the above process to be undertaken (albeit this should remain non contractual). They should take legal advice in the event an employee cannot or will not attend a meeting to clarify their grievance. The appropriate solution will hinge on the facts of the specific scenario.
