AI and the workplace

A fifteen-page grievance just landed. You're fairly sure ChatGPT wrote it.

By Nichola Morgan MCIPD · Reviewed 4 September 2026

It is fifteen pages long. It has numbered headings. It cites a case you have never heard of. And it has arrived from someone whose longest previous email to you was four lines about a car park space.

This is happening a lot at the moment, and the first reaction is almost always the same: a sinking feeling, followed by the thought that this has suddenly become very serious indeed.

It might have. But usually what has changed is the packaging rather than the problem, and it helps enormously to be able to tell the difference.

What is actually going on

Employees have started using AI tools to help write grievances, and the documents that come out look nothing like what most employers are used to receiving.

A grievance written by hand tends to run to a page or two. It is in the person's own voice, it is often a bit emotional, and it is usually about one thing that has gone wrong.

A grievance written with AI help tends to run to fifteen pages or more. It has formal structure, headings and subheadings. It is written in legal register throughout. And it very often takes a single complaint and expands it into a list of separate allegations, because that is what these tools do when you ask them to make something more thorough.

They also contain a distinctive kind of error. These tools generate text that reads plausibly rather than text that has been checked, so the documents sometimes cite cases that do not exist, misquote ones that do, or import legal concepts from other countries entirely. American employment law turns up with some regularity in grievances about jobs in Yorkshire.

This isn't a curiosity at the fringes. The tribunal system itself has noticed — minutes from the Employment Tribunal National User Group in March 2026 record that AI is contributing to more complex pleadings, more applications for reconsideration, and inflated schedules of loss.

Why knowing AI wrote it changes less than you would like

This is the part I most want people to hear, because the instinct runs the other way.

Your obligation to deal with a grievance properly does not depend on who or what helped write it. A complaint about how someone has been treated is still a complaint about how someone has been treated — whether they wrote it alone, a friend helped, a union representative drafted it, or a chatbot produced it. Employees have always had help writing these. The help has simply become much more available and much more fluent.

There is also a trap in reacting to the format rather than the content. An employer who visibly treats a grievance as unserious because it looks machine-generated has just handed the employee something new and rather stronger to complain about.

So, the fact that you can tell it was AI-assisted is genuinely interesting, and almost entirely beside the point.

The real difficulty is length, not sincerity

Nearly every one of these documents has a real grievance inside it. The person is usually upset about something quite specific and identifiable — a rota change, the way a conversation went, being passed over, feeling singled out.

What the tool has done is take that and dress it. One complaint about shift patterns becomes six separate allegations. A feeling of being ignored acquires a legal framework and a schedule of loss. The underlying thing is often quite small and quite fixable, and it is now buried under twelve pages of scaffolding.

Which means the work is separating the complaint from the packaging, and that is not a document-analysis exercise. It's a conversation.

Sitting down with the person and asking them, in their own words, what they want looked at will usually surface two or three real issues in twenty minutes. By setting up a meeting, you can attempt to hear it informally before taking it to a formal route. It also tends to be the step that gets skipped when a document arrives looking this formal, because the formality makes everyone reach for process rather than for a chair.

Answer the complaint, not the document.

The bit nobody mentions: managers are doing it too

I think it is worth being even-handed here, because the conversation about AI at work has a habit of becoming a conversation about untrustworthy employees, and that is not what I see.

Managers are pasting disciplinary details into public AI tools to help draft outcome letters. They are asking chatbots how to word a dismissal. In doing so they are putting confidential information about a named employee into a system nobody at the company controls — and there is now real argument about whether material handled that way retains legal privilege at all. Two decisions in 2026 have touched on precisely this point.

From where I sit, that is at least as significant a risk as an over-long grievance, and it gets a fraction of the attention. If a business is going to have a rule about AI and HR paperwork, it needs to point in both directions, or it will not survive contact with the first manager who reads it.

What tends to help

The employers who handle these well seem to do three unremarkable things.

They acknowledge it quickly and in writing, without commenting on its length or its style. They meet the person early and ask them to explain, in their own words, what they most want addressed. And they respond to the substance they have identified, saying plainly which points they are treating as the grievance and why — rather than either answering all forty numbered paragraphs or ignoring thirty-eight of them silently.

None of that is a trick. It is what a good grievance process looked like before any of this, applied to a document that is trying quite hard to make you forget it.

What this doesn't tell you

Whether a particular grievance has merit, how it ought to be investigated, and what the outcome should be depend entirely on the facts. This describes a pattern, not a route through your situation.

If the document alleges discrimination, harassment or whistleblowing, the stakes are meaningfully higher and it is worth getting proper support early — before the first meeting rather than after it. That is not me being cautious for the sake of it. Those are the cases where the handling in the first fortnight tends to matter most.

General information, not legal advice. Correct as at August 2026; timetables can change.