Employees are increasingly turning to AI tools when they want to put a workplace complaint in writing, and the documents landing on HR desks are looking markedly different as a result. What may have previously been a short, factual account of a difficult conversation can now arrive as a lengthy submission, structured under headings, citing legislation and legal concepts and touching on several potential legal claims at once. It is tempting to treat this as a new problem requiring a new response, but in legal terms it is not. Whilst AI may have changed how grievances are presented, it has not changed what employers are required to do about them. The more useful question is not how to cope with the ‘volume’ of paperwork, but how to keep the investigation proportionate to the substance underneath it.
What the law still requires
The reference point for any grievance, however it was produced, is the ACAS Code of Practice on Disciplinary and Grievance Procedures,. The Code requires that a grievance be properly investigated, that the employee be invited to a meeting to discuss it, that they be allowed to be accompanied, that the outcome be communicated in writing, and that a right of appeal be offered.
The duty to investigate is a duty of reasonableness rather than exhaustiveness. There is no obligation to interrogate every sentence of a twenty-page document with equal intensity. In practice, tribunals expect employers to carry out an investigation that is reasonable and proportionate in the circumstances, taking account of the nature of the allegations and the issues that genuinely require investigation. Further practical detail sits in the non-statutory ACAS guide, Discipline and Grievances at Work, which accompanies the Code and is often the more useful reference point when working through a scoping exercise of this kind.
In practice, this means an investigation can spend most of its time on the allegations that could genuinely affect the outcome, and give less weight to repetition or claims the employee cannot support, so long as that approach is recorded and can be explained if it is ever questioned.
The consequence of getting this wrong has not moved either. Under section 207A of the 1992 Act, an employment tribunal can adjust compensation by up to 25 per cent where a party has unreasonably failed to comply with the Code. An employer who disregards an AI-assisted grievance because of its length, or who investigates only the parts that look easiest to deal with, is exposed to that uplift in exactly the same way as an employer who ignores a grievance written by hand. What tribunals will ask is whether the real issues were identified and reasonably investigated, not who or what drafted the document that raised them.
A grievance is a set of allegations, not evidence
An AI-drafted grievance is not itself evidence of anything, however polished it looks. It is a statement of allegations, produced by a tool that has no first-hand knowledge of the workplace and no ability to verify what it is told. The employer’s task is unchanged by how fluently those allegations are expressed. Evidence still has to come from the usual sources, meetings with the employee, relevant documents, witness accounts and the explanations given by those involved.
This matters because a well-drafted grievance can make a weak allegation sound compelling, while a badly written one may describe something genuinely serious. Legal merit and drafting quality are not the same thing, and an investigation that is swayed by the sophistication of the language rather than the strength of the underlying facts risks reaching the wrong conclusion for the wrong reason.
It also explains a pattern that HR teams are increasingly encountering at investigation meetings, where an employee struggles to explain parts of their own written grievance. Statutory references, legal terminology or additional allegations inserted by an AI tool may not reflect language the employee chose or fully understands. That gap does not automatically undermine the complaint. It simply means the investigation should be anchored in what the employee can actually describe and substantiate, rather than in the strength of the document’s phrasing.
A grievance that cites discrimination legislation, whistleblowing protections and constructive dismissal in the same paragraph can appear more legally significant than one written in plain language, and it is worth resisting that instinct in either direction. The seriousness of a complaint depends on the underlying facts, not on the sophistication of the drafting. A poorly expressed grievance, written without any legal vocabulary at all, may still raise issues that require urgent and careful investigation, while an elaborately framed one may not survive contact with the facts once the employee is asked to explain it.
Confidentiality risks run wider than they first appear
Employees preparing a grievance with an open AI tool may paste in material without thinking through where it goes once submitted. This can include their own contract and internal policies, but it can just as easily extend to confidential commercial information, another employee’s personal data, customer details or documents that are subject to legal privilege. Depending on what has been shared and with which tool, this can raise questions under UK GDPR, breach the employee’s own confidentiality obligations, or, in more serious cases, carry disciplinary implications in its own right. These are separate issues from the grievance itself and may need to be addressed alongside it rather than folded into the same investigation. The same care applies in reverse. Employers should be mindful of their own use of AI tools during an investigation and ensure that confidential information is only processed in accordance with internal AI policies and their data protection obligations.
Guarding against inflated expectations
A document that reads like a solicitor’s letter can create expectations to match. An employee who has produced an AI-generated grievance citing several statutory claims may genuinely believe their case is stronger, and more valuable, than it actually is, simply because of how it has been framed. AI tools asked to produce a formal complaint tend to default to assertive, legally charged language, and a single incident of perceived unfair treatment can arrive presented as discrimination, whistleblowing, constructive dismissal and breach of contract all at once, drawn from events that may span months or years with little obvious connection between them. Left unaddressed, that framing can turn a matter that would once have been resolved through a short conversation into something both sides approach as a formal legal dispute, which makes settlement and informal resolution considerably harder to reach later on.
The scoping conversation covered below is the most effective safeguard against this. Establishing early, and in plain language, what the employee is actually concerned about and what outcome they are looking for gives the employer a chance to reset expectations before positions harden on either side. It also allows genuinely minor or straightforward issues to be resolved informally, rather than being carried through a lengthy formal process simply because of how they happened to be written up.
Keeping the investigation proportionate
The practical starting point is the initial grievance meeting. The grievance meeting can serve as a useful opportunity to clarify and scope the issues before the investigation process. Used properly, this meeting allows the employer to distinguish the factual concerns from the legal language used to present them. Asking the employee, in their own words, what happened and what outcome they are seeking will usually narrow the issue considerably from what the written document suggests.
Recording that scoping discussion and sharing it with the employee is not just good practice. It keeps the subsequent investigation focused on what genuinely needs to be examined, and it creates a contemporaneous record that protects the employer if an employee later argues that something was overlooked. Where several distinct allegations have been raised, categorising them at this stage, separating those that are properly evidenced from those that are speculative or cannot be substantiated once explored, allows investigation resource to be directed where it is actually needed. This is not the same as disregarding parts of the grievance. Employers should ensure that each material allegation capable of affecting the outcome is properly considered, but a shorter, lighter-touch look at the weaker or unsupported strands is legitimate, provided the reasoning for that approach is recorded.
Because AI-drafted grievances often invoke discrimination or whistleblowing allegations, employers may wish to proceed on the assumption that , tribunal proceedings are more likely to follow. An investigation plan, notes of meetings, the evidence considered and the reasoning behind the conclusions reached should all be documented as though the matter may be tested later, because the elevated legal framing of these grievances makes that more likely rather than less.
Preparing policies and managers before a complaint arrives
Much of the difficulty caused by AI-drafted grievances can be reduced before one ever reaches a manager’s desk. Grievance policies are worth revisiting with this trend specifically in mind. A policy that actively encourages employees to describe their concerns in their own words, and confirms that the employer will investigate the substance of a complaint regardless of how it has been framed or which claims it references, gives HR a documented basis for the scoping approach described above and signals to employees that a plain, honest account carries just as much weight as a polished one.
Line managers need a specific steer too, since they are often the first point of contact for a grievance, particularly in smaller organisations without a dedicated HR function. A long, assertively worded, legally referenced document can be intimidating to a manager who has not encountered one before, and the instinct to either over-react to it or dismiss it out of hand is understandable but unhelpful either way. Managers should be told plainly that the length or tone of a grievance has no bearing on the standard of investigation and response required, and that the same principles of reasonableness and proportionality apply whether the document in front of them runs to two pages or twenty.
Where AI has a legitimate role for employers
The instinct to keep AI out of the employer’s side of the process entirely is understandable but slightly overcautious. Used to support an investigation, AI tools can be genuinely useful, summarising long witness statements, building a chronology of events, flagging inconsistencies across documents or helping organise material for disclosure, provided the tool is approved for use, appropriate confidentiality safeguards are in place, and processing complies with data protection obligations . None of that involves the tool making a judgement about the merits of the complaint.
The distinction that matters is between using AI to support decision-making and allowing it to produce the decision-making itself. Drafting the outcome letter, or reasoning through whether an allegation is upheld, should remain the decision maker’s own work. An employee is likely to recognise AI-generated language in an outcome letter, and a decision maker who cannot explain the reasoning behind their own conclusion, including under questioning at tribunal, has effectively delegated a legal responsibility that cannot be delegated.
What tribunals will actually ask
AI is changing the language in which workplace grievances are written, but not the standard against which employers are judged. Tribunals will keep asking the same questions they always have. Did the employer identify the real issues? Was the investigation reasonable and proportionate in the circumstances? Did the process follow the ACAS Code? Employers who keep their attention on those fundamentals, rather than on how impressively a grievance reads, are unlikely to be caught out by whether it was drafted by the employee, a representative, or an AI tool.
