AI in Workplace Investigations: Useful Tool or New Risk?
This is the one where the conversation has genuinely shifted in the last year, not because employers started using AI to run investigations, most haven't, but because employees started using it to raise them. That's changed the job on the employer side considerably, whether the business has noticed yet or not.
The volume problem
UK employers handle somewhere in the region of 1.7 million formal disciplinary cases and close to 375,000 grievances every year, and a straightforward investigation already takes around four weeks and 60 hours of investigator time before AI enters the picture at all. What's changed is the shape of what's landing on the desk. Employees are increasingly using tools like ChatGPT to draft grievances, appeals, and even tribunal claims, and the result is often longer, more legalistic, and harder to pin down than the kind of complaint HR teams are used to handling.
Why that's a genuine problem, not just an annoyance
A survey reported through People Management found the substantial majority of HR directors had encountered issues linked to employee AI use in disputes, and a striking proportion of those AI-generated claims contained inaccurate or outright incorrect information. That matters because an employer can't simply dismiss a complaint for being AI-drafted. Under the Acas code, it still has to be investigated properly, however it was produced, and a tribunal can adjust an award by up to 25% for a business that doesn't. The extra length and legal framing doesn't change the fundamentals of what a fair investigation requires. It just makes it more work to get there.
Where things can go quietly wrong
A few patterns worth knowing. AI can inflate or embellish allegations the employee didn't originally intend to make, sometimes without them fully realising it's happened, what's sometimes called validation bias, where the AI's own output starts to feel like confirmed fact. Positions can harden earlier than they would in an informal conversation, because there's been no informal conversation, just written correspondence run through a tool. And there's a genuine confidentiality risk when an employee pastes internal documents or meeting notes into a public AI tool to help draft their case, that's very likely a breach of confidentiality in its own right, separate from whatever the underlying grievance is about.
Can AI actually help run the investigation itself?
Cautiously, in limited ways. Current tools aren't yet reliable enough to transcribe witness interviews word for word, and mistaking one phrase for its near opposite in a transcript is exactly the kind of error that can quietly derail a fact-finding exercise. Where AI genuinely helps is lower stakes: building a timeline across a large volume of documents, or cross-checking human notes rather than replacing them. What it can't do, and won't be able to do for a while yet, is read a room. Whether someone's telling the truth in an interview is still, for now, a human judgement.
What this means practically for a South Yorkshire employer
If a grievance lands that reads like a legal filing rather than an employee describing what happened, that's not a reason to react defensively or escalate too quickly, it's a reason to get the person in a room, clarify what they're actually concerned about in their own words, and investigate the facts underneath the language. Update the grievance and disciplinary policy to reflect where AI now sits, both in how complaints get drafted and in what's acceptable for staff to input into public AI tools during a live process. And where allegations become unusually broad or start touching on discrimination or whistleblowing, get advice early rather than treating it as business as usual with extra paperwork.