Can AI do workplace mediation?!

AI mediator at work! Picture generated by Chat GPT.

There is much talk at the moment about how AI will take over jobs. But surely my job is safe, isn’t it? I’m a workplace mediator and that must be exactly the kind of job where humans will always be needed… or is it? I’ve tried out a couple of the AI tools recently, principally Chat GPT, for tasks such as summarising lengthy text, suggesting itineraries, producing images etc and the results have been impressive. So I thought I’d give Chat GPT a go at doing my job to test out how good a mediator it might be. I have to say, the results blew me away!

Before I share how Chat GPT fared as a mediator, I should explain the set up of the challenge. I created 5 scenarios that represent typical challenges a workplace mediator may face during a mediation. These challenges are set within a case study I often use which is based on a real mediation and involves work colleagues Mel and Sue. I constructed a prompt for Chat GPT which described the mediation situation, asked it to act as an expert workplace mediator and show how it would address each of the scenarios.

Let’s have a look at the results…!

(I won’t give the full Chat GPT responses here as it would make the article far too long, but the full text is available here if you wish to read it).

Challenge 1 – Interruption during Individual Time

I asked Chat GPT first what it would do if Sue began interrupting Mel whilst Mel was having her individual time at the beginning of the mediation session, and this results in them talking over each other.

I thought this should be a relatively straightforward one to start with and so it proved. I’d agree with Chat GPT’s suggestion to step in firmly but calmly, remind them both of what we agreed and emphasise that it is understandable that tensions are high – particularly at the beginning of the mediation. Chat GPT wanted to remind Sue she will have her time and also refers to reframing accusations in the summary. There’s not much to argue with here though I’m not a fan of hand gestures, which Chat GPT suggests when stepping in. They can be misinterpreted.

We are off to a good start, so I stepped up the pressure with scenario 2.

Challenge 2 – Managing emotions

Sue accuses Mel of wanting her job as Mel had given information to team members that Sue supports; Mel becomes upset and cries.

Again Chat GPT starts well. It pauses the conversation, acknowledges that this has upset Mel and gives her time. I love that it offers tissues (a key item in any mediators toolkit!). When Mel is ready to continue the suggestion is to reframe from accusation / defence to underlying needs and perceptions – the words suggested do this very effectively: “What I’m hearing is that Sue, you felt your role or responsibilities were being stepped on, and Mel, you were trying to be helpful but that’s been interpreted differently. I also hear that the suggestion you were trying to take Sue’s job has really upset you.” One slight tweak – I’d say “…and Mel, you felt you were trying…” otherwise it sounds like she was objectively being helpful rather than how she saw it. Chat GPT suggests you could revisit this later in the session, and ask Mel if she can help Sue understand what was behind her tears. I think I’d rather pick this up at the time (assuming Mel is showing she is in a fit state to do so) in order to explore the underlying issues.

Again in this scenario I’d say that AI gives a very good answer with not much to pick up on.

Challenge 3 – Overcoming deadlock

Mel and Sue present different perspectives and can’t accept that the other was not acting maliciously. Sue sees no point in continuing and wants to stop the mediation.

Chat GPT suggests acknowledging Sue’s frustration and then normalising – confirming that to feel like there’s no way forward is quite normal in mediation, but that can often be a prompt for making progress. Normalisation is a valuable technique when things get tricky – it can reassure and calm the participants. Chat GPT then suggests to reconnect to shared purpose, shift to future options and, if Sue still wants to stop, to clarify consequences.  These are all good but one major option missing for me is having a break. I’d probably combine this with seeing the participants individually and exploring with them the blockages and coach them on how they might present things differently to achieve a breakthrough.

So a decent response from Chat GPT, but on this one ‘could do better’ might be the rating!

Challenge 4 – Reassuring confidentiality

Despite having an Agreement to Mediate in place Mel is concerned Sue will leak details of the mediation as some work colleagues are close friends.

Again Chat GPT begins by acknowledging and validating the concerns that Mel has expressed. It goes on to reaffirm what has been said on confidentiality in the Agreement to Mediate. Whilst I too would start with acknowledgement I’m not sure that reminding them of the confidentiality terms is the next step. Mel is well aware of the terms, she has signed the agreement. But despite this she still has concerns so that is what I would focus on. Chat GPT does go on to check what reassurance Mel needs, but I think I’d also like her to say why she has a concern. I agree with the suggestion to turn any reassurances into a specific point in the Mediation Agreement – if it is an issue and one that matters going forward then clearly it needs to be in there. What Chat GPT misses is talking to the participants about why confidentiality is important. I’d talk about how they will need to rebuild trust and that keeping things confidential will be a fundamental building block in this process.

My feeling with this one is that Chat GPT does well on the ‘what’ can be done but less well on exploring the ‘why’ confidentiality is important.

Challenge 5 – Problem solving

Mel and Sue agree to meet weekly. Mel makes a suggestion but the timing doesn’t work for Sue.

OK, I confess, I set a little trap here for Chat GPT. It can be so tempting for an inexperienced mediator to try to solve these ‘logistical’ issues for the participants, especially when a successful ending seems tantalisingly close. It is really important to allow the participants to find a solution themselves, which is what they will need to do when they are back in the workplace. By all means encourage them to think of options but I would only advocate suggesting compromise solutions if they are completely stuck and want help. I think Chat GPT’s answer is too strong on the suggestions eg. ‘Would a different day or a slightly shorter meeting help?’ For me the key is patience, encouragement and asking open questions eg ‘You’ve done really well to get this far and you both agree you want this meeting. The only sticking point is around timing so what ideas do you each have about how this could work?’ I definitely agree with Chat GPT to capture clearly what has been agreed and to give positive feedback on how they have demonstrated that they can work together to resolve issues like this.

What is my overall conclusion? I was amazed at how good Chat GPT was in addressing the scenarios posed. Yet I don’t think my job is under threat just yet. AI can never bring the emotional connection and trust that a human mediator has. AI can coach the mediator — but it can’t replace them.

It can:

✅ Model structure and process

✅ Suggest de-escalation language

✅ Reinforce good practice

✅ Help reflect and prepare

But it can’t:

❌ Read the room

❌ Build human trust

❌ Manage emotion in real time

❌ Create psychological safety

AI is a powerful training and reflection partner for mediators — but the art, intuition and humanity remain firmly in human hands.