You can feel it before you can prove it.

 

The work gets done faster than ever before, plus it’s well written, nicely structured and is free of errors. But something feels a bit off…nothing about your work is “wrong”, exactly, but it sounds like something you’ve read before – like a moment of literary déjà vu – and when you put it next to a competitor’s piece from the same week, the two could swap brand logos and nobody would notice (or complain).

 

That feeling now has a name and – more usefully – a set of numbers attached to it.

 

 

In 2024, two researchers at Exeter University and UCL (Doshi and Hauser) ran an experiment with 300 writers and 600 people to judge their work. Some of the writers were given story ideas generated by AI. Some weren’t given any and had to generate ideas with their own brains (the old fashioned way). The stories written with AI help came back rated up to 27% better written, 23% more enjoyable and 15% less boring.

 

But then the researchers measured how similar the stories were to one another…and things got interesting: the AI-assisted stories were 10.7% more alike.

 

I go deeper into that specific finding, including the exact study numbers, in Does AI make writing more similar? What the Science Advances study found.

 

In other words, they were better…but more the same.

 

I call this the AI Averaging Bias, and it’s the clearest way I’ve found to explain AI content homogenisation (I define exactly what it is and where the term comes from in What is AI averaging bias?). If you run a content team, this is the most useful finding published in the last two years. Not because it tells you something you didn’t already suspect, but because it hands you the evidence for the thing you’ve been trying to raise internally, without sounding like the person fighting to prevent progress.

 

You may not be against the tools (I’m certainly not – I use them every day and so does every team I work with) but there is a key detail in that study almost nobody quotes, which changes what you should do about it on Monday morning.

 

The part that gets conveniently forgotten

 

Here’s how the finding usually turns up in a LinkedIn post: “AI made writing 27% better but 11% more similar.”

 

Accurate, as far as it goes, but it stops about eight words too early.

 

The full version states AI-assisted stories were rated better “especially among less creative writers”. That’s the key point and one the researchers made clear in their study’s abstract. Leaving this part out is highly misleading as it changes the picture significantly.

 

The 27% is a ceiling, rather than the average – these gains belong to the writers the study had already rated least creative, using the maximum amount of AI help available to them.

 

Further, the writers who scored highest on creativity before the experiment started gained around 8% on novelty and 9% on usefulness. Again, these are real improvements but roughly a third of what the weakest writers got.

 

So read those two numbers next to each other, because this is the whole article in one line:

 

AI lifted the weakest work by a lot and the strongest work by a little.

 

 

Unskilled AI use makes your whole team better…but also more average

 

Think about what this finding does to the strength and quality of your team.

 

Your least confident writer gets pulled up towards the middle. Your best writer, the one whose words you’d recognise instantly, also gets pulled up…but only by a little bit (comparatively).

 

The gap between the strongest and the weakest narrows. Essentially, everyone starts to gather in the same quality cluster, but not because anyone did anything wrong, but because that’s the shape AI tools create.

 

That’s why “the average” is the right word for it and why I keep using it, even though it sounds like a criticism of individual people (it isn’t). Nobody is being average – the system is making people average.

 

The thing that makes this issue hard to identify is that it’s completely invisible when looking at only one individual piece of work from one individual company.

 

You read a draft and the draft is good – in fact, it’s better than what that person was producing previously. Everything about your own experience of the work says things are improving…because they kind of are.

 

What you can’t see is the other 99 drafts, written by 99 other companies, which are all converging towards the same middle-ground as you, but from 99 different directions…because everyone is using the same tool to write about the same things in the same way.

 

What AI content homogenisation looks like in a real department

 

“Homogenisation” is a fairly abstract word for something that’s actually very easy to spot once you know where to look.

 

Picture a content team of 14 people, where nobody has been told what to do about AI: there’s no policy, no shared standard and no budget has yet been spent on AI training. What happens over the next 18 months is fairly predictable:

 

• 2-3 people become enthusiasts and get genuinely good at AI, mostly in their own time
• 5-6 use AI daily but in a shallow manner, pasting in a brief and taking the second draft
• The rest use it occasionally and feel slightly guilty about it
• A couple think AI is beneath them and openly say so
• Nobody compares notes, because there’s no forum in which doing so is someone’s job

 

 

Every one of those 14 people is producing better work than they were, the team’s output is faster and the numbers look good on your dashboards.

 

But the 2-3 enthusiasts have, over time, become less distinctive because they’re using the same AI tool as everyone else and nobody ever taught them how to make it sound like them.

 

The 5-6 shallow users are producing perfectly competent work…which could just as easily have come from any of your competitors.

 

And the sceptics, who are often your best writers, are opting out of a tool that would genuinely help them…which means the strongest voices in the building are the ones least represented in what actually ships.

 

This disaster is not immediately apparent. It emerges over time, as a slow flattening of quality, hidden by a good-looking dashboard.

 

Thankfully, all is not lost.

 

The 3 things that actually separate distinctive AI use from average AI use

 

I’ve now run this training with language schools, marketing teams and editorial teams and the same 3 important things come up every time:

 

Important Thing 1. Knowing the AI Averaging Bias Exists

 

This is the most important point, hence why it’s number 1 on this list. A team that has never heard of the AI averaging bias can’t avoid it because they don’t know there’s anything to avoid. That’s why every session I run starts by presenting and defining the AI Averaging Bias.

 

The goal of this is to get people to stop asking “how do I get this done faster?” and start asking “how can we get this to sound more like our team, our department and our brand?”

 

Adjusting your mindset to the latter, rather than the former, will very quickly help you take a huge leap forward in terms of quality of work. It takes a little bit more time and requires more energy at first, but the benefits will start paying dividends almost immediately and long into the future.

 

Important Thing 2. The Things That Made You Unique Are Now Worth More Than Ever

 

An AI model given a generic brief returns a generic draft (junk in = junk out). This sounds obvious, but most teams miss this step as it’s not clear to us exactly how AI tools work (put crudely, they use maths to predict which word comes next).

 

But the main thing to know is that if everything you put in could have been written by any old Joe in your industry, anything and everything it produces will read like it was.

 

The teams who escape the AI averaging bias are the ones putting their own, unique material in:

 

• The client’s actual words from an actual call
• The number from last quarter that surprised the whole team
• The objection that keeps coming up in sales meetings
• The house style guide
• That one thing the CEO always says

 

That’s the raw material no competitor has and it’s the only thing that makes an output unmistakably yours. This is your secret weapon for maintaining your individuality in this sea of sameness.

 

Important Thing 3. Keep The Debates Going

 

This is the trap nobody sees coming.

 

In a healthy team, drafts get picked apart, dissected and inspected to an inch of their life. One person says the intro is too wordy, someone else thinks the opposite and a third person wants to change it completely. This is actually what helps the work get better.

 

However, when AI-created work arrives perfectly fluent, highly polished and ready to go-live, it doesn’t invite the healthy debates which gradually improve the team over time. It only invites instant approval. It sounds right and it looks finished, so the conversation that used to make your work unique to you, eventually disappears and because there is less friction in the team, everyone feels like progress is being made.

 

This is usually the first sign your work is being dragged into the middle, with everyone else’s.

 

Thankfully, it’s the cheapest of the 3 to fix and all you need to do to solve it is to consciously decide to continue debating about, pulling apart and picking at your content.

 

What Happened When I Trained a Group of Language Schools

 

I ran a 6 week AI training programme for member schools of the IALC and it’s the best example I have of what changes when a mixed room gets properly trained to use AI rather than just given access to AI.

 

When we started, everyone was at wildly different levels: some were complete beginners (one person kept calling it “ChatMVP” instead of “ChatGPT”, which I still think is a better name), others were already using AI daily (but without any real idea of whether it was helping them or quietly hurting them) and a few were totally indifferent.

 

When we finished, attendees were building things I hadn’t even taught them. One school built an AI-powered chatbot to answer questions from prospective students. Nobody asked them to (it wasn’t on the curriculum), but they knew their own specific problems better than anyone, so were able to use AI to help them find an above-average solution.

 

 

That’s the bit worth paying attention to and is why I keep telling the story:

By the end of their AI Training Programme, the IALC member schools were building things they hadn’t been taught. That’s when training starts becoming a unique strength your business truly owns.

The reason it worked wasn’t because of something generic AI created. Every one of those schools already had access to the same tools before we started and having this access hadn’t moved them an inch. It worked because the programme was built around the problems those specific schools actually had, in their own words, from their own experiences.

 

This is the whole argument. AI doesn’t make a team distinctive. Trained people using AI with skill and intention makes a whole team distinctive. The tool just does what it’s told (albeit it at light speed), in whatever direction it’s pointed.

 

You just need to know where to point it to get the best results.

 

What To Do First Thing On Monday Morning

 

Here are 4 things you can – and should – do right away, in order of how quickly you can do them.

 

1. Do a blind taste test: Take a recent piece of your team’s work and a comparable piece from your key competitor, cover the names so you can’t see whose is whose, then ask someone outside the team who created what. If they can’t tell, you’ve fallen into the dreaded pit of the AI averaging bias. For 3 more signals to check, see How can you tell if your team’s AI output is becoming generic?
2. Count the specifics: Take 10 articles you published last quarter and count how many contain information only your organisation could have supplied – a named client, a real sales figure, an actual quote from your main customer…anything unique to you. The number is almost always lower than people expect but it gives you a much better measure of how effectively you’ve been able to protect what makes your brand stand out from the crowd.
3. Tell your team about the AI averaging bias: Not as a warning, more as an interesting piece of research (which it is). This will help change your team’s mindset and ensure the conversation is a positive one. It gives the sceptics a reason to engage and the enthusiasts a reason to be more deliberate.
4. Make a conscious decision: are you buying tools or building skill? Both of these are genuinely different things and only one of them will remain profitable in the long term.

 

If your team is already using AI, the question should never be “who is using it the most?”

 

It should always be “is everyone using it in a way that helps their work get better, stay human & fulfils their potential?”

 

Related questions:

• What is AI averaging bias?
• Does AI make writing more similar? What the Science Advances study found
• How can you tell if your team’s AI output is becoming generic?

 

If you’re watching all of this happen and wondering how to make AI work for your business, book a call with me today – I train and consult businesses and teams to use AI with confidence, skill and intention.

 

Reference: Doshi, A.R. and Hauser, O.P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28): eadn5290. DOI: 10.1126/sciadv.adn5290