Is anyone reading your AI reports?
In July, Deezer reported that more than half the tracks uploaded to its platform each day are now fully AI-generated. That’s around 90,000 a day, up from 10,000 at the start of 2025. If, like me, you’re a fan of original music and believe the artist behind it matters, that sounds alarming. You might be relieved to hear that hardly anyone is listening. AI-generated music makes up only 1% to 3% of streams on Deezer, and the company has flagged 85% of those streams as fraudulent. So we have music made entirely by AI, streamed on repeat by bots.
Maybe I’m just the latest grumpy old man who doesn’t understand what the bots are listening to. But like every grumpy old man before me, I’m not giving them that much credit.
Producing music has never been easier, but getting anyone to listen is as hard as it ever was. This might seem a strange place to start a blog post for businesses that aren’t recording studios, but I think the Deezer story raises two points that every business can take heed of.
Is anyone reading what your AI produces?
I want to be careful with this one, because the best way to learn how to get value from AI tools (even better than hiring an AI trainer) is to experiment with them. Find the repetitive tasks in your organisation and see whether AI can take them on. Some of those experiments won’t go anywhere, and that’s fine. It’s how you find out what the tools can do.
The trouble starts when an experiment quietly becomes a routine. Anyone who has spent time with these tools has set up something that looked clever at the time, like a weekly summary, an automated report, or a daily digest of the inbox. If we’re honest, some of that output isn’t driving anything. Nobody reads it, or people skim it out of politeness, and no decision changes because of it.
At the moment that probably isn’t costing your business much, but AI may not stay this cheap. Axios reported in March that some of the big AI companies are still losing heavily to win customers, with OpenAI projected to burn $14 billion this year. When prices rise, every workflow nobody needs gets more expensive. There’s also a cost you’re already paying: someone has to check AI output, and checking a report nobody reads wastes their time twice.
Try this today. List every task AI does for you or your team on a regular basis. For each one, ask who reads the output and what decision it changes. If nobody can answer, switch it off and see whether anyone notices.
If you’re wondering how this post ended up on Jaws, I’m guessing this painting of Quint in my living room primed me.
Then there is the Jaws problem.
When people talk about checking AI output, nearly all of the conversation is about accuracy: whether the facts are right and the sources are real. That matters, and there are plenty of examples of companies caught out by AI-generated content. But there’s another part of checking AI output that gets far less attention, and the best illustration I know is the theme from Jaws.
When John Williams first played Spielberg the two-note theme, Spielberg reportedly thought he was joking. If an AI had produced one of the most iconic themes in film history, I strongly suspect Spielberg would have turned the effort level up and hit regenerate. Williams, however, could explain why two notes would work, and he was prepared to defend them.
Every business relies on people making decisions, and the good ones sometimes make calls that look wrong on paper. Someone who knows the business and its customers will occasionally back an idea that goes against conventional wisdom, the way Williams backed two notes. AI tends to pull the other way. It gives you the most likely answer based on everything it has seen, which makes it very good at telling you what most businesses would do. Follow that advice every time, and you’ll perform like the average of all of them. Not every unconventional call pays off, but a business needs people who know it well enough to take calculated swings, and the confidence to overrule the AI when they do.
Deezer gets 90,000 AI-generated tracks a day, and almost nobody wants to hear them. A business that runs on AI’s default answers is competing to be number 90,001.
When output is close to free, the valuable part of the job is deciding what’s worth making. Most teams I work with already have that judgement. What they need is the habits and confidence to use it. If you’d like help with that, I run practical AI workshops for non-technical teams across Ireland. Book a free 20-minute call, and we’ll work out where to start.