The Dark Revenue Problem: Publishers’ €10 Billion Blind Spot

Publishers are generating roughly €10 billion in AI training value every year.

They can’t see it.
They can’t track it.
They definitely can’t invoice it.

I’ve said before that amounts to stolen revenue, but that’s only half right.

Rather, it’s dark revenue; value created from their content that never shows up on a statement.

Like dark energy or dark matter in physics, we only know it exists because of what it does to the system around it: AI products that mysteriously talk like seasoned journalists, or produce excerpts verbatim from paywalled content.

I’ve had hundreds of conversations about this with publishers, AI engineers, and media execs. Two in particular stuck with me.

Both were with major financial publications.

We spoke under Chatham House rules, but you’d recognise them instantly.

I asked them both the same question:

“How do you know when your content shows up somewhere it shouldn’t? And what do you do about it when you see it?”

What I wasn’t expecting was both to give almost identical answers.

Someone on the team happens to see their paywalled story on another site.
Or a reporter pastes an AI chatbot answer into Slack and says, “I’m pretty sure this is ours.”
Maybe a subscriber emails them a screenshot.

That’s the “monitoring system” at some of the most sophisticated newsrooms on earth: luck, screenshots, and the kindness of strangers.

No systematic way to discover where their work appears online or inside LLMs.
No internal playbook for what happens next.
No budget to fix it, because you can’t price a problem you can’t measure.

If that’s where they are, imagine the rest of the industry.

This is the dark revenue problem. And it’s where publishers are fighting the wrong war.

They’re trying to block AI, when they should be building toll roads.

Everyone points at Spotify and says, “Music solved piracy. We’ll just do that.”

Not quite.

If it were that easy, everyone would do it.

Music has a clean attribution unit. When you stream a track, one work gets credited. Anyone can understand who wrote “Yesterday”, and recognise a cover version.

Text in AI is a blender.

Ask a model for an explainer on climate risk in banking and the answer might pull from:

Hundreds of fragments. Maybe thousands. One answer.

Who gets paid?
The outlet that broke the story?
The one that explained it best?
Everyone? No one?

This is why “just do what Spotify did” does not transfer directly.

It’s more of an analogy form which to draw inspiration.

What is useful from music is the sequence:

ASCAP didn’t wait for the internet or streaming. They built detection and collection mechanisms when radio was the supposed existential threat. By the time the format changed, the rails were already there.

Publishers are trying to skip the boring step. They want streaming royalties without stream-level infrastructure.

That’s not strategy. That’s a wish.

The late, great Charlie Munger liked to say: “Show me the incentive and I’ll show you the outcome.”

Right now, AI companies are looking at three options:

No prizes for guessing which one is easiest to execute.

The first option is cheap and fast.
The second is expensive and slow.
The third only exists in a few verticals.

So they scrape, settle, and move on.

What they’re absolutely not going to do, which some in the industry appear to think, is they’ll patiently sit on their hands and hope “fair use” magically stretches to cover everything.

No.

Suing them might feel satisfying, but it doesn’t change the incentive structure. The music industry tried the “see you in court” strategy for a decade. Nobody looks back on that era and says it was their finest commercial decision.

What actually changed behaviour was when legal access became cheaper and simpler than cheating.

That’s the real job here: create conditions where paying publishers is the low-friction, rational move.

Litigation changes the risk calculation at the margin, sure – the New York Times suit against OpenAI and Microsoft, the French cases against Meta, and the $1.5B+ Anthropic settlement all matter.

What those cases don’t supply is a way to pay thousands of publishers consistently.

For that, you need something blunter: a setup where paying for clean, trackable content is cheaper, safer, and simpler than gambling on scraped data.

Not because anyone suddenly finds their conscience, but because the spreadsheet says so.

To get that flip, you need industrial infrastructure. Boring, but effective.

Toll roads don’t stop people driving, they simply charge those who use the road most.

To change IP incentives, publishers need one boring, essential thing:

Digital fingerprints for text.

Every article, column, and long-form piece needs a durable machine ID and ownership metadata that can survive paraphrasing, summarisation, and blending.

When an AI system trains on, retrieves, or remixes that content, those fingerprints should show up in the exhaust:

Think of it less like DRM and more like metering.

Not a wall to keep everyone out, but a turnstile that counts the people coming through.

Once usage is trackable, a lot of “impossible” conversations become straightforward commercial questions:

That’s the infrastructure gap. Until it’s plugged, dark revenue stays dark.

Here’s the part that should bother anyone who signs budgets or sits in a board meeting.

Deals are already being signed.

While most of the industry is still arguing about prompts, AI companies are quietly locking in supply deals with whoever can deliver:

Sometimes that’s a publisher.
Often it’s not. It’s a platform, a data broker, or a weird aggregator you’ve never heard of.

Those early contracts are setting the reference points for value, establishing the baseline economics for the next decade:

If you arrive after those numbers are already baked into product roadmaps and budgets, you’re not a strategic partner. You’re now a line item.

If you work in publishing, this isn’t a think-piece. It’s a checklist.

This quarter, you still have time to:

If you’ve read this far, you’re already ahead of most of the industry.

You’ve seen the outline of the missing money, the dark revenue is already there.

The question is whether you’re going to help design the meter, or wait for someone else to hand you their version of the bill.

If you want help with any of this, or simply have a question, feel free to book a call or reply to me directly.