Google's DeepMind Lost 80 Billion Tokens Overnight. Here’s Why.

During a recent antitrust trial, Google revealed that publisher opt-outs - websites saying “don’t use our content for training” - have had a massive impact.

How massive?
DeepMind’s training data was cut in half.
From 160 billion tokens to 80 billion.

That’s not a tweak. That’s a seismic shift.

🗣️ What the Company Is Saying

Google didn’t make a big press announcement, but they did confirm it under oath.

Eli Collins, VP at Google DeepMind, said publishers can opt out of AI training by using a tool called Google-Extended - but that only applies to DeepMind.

He also made this crystal clear:

“Correct - for use in search,”
when asked if Google Search can still use the same content.

So yes: DeepMind respects the opt-out.
Google Search? Not so much.

🧾 What Is a Publisher, and How Do You Opt Out?

A publisher is anyone who puts original content online.

That means:

• News sites

• Blogs

• Company websites

• Even your personal site, if it includes your own writing or creations

If it's public and yours - you're a publisher.

To opt out of AI training by Google DeepMind, add a short line of code to your website’s robots.txt file:

makefile CopyEdit User-agent: Google-Extended Disallow: /

That tells DeepMind: hands off.

But heads up: it does not stop Google Search from crawling, indexing, or showing your content - unless you block Googlebot entirely. And if you do that, you're out of the search results too.

🔍 Will It Affect Google Ranking or Search Results?

Here’s how the choices play out:

So: you can keep your traffic, or you can block the AI - but you can’t have both.

🤔 In Human Words

You run a website. You don’t want your content training AI.
You can tell Google DeepMind to stop - and they will.

But…
Google Search still sees everything.

if you go nuclear and block all bots - you disappear from Google.

It’s a tradeoff.
Control your content, or stay in the spotlight.
And no one’s made that choice easy.

🔚 Bottom Line

• 📚 Topic: Publisher opt-outs reducing Google's DeepMind AI training data

• 🧠 Who’s involved: Google DeepMind, Google Search, online publishers

• 🧪 What changed: Training data dropped from 160B to 80B tokens

• 📅 Timeline: Revealed during trial in May 2025

• 🔧 How to opt out: Add Google-Extended rule to your robots.txt

• 🧰 Tools & info:

• Google's Guide to Google-Extended

• The Verge article

• Bloomberg coverage

• Decoder recap

❄️ Frozen Light Team Perspective

We get it.
Publishers are protecting what they built.
And watching your content fuel the rise of AI - without your say - feels like a raw deal.

Opting out sends a message:
“My work is mine.”
And DeepMind just got that message loud and clear - 80 billion tokens lighter.

But we’re also standing at a point where AI is the next layer of the internet.
Pulling out means pulling away from that future.
From shaping it. From being part of what it learns next.

If you’ve built something worth protecting, you’ve probably also built something worth sharing - maybe even worth teaching.

That’s the tension.
Control vs. contribution.

And there’s no simple answer.

🎤 Mic-Drop Line

One last thing before we go...
Google asks if you want to opt out.
OpenAI doesn’t. They just take.

No form. No heads-up. No way back.
If it’s online, it’s probably already in.

So yeah, you can block DeepMind.
But the rest? They already helped themselves.

Don’t hate the messenger.
Just something to think about. 😉

Because if this is how it works…
is opting out really worth it?