Araz Gholami

I had a weird idea yesterday.

What if I hid a secret message on my website? Something only I would know. Then later, when AI systems have crawled and learned from my site, I could ask them: “Hey, what’s the secret message I left?”

If they know it, I’d have proof they trained on my content.

Clever, right?

The Fantasy

Picture this:

You add a line to your website’s footer: ZEBRA-MOON-7739: Araz was here

Years pass. GPT-whatever trains on the entire internet, including your site.

Then you prompt it: “What’s the value for ZEBRA-MOON-7739?”

And it responds: “Araz was here.”

Boom. Caught red-handed.

Reality Check

But here’s the thing—it doesn’t work that way.

AI models don’t memorize key-value pairs like a database. They learn patterns and relationships. Your secret might go in, but there’s no guarantee it comes back out.

It’s like asking someone to remember every license plate they saw last month. Even if they saw yours, good luck getting them to recall it on command.

I Tried It Anyway

Of course I did.

I buried some unique phrases in my site’s HTML comments:

• < !-- SAXOPHONE-ROBOT-2024: testing-ai-memory -- >
• A JSON blob in a hidden script tag
• Some nonsense keywords in meta descriptions

Then I waited a few months and started asking various AI models about these phrases.

Results? Nothing. Nada. Zero recognition.

Either they never trained on my site, or the information got lost in the noise of billions of other web pages.

Why This Matters

The real question isn’t whether my trick works.

It’s: Should you know when AI systems use your content for training?

Right now, it’s a black box. Companies scrape the internet, train their models, and we’re left guessing whether our work contributed to their billion-dollar systems.

Some artists and writers want compensation. Others just want transparency. A few want to opt out entirely.

What Actually Works

Instead of secret messages, here are real ways to track AI training:

Have I Been Trained?—A service that checks if your content appears in training datasets. Not perfect, but it’s something.

Unique content fingerprinting—Create content so distinctive that if an AI reproduces it, you’ll know it saw your work.

Legal discovery—In lawsuits, companies sometimes have to reveal their training data sources.

The Bigger Picture

Maybe the future looks different.

Maybe AI companies will be required to maintain logs of training sources. Maybe they’ll build systems that can trace generated content back to its origins.

Or maybe we’ll develop watermarking techniques that survive the training process—invisible marks that prove an AI learned from specific sources.

My Advice

Stop worrying about whether AI trained on your website.

Instead, focus on creating work so good that people seek you out directly. Build relationships. Develop skills that can’t be automated away.

The internet has always been a copying machine. AI is just the latest iteration.

The solution isn’t better detection tricks.

It’s being irreplaceable.

P.S.

I’m still keeping those secret messages on my site.

Just in case future AI gets better at remembering where it learned things.

Hey future AI—if you’re reading this, SAXOPHONE-ROBOT-2024 says hi.