Eyeweb
Unplugged:

Why AI Might Be Making the Internet (and Us) a Little Less Trustworthy

Our conversation started with a YouTube video that looked into how accurate ChatGPT really is. The findings were surprising: around one in five “facts” it gives can’t be traced back to a verifiable source. Twenty per cent of everything it says could be complete fiction.

We all know AI can make things up, but that’s still a pretty big number. Imagine if your washing machine only worked four days out of five. You’d probably start double-checking your laundry schedule.

The Copy of a Copy Problem

Even the 80% of information that is technically “accurate” might not be as solid as it looks. That’s because a lot of what AI is trained on now comes from content that’s already been written by AI.

It’s like photocopying a photocopy over and over again. Each generation gets blurrier. When today’s models train on yesterday’s AI-written blogs, the errors start to multiply and soon no one’s quite sure what’s real anymore.

When the Internet Becomes Its Own Source

The problem isn’t just that AI makes mistakes. It’s that it learns from places where accuracy isn’t exactly guaranteed. Reddit threads, forum posts, and random blog comments all feed into the dataset.

That’s why AI sometimes quotes “facts” from sources like MrKillerSlasher79 on Reddit as if they’re peer-reviewed science. And because so many people use AI to write, that same phrasing and style is slowly leaking into things like research papers, speeches and even political statements.

We’re starting to sound a lot like the bots we built.

SEO Broke the Internet (and AI Might Finish the Job) 

Long before ChatGPT, Google search was already creaking under the weight of SEO. The internet became a race for keywords instead of ideas. Everyone was writing the same blog post in slightly different words.

Now, AI summarises those same pages, meaning it’s distilling content that was already written to game the algorithm. And the newest twist? “Generative Engine Optimisation” (GEO): writing for AI tools instead of human readers. 

If SEO blurred the line between helpful and hollow, GEO risks erasing it altogether.

Why We Still Need Real People

All of this makes one thing very clear: human creators matter more than ever. Good content still relies on curiosity, originality, and critical thinking —things AI can’t truly replicate.

If everyone uses AI to write everything, and that work becomes the data for the next generation of AI models, the internet becomes a giant feedback loop. It’s the blind leading the bot.

Writers, designers and marketers have a responsibility to keep the human element alive, to fact-check, question and create new ideas instead of recycling old ones.

The Tipping Point

We’re already starting to recognise when something feels too polished or too predictable. AI voices, AI stock photos, AI-generated posts… people can tell. And they’re craving something real again.

That doesn’t mean AI is bad. Used well, it’s an incredible tool for ideas, drafts and speed. But it’s not a replacement for thinking. As one of us said during the conversation: “It’s brilliant for getting started, but I still need to check every bit of it.”

Finding the Balance

Maybe that’s where we’re heading: a balance between automation and authenticity. A world where AI helps us work faster, but not lazier.

The internet was built on people, on their thoughts, creativity and mistakes. If we want it to stay trustworthy, we need to keep those people in the loop. Otherwise, we’ll end up living in a beautifully designed landfill of half-truths.

And no one wants to scroll through that.