Spotting AI-generated junk sites in the wild

I spent twenty minutes yesterday reading a recipe for sourdough starter that eventually pivoted into a lecture on the geopolitical stability of salt mines. The site looked professional enough at a glance. It had the clean white space and the high-resolution photos of dusted flour we have been trained to trust. But the text felt like a fever dream filtered through a corporate HR manual. This is the new baseline for the internet, a landscape littered with digital strip malls built by scripts and populated by ghosts.

You can usually smell the rot before you find the first hallucination. Look at the images first. AI-generated junk sites rely on stock photography that feels too expensive for the quality of the writing. They favor those hyper-saturated, uncanny valley scenes where everyone is smiling at a salad or a laptop with a fervor no human has ever felt. If every single image features a person with exactly thirty-two visible teeth and no pores, the site was likely assembled by a bot in a dark server room.

Practical skepticism requires looking at the clock. Check the publication dates against the content itself. You will often find a site claiming to review the best tech of 2024 with a footer that says the site was founded in 2011, yet the earliest post is from three weeks ago. These sites are frequently built on expired domains to steal the original owner's reputation. If the site claims to be a decade-old authority but has the archival depth of a puddle, keep moving.

The paragraph structure is the loudest giveaway. Human beings are messy writers who get distracted or excited. We write one-sentence zingers and then follow them with long, rambling explanations that shouldn't pass an editor's desk. AI is pathologically polite and uniform. If every paragraph on the page is exactly four lines long, you are reading a sequence of tokens optimized for an algorithm, not a person. It is the visual equivalent of a dial tone.

Search for a name. A real person usually wants credit for their work, or at least a fake persona with a convincing backstory. Junk sites use generic bylines like Staff Writer or Admin. If there is a name, click it. If that author has published forty-five articles in the last twenty-four hours across topics ranging from lawn mower repair to crypto-mining, they do not exist. They are a line of code wearing a human mask.

Finally, check the receipts. These sites love to cite studies that sound plausible but evaporate under a basic search. They will mention a 2022 University of Zurich study on caloric density that simply does not exist. They count on you being too tired to check. If the link just leads back to their own homepage or a dead 404 error, the information is fiction. Trusting these sites is like asking a magic mirror for medical advice; it will tell you what you want to hear until your hair falls out.

The internet is becoming a library where the books are written by the shelves. We are currently in a war of attrition against filler content designed to harvest your attention for three cents of ad revenue. You have to be a difficult reader. If the prose feels like it was written by someone who has never actually touched the object they are describing, it probably was. Stop giving your time to machines that don't know how to do anything but repeat the last thing they heard.