April 30, 2026
Booked, bothered, and ratioed
I scraped 1.94M Airbnb photos for opium dens, pet cameos, and messy kitchens
A giant Airbnb photo scrape turns into a comment-section brawl over ads, ethics, and "drug-den vibes"
TLDR: Someone analyzed nearly 2 million Airbnb photos and thousands of reviews to find strange patterns in rentals, but commenters quickly turned it into a fight over ethics, branding, and sloppy labels. The big debate: clever internet experiment, or creepy ad campaign with classist vibes?
A developer went full internet detective, scanning 1.94 million Airbnb photos and thousands of guest reviews to hunt for everything from pet cameos to supposedly sketchy interiors. On paper, it sounds like a weirdly irresistible data project: public listings, public reviews, lots of number-crunching, and a parade of bizarre findings. But the real show started in the comments, where readers basically yelled, "Wait, is this research... or an ad?"
That became the main event. Multiple commenters were instantly suspicious of the giant Burla branding, calling the whole thing content marketing dressed up as quirky analysis. One person bluntly said it was "plain and simple" promotion, while another dragged in Inside Airbnb's guidelines and asked whether scraping all this data crossed a line. Others went even further, tossing around the L-word: lawsuit. Not exactly the cozy vacation vibe Airbnb usually aims for.
Then came the methodology meltdown. Readers were especially annoyed by labels like "drug-den vibes," saying the project seemed to confuse dim lighting, clutter, or plain old poverty with criminality. One commenter was horrified that the site treated miserable or unsafe guest experiences as the "funniest" reviews, asking the obvious question: where, exactly, is the joke? So yes, people were fascinated by the strange findings — but the bigger entertainment was watching the community roast the project for being tacky, shaky, and maybe a little too pleased with itself.
Key Points
- •The article says it examined every public listing in Inside Airbnb's open dump across 119 cities and four quarterly snapshots.
- •It states that 1.7 million listing photos were scored with CLIP and suspicious images were rechecked with Claude Haiku Vision.
- •The article says every review was scored and the 12,000 strangest reviews were reranked with Haiku.
- •The compute workload was run on Burla using a dynamic cluster that scaled to about 1,700 CPU workers and 20 A100 GPUs.
- •The article says its findings cards use bootstrap 95% confidence intervals on 365-night calendar occupancy as a demand proxy.