Newsbox Hunter: Using Street View and AI Vision to Find Abandoned Creative Loafing Boxes (Poorly)
Newsbox Hunter: Using Street View and AI Vision to Find Abandoned Creative Loafing Boxes (Poorly)
I had a simple question. So simple, really. Are there still Creative Loafing newsboxes sitting on Atlanta sidewalks, rusting away, years after the paper stopped printing? Those distinctive green boxes were everywhere intown for decades, and some of them have to still be out there. So I built a tool to find them. This one's part of the vibe-coded series, and honestly, it's as much a love letter to a paper that shaped how a lot of us understood this city as it is a technical project.
What does the tool actually do?
It pulls Google Street View images across Atlanta's bar and restaurant corridors, runs each frame past Claude's vision API alongside reference photos of an actual CL box, and flags anything where the logo is still legible enough to be a real match. The whole thing is a Python script, a Google Maps API key, and an Anthropic API key, nothing more elaborate than that.
I love exploring these tools but I have read horror stories about API calls going haywire.
Where did the first approach go wrong?
I started by scanning Little Five Points with a straightforward grid of bounding boxes, which worked but burned through a lot of API calls scanning empty blocks and parking lots that were never going to have a newsbox on them in the first place. Sure, I could spend $10,000 scanning every single street in the city, but that doesn’t seem like a good use of funds.
From there I tried to recover Creative Loafing's actual old distribution list from the Wayback Machine, since presumably that would tell me exactly where the boxes used to be. The page itself archived fine, but the map pins didn't, since that location data lived in a live database that was never actually captured in the snapshot. I did find the paper's old Google Maps API key sitting in the page source, which was a fun little artifact of internet archaeology, but obviously not something I could use for anything.
What actually worked?
Querying Google Places along the specific intown corridors where Creative Loafing was always distributed, streets like Euclid, Moreland, North Highland, Ponce, Flat Shoals, Edgewood, and the Decatur square, and pulling every bar, cafe, and music venue along them as scan targets. Current businesses standing on the streets where the boxes actually used to live turned out to be a much better proxy than an empty grid scan, since a newsbox that's still out there is far more likely to be near a business that's been in the same spot for a long time than out in front of a random address.
I was explicit in asking to make sure it gave me an estimated cost before it ran a search.
Ok hold up there partner, no thank you
How does the actual detection work?
The prompt sends three reference photos of a real CL box, a side view, a front view, and an angled shot, alongside each Street View frame, and asks Claude specifically to match the logo. Anything where the branding is too faded to confirm gets filtered out rather than counted as a hit. Results plot onto a map in a custom viewer with a direct Street View link for each one, so you can actually verify a potential find before going out to look for it in person.
I tried to find some good images straight on, from the side, in different contexts.
Even spray painted or a little messed up seemed fine to me.
Sure! Why not a Tampa one too.
How accurate is it?
Ok so this was my favorite part that really calls attention to how weak these tools can be.
The tool would give me a more user-friendly view of the results and open a direct street view link to where it was spotted.
Oh look, a HIT!
I think this orange box is what it spotted – but it’s just a box.
Oh look, another HIT!
Or, another miss. I’m not even sure what this is describing. It seems to be describing the box in the other image.
Street View imagery can be years out of date, and Claude occasionally flags something with similar geometry, a green recycling bin, a utility box, that isn't actually a newsbox at all. But as a starting point for an actual walking survey of these corridors, it's genuinely useful, it narrows a whole city down to a specific, checkable list instead of just wandering and hoping.
FAQ
Does this find newsboxes in real time? No. It works from existing Street View imagery, which can be years old, so a hit is a lead worth checking in person, not a live confirmation.
Why not just search Creative Loafing's old distribution records directly? The paper's old distribution map lived in a database that was never captured by web archiving tools, so the underlying location data no longer exists anywhere accessible.
Does it only work for Creative Loafing boxes? The detection approach, matching Street View frames against reference photos with an AI vision model, could work for any distinctive, recognizable street object. This particular build is tuned to CL's specific box design.