Trying to Find Stolen Bikes for Sale on the Internet

06 Nov, 2025

Bike Theft in London:

It doesn't take long on Reddit to come across countless posts about bikes being stolen in London:

LondonCentric (an excellent London specific newsletter) has produced a number of reports on bike theft, this from September 4th '25 is particularly striking:

(search for “Bike theft is practically legal” on https://www.londoncentric.media/p/more-chance-of-a-lottery-prize-than)

I'm a keen London cyclist and have invested in a "LITELOK X1" to try to protect my bike, but we seem to be in a rather wild situation where you have to purchase a lock more expensive than many second-hand bikes, and even then - thieves can cut through bike stands and just walk off with your bike:

I do love my bike, so I've even signed up to BackPedal, a very cool bunch of folks who install a GPS-tracker in your bike and have a high bike retrieval rate. It's worth watching their videos:

The site Stolen Ride has some decent tips about how to lock your bike.

So, leaving your bike on a London street, even with two expensive locks (as Stolen Ride suggests) and a GPS tracker installed still feels like playing a game of chance - will your bike be there when you get back? If it's gone, will you get it back? Is it actually worth having a decent bike in London? Play the game, leave your bike, roll the dice and find out! 🫣 🎲 🚲

Finding Stolen Bikes:

All of this got me wondering - once a bike gets stolen, what happens to it? From the reading I've done:

They might get resold on sites such as Cash Converters, of course Facebook Marketplace and GumTree
They might get shipped abroad
They might get broken down into parts and then resold.

And this led me to thinking - I've just done some experimenting with a neural network capable of comparing and labelling images (Fiddling with the Open AI Clip Neural Network), so might it be possible to identify stolen bikes for sale on the internet?

Sources of Images:

A list of stolen bikes at Bike Register
And many bikes for sale on sites such as Cash Converters, GumTree and of course Facebook Marketplace

Results of Scraping:

Stolen bikes: I managed to download details of 59,049 stolen bikes (via a semi-automated process); 31,751 with images (53% of all downloaded details of stolen bikes)
Bikes for sale: I managed to find a way to download a few from Cash Converters and Facebook Marketplace: 771 - all with images. I didn't bother with GumTree as I just needed to download some for sale from somewhere to try this out.

Facebook Marketplace is the largest source, but from what I've read somewhat difficult to scrape, so I've got another fairly semi-automated manual process - hence the low numbers of total bikes for sale.

What Did I Build?

So I managed to build something that goes through every one of the 771 bikes for sale and do an image comparison with the 31,751 stolen bikes with images.

I tweaked the comparison score so that it's a combination of:

The OpenAI’s CLIP similarity score - weighted at 70% of the overall comparison score
Similarity between the bike's names - weighted at 15% of the overall comparison score
Similarity between the bike's colour - weighted at 15% of the overall comparison score

For each bike for sale, the final report flags up the top five stolen bikes with an overall comparison score of over 89%.

Some Example Results

With all these images below please do click on them to see a larger higher-resolution image.

First Example Comparison:

This appears to be a very good comparison: same brand (Carrera), same colour (dark grey) and what does appear to be a very similar image comparison. Certainly looks like a solid contender for a stolen bike being resold.

Second Example Comparison:

This again appears to be a pretty solid comparison matching brand name, type (commuter bike), colour - and initially a pretty good image comparison. However, there's one or two minor differences (mark on the frame, wheel colour is slightly different) that makes you question the match.

And Third Example Comparison:

So, here's an example that hasn't worked that well. There's a match with the brand (Rockrider) but not the model, the type is different ("BMX Bike" v "Mountain Bike"), and although CLIP has identified similar colours the colours are actually different. Visually though, the bikes do look somewhat similaar.

So Did I Find a Stolen Bike for Sale? And What Does All This Mean Anyway?

Did I find a stolen bike for sale? Honestly, not sure. Some of the comparisons seem remarkably good and possibly worth investigating further, so it's possible I've found one (I have tried to contact an owner via Bike Register) but it's really for an owner to say for sure that it's a match.
The image matching shows real potential. There's work to be done to really hone it to perfection e.g. colour, brand, type, state of repair identification could be improved. I'm sure someone who's a good data scientist with neural network expertise could work some magic.
I'm also pretty confident that if one could wave a magic wand and have access to all bikes for sale in the UK this technique would be able to flag up stolen bikes for sale. It would still require human intervention to validate the analysis though ....
The hard part of building something that proactively tries to find stolen bikes would be the part regularly scraping sites such as Facebook Marketplace and GumTree at significant scale. I'm sure those sites won't want to be scraped, and I doubt very much that they'd be interested at providing any kind of feed.
And then there's the question of avoiding a false positive match thereby potentially accusing someone completely innocent of selling a stolen bike?

A Possible Next Step?

So, perhaps a first step could be if sites such as Bike Register or Immobilise could provide an image search so that when someone is buying a bike online they could check to see if a bike for sale is possibly stolen? Something that could flag similar results to those shown in the above examples e.g. % image similarity, name, colour, brand similarity ? This alone would be quite helpful.

Useful Links

What's Next for Me with This?

I would like to see if there's any interest to take this further in whatever form that might take. It would be great to do something about bike theft in London.

I did try to contact the people at Bike Register for any comment on the contents of this post, but sadly not yet received any reply.