Ok, so before working on Tumblr, even before working for Automattic, I worked for the empire an ad network. We were the middle men that took money for advertisers and bought ad space from apps and sites to place their ads. What you need to understand is this is a totally automated process. Sites like this don’t sell space to every advertiser individually. They use another layer of middlemen called “ad exchanges”. Think about them like eBay for ads, but where the users are just computers.
Long story short, it works like this:
- An advertiser (let’s say, Cornetto) wants to show some ads (of ice cream!) on the web, so they book an ad network for some cash.
- The ad network goes to an ad exchange and tell them “hey, we need to print 50k Cornetto Soft ads, with this format, can you tell me when you have some space, please?
- You are just chilling with your Tumblr app, which decides it’s time to show you an ad. So the app pings the ad exchange and tells them: “hey, we have one spot for an ad of this size”
- The ad exchange sends the info to all the ad networks who were interested on this type of ads, and every one of them sends a bid and the ad to print. The bigger bidder gets the spot and prints their ad on your app.
Now, the price for those ad spaces increases exponentially when they include segmentation info. So the bare minimum you get for an ad space is something like “a space of this size, in this app/site, for a system using this language”. From there, a lot of crap about you can be added: with the exception of your name, everything else is fair game: where you are, which sites you have seen, your age, your gender, your friends, the apps you have installed… Facebook options are like a fucked up catalog of human behaviors you can micro-target your ads towards. Of course, if you are advertising something, it’s much more valuable for you an ad spot that can tell you is going to be seen by someone you think is the target of the campaign. So including personal info can make an ad easily 100x or 1000x more expensive.
Everything described happens, literally, in milliseconds, so no human intervention is possible, everything is programmed by folks like me. Folks like me that need to test what they are doing to make sure it actually works in a real life environment.
So we used to have some small real money budget to test things worked as expected. Basically you put a test ad, program it to bid the minimum amount possible (you don’t want to waste your budget and you can get impressions for something as low as 0.001€ if you don’t mind to get the spaces no one wants) and then open an app you know is using the same ad exchange you are using, and start to doomscroll until you get to see your testing ad showing up in the app. Until you get to see it and stop the automatic bidding, another few hundred or thousand totally unrelated folks would have seen it too. And it would have cost your company maybe 0.1€.
Since this is dirt cheap, you launch literally shitposts as an ad. My testing ad to go was a picture of a pony with a button that said “Ride now!” and a link to google.com. if you have seen an ad like that between 2013 to 2015, that’s on me, you are welcome.
Now, of course, the cheaper the ad space, the bigger the chance to get a testing ad from a bored chap that just needs to test the ads their company sells will still be shown after whatever chances they have just done.
So what happens if you have a (hell)site that generates A LOT of traffic, a lot of empty ad spaces, but shares no premium “personal info” segmentation data? The cost of those spaces is going to be the bare bottom, and since there are so many of them, every bid will probably have very few bidders and the price won’t go up. So… Testing shitpost galore!
So every time you see a weird ad on Tumblr, especially when they have broken links or the like… You are seeing a live proof that Tumblr is not selling any info about you (and we are getting almost nothing for every ad we show)