cliffordbanes

"Casey Liss:

So I have a question. So one of the things you mention in your blog post is tracking pixels effectively invisible but when Overcast or some other client goes to get that image to display in show notes for example that inherently passes some information about that user to the server that is serving that image. So it's kind of a backdoor to getting a little bit of information about users without their consent. I have heard of tracking pixels and I've seen it used plenty of times on the web. I had no idea that this was a thing in the show notes for podcasts. I mean it makes sense. I'm not saying it's unreasonable from a technological perspective but I had no idea that was a thing. How did you even find out about this?

Marco Arment:

Well you know I follow the podcast industry news and tracking pixels have been a thing in RSS feeds for a long time. I think feed burner even offered them to the masses but there has been various feed hosting and analytics platforms and packages that have offered tracking pixels for a while. The podcast industry, I wrote about this in the post, the fundamental behavior of a podcast player is that no publisher can break without cutting off a large portion of their audience is that they publish an RSS feed.

In that RSS feed are URLs of audio files and the podacast player at some point if you are subscribed in your podcast app, at some point the podcast app downloads that audio file that is listed in the RSS feed for the episode that you are going to listen to. And then after the publisher hears nothing from the player so the kind of information void here is the publisher of that podcast all they know is that they served a copy of this audio file to a person at this IP address at this time, that's it. They don't know if you've listened to it, at all, if you do they don't know, if you've listened to at all, you could have downloaded it to your phone and listened to it six months later. They don't know if you've listened to the first five seconds and then deleted it or never went back to it. They don't know if you've skipped the ads which a lot of them want to know. They don't know how far you got. They don't know if certain segments of the show are boring to people and they always skip over them although what they really want to know are ads. They want to know content decisions they wan to

As podcasting becomes a bigger and bigger business the big players in the business are people coming from the big content world, from the web, from TV, from radio, etc and they are used to the kinds of things they can do on web pages. If you are running a web publication you can track everything about what people can do while they read.

You can track how long they spend on the page, how far they scroll down the page, how quickly they scroll, whether they are that kind of person who highlights text as they read or not and what parts they highlight whether they click around the page.You can measure so much and then you can embed tracking codes with cookies and things like that that follow those people around the web so you can say not only does this person click around when they read but this is a person that a little while ago went to amazon and looked up patio chairs and now they are over at my site reading about sunglasses and it sounds like they are maybe going to the beach.

You can make those kind of correlations and then later on the things that your website tells Facebook about them can be used to advertise to them on Instagram about the article they read about on your site three hours ago. So it's this whole crazy web of data and also this last for things like they can algorithmically generate articles. Bots can generate articles that they know people will like based on previous data. They can analyze how those articles perform in real-time and they can A/B test different arrangements of sentences and paragraphs, different kind of constructions, different topics and they can A/B test that to generate even more articles that are even more effective later and podcast publishers want to bring that kind of world to podcasts.

Now I should clarify. Not all podcast publishers. Not even most podcast publishers. A few of the biggest ones. Like the biggest ones that have massive companies, massive budgets, massive listener counts. Big Podcasting basically. They want more data so they can sell more accurate ads, more targeted ads for higher rates to big brand advertisers like Coca Cola who want that kind of data. That's what they want. Nobody like us wants this. Oh, and also the big publishers want to be able to make content decisions based on that kind of data so they wanna know what oh, if we structure the intro this way and if we do this segment before this segment listenership drops 4 percent but if we re-arrange them then the data shows that this is better this way because people will hang on 2% longer time or something like that.

And what that leads to is in my opinion what the web has now which is a lot of spam and a lot of garbage content, a lot of micro managing of creative content and creative people and consumption of that content by data people, a huge amount of privacy invasion and tracking that is creepy and that people don't want or like or even know about. And just a kind of general cut-throat, bargain basement, penny scraping ad market that right now podcasting has none of these things and podcasting is awesome right now. And we've gone like decade having none of these things so I don't know why anybody wants this, well I know why they want it but I think they are misguided and wrong.

Because the big publishers want this and because podcasting is hot and there is a lot money being made here. Both the big publishers want and also there's a ton of ad-tech companies that is trying to break into podcasting. Basically they smell money and they are like 'oh, we'll use the tools we have to expand into this market' so the tools that ad-tech has are tracking, a/b testing and stuff like that. This is why a lot of podcasts will serve you local ads. It's just another part of this movement where they literally inject ads into the podcast at the moment you download it. It's tailored to your IP address and anything else they know about you from that. And ad-tech companies are selling all sorts of wonderful solutions to podcasts publishers saying 'we can solve your data problem, we can track people past the download with our proprietary technology'. And most of it is total BS, most of it is totally like they are not even knowing anything here. They are not really getting any data or they are getting data from very few people.

So I'm almost on the look out whenever any ad-tech company says we've solved podcast data, we can tell you more about what your customers are coming from and how far they are listening and everything. It all boils down to a few small things they can do.

Number one they can embed images in the feed, they can embed tracking pixels and they can assume if someone downloads the podcast and then the tracking pixels show that they got they can assume that person is playing the episode and has viewed the show notes. And then they can extrapolate from that what percentage of people are viewing that or whatever and that's fine.

And number two they can try to slow stream the file to you basically to try and measure how far you've listened. Instead of sending you the file as fast as you can download they if they can tell that you are streaming it whether or not you are in a web browser or an app they control or if they can somehow figure out through heuristics that you might be streaming instead of downloading for some reason. They will serve it to you just a little bit a head of the play head so that if you cut off the stream after listening for five seconds it hasn't served you much more than that so they know how much they've served so they know how far you've listened.

That's basically every podcast ad-tech thing. It's dynamic ad insertion, tracking pixels in the body and slow streaming files. That's about the extent of it. And again there's lots of effort to change this. I don't know how that is ever going to get off the ground but it's fine.

I should also point out by the way that a few months ago Apple launched podcast analytics. Forever now the podcasters have been begging Apple to give them more data because Apple is the biggest podcast player out there. For most general audience Apple podcast is something like 60 or 70 percent of their audience. So if you have data but you don't have data from Apple podcast your data is not that useful.

Apple actually gave people data. They didn't give personal data but now as a podcaster ever since November or so you can login to the Apple podcast backend and you can see what percentage of people listen how far into your show. And you can see what segments they skip. It's funny you can look at the graph of an episode and you can see the ad breaks because you can see that it's a small drop off for the ad breaks. There's a slow curve or a slow slope down as you get to end of the episode where you can tell people are like you know not everybody finishes the episode, of course, right. But the funny thing is we've had since October, November, December and that basically told us basically everything that podcasters say they wanna know. It didn't give any personal info which is what they really want.

But it did say you know what it turns out most people who download it listen to it. Most people who listen to it listen to the end and don't skip the ads. That's about all we needed to know which guess what we already kind of knew that. We already kind of thought that because our advertisers use things like special urls and coupon codes and everything. And that's how they can gauge roughly how many people are hearing their ads and responding to them. That's why podcasts have repeat advertisers. Because they can measure these things in the old fashioned way and they don't need all this tracking data. No one needs all that.

Anyway all this is a long winded way of saying that the podcast industry is trying very hard to get more data about people and I am in interested in supporting none of that. I would like to keep giving them the exact data they have gotten forever which is when you download the mp3 of the episode they by the nature of a download get your IP address and they get a user string that says they're using Overcast. That is all I wanna tell them. That is the minimum. Technically I could remove the use agent string but they are not really using that for analytics purposes anyway but rather using your IP address.

I think that's fine. I don't think I'm hurting anyone by doing that that's how the industry always worked and I'm just trying to prevent new things from invading our privacy in weird ways we didn't consider. So in Overcast 4.2 I blocked tracking pixels.