There Is An Algorithmically-Driven Reason You Like Celebrity Pics on Instagram

So yeah, your circle of friends' behavior mathematically demonstrates they like Kim Kardashian photos.

Despite their fervent denials to the contrary. Deal with it. JL


Ashley Carman reports in The Verge:

Instagram realized that people were only seeing 30% of the posts from the accounts they followed. And so, being owned by Facebook and the end goal being to keep people inside the app, because that’s how people see ads, they rolled out changes to make it algorithmically driven to surface things. Because the signal is relative to your other content, how is this post performing, and that determines how high it gets served in the feed, then that effect of that thing being more popular is going to be amplified.

We aren't stopping here at Why'd You Push That Button? HQ, aka The Verge's offices. We've still got more episodes, and this week, we're asking: why do you like celebrity photos on Instagram?

This question might sound familiar if you're a Verge reader. I asked it months ago in this iconic post: "Why did my boyfriend like Emily Ratajkowski's butt on Instagram?" We now have the definitive, audio version of the article. We're going to get through it together as a family.

For this episode, Kaitlyn and I talk to my boyfriend Chris to get a final answer about his butt-liking behavior. We also chat with a certified Instagram influencer named Lisa Ramos, our dear friend and Verge collaborator Lizzie Plaugic, and Verge editorial director Helen Havlak. (I've taken to calling her the algorithm whisperer, though, so maybe she should consider that as an alternate title.) By the end of this episode, you should understand why you double tapped that photo of Kim Kardashian and why you keep seeing content tangentially related to Kim. You should have a clearer understanding of what a like means, in a philosophical sense, and how to feel about your boyfriend liking a model's butt. We grappled with these questions so you don't have to.

We also have an email now: button@theverge.com. Email us your ideas, your critiques, your thoughts, and, if you're feeling bold, an audio recording of a button pushing that went wrong. We're hoping to do a Thanksgiving special in which we play back your audio files and read emails. We just want to have a good time.

You can read our interview with Helen below. You can also find us everywhere you find podcasts, including Apple Podcasts, Spotify, Google Play Music, and our RSS feed.

Kaitlyn Tiffany: We’re curious if Chris liking the butt pic serves any real function as far as him seeing more butt pics, him seeing butt pics before he sees other pics, etc.

Ashley Carman: I think he feels, from when I’ve talked to him other than him being on the podcast — he didn’t actually think about it. He thought he was liking this butt in a vacuum.

Helen Havlak: Yeah, the thing about Instagram is that it used to be this purely reverse chronological feed of everything anyone you followed posted. They basically realized

though, that people were only seeing like 30 percent of the posts from all the accounts they followed. And so, Instagram being owned by Facebook and the end goal being to keep people inside the app, because that’s how people see ads and that makes money, they rolled out a bunch of changes last year to basically blow up the old style of the Instagram feed and make it algorithmically driven to surface things that were quote unquote “the best” for that particular user.

They’ve talked a little bit about the signals that go into what they decide to surface. That includes things like the likelihood that you’ll be interested in the content — we’ll get into that a little bit in a second. Your relationship to the person posting — if it’s your best friend it’s likely to surface at the top. If you are obsessed with, for example, Juniper the fox... I love foxes, I don’t know what to tell you, that bounces to the top of my feed. And the timeliness of the post, all go into that. They gave a little more detail to Business Insider about what the popularity of the post or your likelihood to engage with a post signals are. Those are things like, have you actively searched for someone’s profile in the past? Say you’re logged onto Instagram, you don’t see something, you think, “I’ve really missed this celebrity and I want to see what they’re up to.” That would send a signal to Instagram that they should be surfacing that content to you higher up. Things like direct shares. If you are often sharing from someone’s profile to your friends that will also send a signal. And then the popularity of the post, in terms of likes and comments. That’s probably relative popularity. So, something with 2 million likes versus your best friend’s post with 200 likes — if that 2 million like thing is Kim Kardashian’s post, and it’s performing 200 percent better than a normal Kim Kardashian post, that’s likely to go to the top of your feed. And if your friend’s post, say your friend only usually gets 100 likes on her post, if something suddenly gets 200 likes, they’re likely to serve that up.

Another thing they don’t explicitly talk about, but that’s probably a factor is time spent on the post. So how long on average do people spend looking at it. The end goal is to keep people in Instagram so that they spend money. Hypothetically, if a bunch of users are spending a really long time just staring at a butt.

AC: Which is definitely happening.

HH: Just a really long time.

KT: Just quiet, contemplative…

AC: Meditation time.

HH: That may send a signal that that post should be going to the top of everyone else’s feed, because guess what, it gets people to spend a lot of time on it and stay inside of Instagram. So all of those signals can go into it. And then if you look at how Facebook and Instagram target ads, it used to be that they called these “affinity groups.” They would say, “if you like this page, you’re also likely to like this other set of interests and pages.” Now they call these a “lookalike group,” but they may say, “okay, the accounts you follow and the way you interact with them is really similar to this other group of users and how they interact.” So, if users who look like you in their behavior and their interests are all liking this butt photo, then they may be more likely to serve it up to the top of your feed.

AC: Okay, we were also wondering for some of these influencers who do tend to post a lot of butt pics, if they notice that their butt pics are getting a lot of likes, it is beneficial for them because it surfaces them higher in the feed.

HH: Yeah, and I think they’re probably looking at, it’s relative to their other content. The algorithm tends to magnify the difference in engagement between the content you share. Right? If everything you posted got treated equally in the feed, then I think you wouldn’t see such a big discrepancy between butt photos: tons of engagement, non-butt photos: really low engagement. But because the signal is relative to your other content, how is this post performing, and that determines how high it gets served in the feed, then that effect of that thing being more popular is going to be amplified. It determines whether you’re in that 30 percent of posts people see, or you’re not.

AC: Lisa [Ramos], the [influencer / model] that Kaitlyn interviewed, she mentioned that she doesn’t care about likes.

KT: We discussed this, and I was like, this seems hard for me to believe as a person

who cares about likes. But she said, “if I post a bikini photo, I obviously know it’s going to get way more engagement than if I post an inspirational quote or something.” Even though she said she doesn’t care about likes, she said something like “I definitely understand the temptation to serve people content that they’re going to engage with.”

AC: Also to be a brand, you need likes to survive, I would assume. I don’t know how comments factor in. Comments don’t weigh as heavily as likes, do they?

HH: I would assume that comments are actually weighted more heavily. A like is easy, you double-tap and move on. A comment is a bigger level of engagement with something. And if you look at something explicitly calling out something like direct share, that’s probably an even higher weight because it means you care about something more. Probably comments are actually weighted a little higher than likes, and those don’t always have the same ratio, right? It can depend on what the post is. But I think in general, things with more comments are going to send a good signal that they’re engaging in a different way.

KT: James Vincent, our coworker who lives in a different continent that is escaping me right now —

AC: He lives in Europe.

KT: Okay. He tweeted something yesterday about how comparing Explore tabs on Instagram with people is like doing tarot readings. I feel like that is a thing I am very curious about. If you look at someone’s Explore tab, is that just the sum of everything we’ve been talking about? That’s pure algorithm?

HH: The Explore tab I think is pure algorithm, and it’s probably largely driven by this lookalike thing, where it’s looking at the accounts you follow and then it’s looking at other people who follow these accounts also look at these kinds of contents and these accounts. And it’s trying to encourage, suggest new accounts and new content to you that it thinks you would like. So my guess is that that would be more driven by this lookalike audience thing, of people who tend to have similar behavior to you and tend to follow the types of accounts you follow — what else are they doing that you may find interesting?

KT: So you gotta look in Chris’ Explore tab.

AC: That’s what’s funny, I love to think about the idea that Chris is defined in Instagram’s little user groups as like, “butt-liker.” “Loves butts.”

KT: Chris is sincerely one of the nicest people I’ve ever met in my life and is not at all what you’re picturing from knowing only that he likes butt pics.

AC: I feel like I’ve portrayed him in this terrible light, and he is very nice. But you know, clearly he just liked @emrata’s butt. And that’s fine.

KT: This is a time for me to share something about Chris. A couple of weeks ago when we were at one of Ashley’s friend’s birthday parties, he took out his phone to show our mutual friend something on Instagram, I don’t even remember what it was, and clicked Search, and his top suggested search was our girl Ashley Carman.

HH: Aww.

KT: So, he is redeemed in my eyes. And I texted her under the table, as soon as I saw that.

AC: It is true, she did do that. Do we have any other questions for Helen? I’m trying to think. Do you know about image recognition? Like, does Instagram — obviously, you know, you’re hinting at, if you like one model’s butt, there’s probably other models who post butts or whatever. Body pics. There’s a higher likelihood that people who follow this one model also follow this other model, and it just keeps going. But I’m wondering if Instagram has explicit image recognition where they can literally be like, “This is a butt.”

HH: The technology is definitely out there. If you look at, in Google Photos, how you can say, “make an album of only photos of my dad,” and it will just do it for you, the technology exists. I haven’t seen anything that suggests that Instagram is using that in its algorithm yet. Now, if Instagram could say “Look at people who engage with puppies” and look at that in the image and serve you more puppies, I think they probably would. I haven’t seen anything that indicates that they’re doing it yet. And I think it would probably slow down the speed enough right now that it’s probably not happening yet. But I think it’s probably only a matter of time.

KT: Also to be completely honest about the diversity of the things that people post on Instagram, I wonder if image recognition would be that useful. It’s like, yeah, here are the categories of posts, it’s people’s faces, people’s pets—

AC: Avocado toast.

KT: People’s food. People’s skyline views.

HH: The technology exists, it’s possible that it’s being used. But I think signals like, on Instagram you also know the account you follow, you expect things from them. So should that signal be higher than “what account did it come from and what’s your relationship to that account?” Probably not. So, even if that does get incorporated in, or even if that is incorporated in now, which isn’t totally inconceivable, I don’t think it would be as high a signal as things like your relationship to the poster, engagement on the post. The discrepancy between how posts perform is higher now in this algorithmic timeline.

AC: We’re all being quantified. Everything is quantified. Okay, well Helen, you’re the best, you’re so knowledgeable, it scares me. You know so many things!

HH: I’m happy to be on this brand-new podcast.