Hacker News new | past | comments | ask | show | jobs | submit login
How Instagram’s algorithm works (techcrunch.com)
129 points by laurex 10 months ago | hide | past | web | favorite | 66 comments

Instagram's algorithmic feed is hot garbage. I went through a two week period where it would ONLY show posts from three or four of the ~2,000 accounts I follow. Those accounts also happened to belong to big-ish companies, compared to most others I'm following, which are individuals or small organizations. It got so bad I had to unfollow them just to get any other content in my feed. And because I can't trust the feed, I keep a mental list of accounts (maybe 5-6) to "check on" manually for content, which is way more work than I should have to spend on a photo-sharing app.

If I could switch back to chronological feeds, in exchange for double the ads or something, I'd do it. It'd waste less of my time.

Also, Instagram banned me from following via the app for about five months with no communication or explanation. I think it thought I was a bot, but can't be sure. Not the customer but the product, indeed.

> And because I can't trust the feed, I keep a mental list of accounts (maybe 5-6) to "check on" manually for content, which is way more work than I should have to spend on a photo-sharing app.

I use a third-party app for Twitter, since they've abandoned any relationship with chronological reality, but even so I keep a separate list of people whose tweets I absolutely don't want to miss that I check every few days.

Gosh, I miss RSS.

rss is alive and well! why did you stop using it?

My Instagram has turned into a National Geographic browser.

Mine has turned into hot girls and yoga.

You click on 1 too many photos of a girls chest popping out of her shirt and you lose control of your IG.

The real question here is whether the algorithm is correctly reflecting your interests or not. Are we our ego or our id?

Suppose my ranked preferences (by assertion, watch time, and also ad interaction, and) of context mix may be 30% yoga, then 15% yoga, then 100% yoga.

Suppose IG only offers 100% yoga or 10%, based on my click activity.

Ego and id may be fully compatible but it IG is still failing for my use case.

I used to like following National Geographic, but they post so much it takes over my feed. Ended up unfollowing solely for that reason.

I've started following specific National Geographic photographers that I really like instead.

Agreed. I picked up an interest in golf lately so I followed maybe 1-3 players and now that's all I see. Everything in my feed and the explore is golf. Which, granted could be worse and I don't hate it but now I feel like it's no longer a social app just a golf feed.

I prefer it to show me things I hadn't seen since the last time I visited, in chronological order. Close to what Slack does and shows you a _New Messages_ indicator.

With that being said, in case of Instagram Im kind-of OK with it as it's just pictures.

In case of Twitter, if I was a product manager, I'd file a bug report with highest priority. The kind of bug that developers would not sleep until it's fixed.

I get a series of tweets about an ongoing event in a random order.

It's like watching a goddamn movie with scenes shuffled randomly. It makes no sense.

> It's like watching a goddamn movie with scenes shuffled randomly. It makes no sense.

Have you seen Memento?

Jokes aside I never understood why twitter does not allow the user to choose the sort order of the feed. This is a reoccuring theme for social networks (facebook feed, youtube subscriptions etc) so I guess there is some kind of motivation behind it.

These services want to show you cancer - sorry I mean ads - as much as possible, and for that they want you to stay on the app as long as possible.

With a sane chronological feed, once you scroll down to the last tweet you remember reading, you know you’re done and can leave the app and do productive things.

With a bullshit “algorithmic” feed, you never know when you’ve caught up, and the “fear of missing out” will make you stay there scrolling way longer just in case there’s one tweet somewhere down there you haven’t seen yet - this gives them the opportunity to show you more cancer.

Just curious, but do you think a paid service that served photos and no ads, and allowed the customization of feed order, would be able to compete? I've presumed the reason folks continue to flock to Instagram is that not only is such a service not available, but the problem that they would likely have to pay before seeing if there's anything to see there.

I’m not sure it would be able to compete now - displacing an established player is very hard; App.net tried that a few years ago with Twitter; but ultimately not many people cared (despite the free tier) and the few paying users weren’t enough to make it sustainable. The issue wasn’t really with it being paid, it was simply about network effects - everyone was on Twitter.

If a new free social network becomes mainstream, they could however introduce a paid tier as part of their monetisation strategy alongside ads. This would give users a choice, be exploited by ads and see cancer every day, or pay your way out of it.

If anything, Instagram would’ve been the perfect opportunity to offer such a paid plan, as a lot of professional photographers use it, and they wouldn’t mind paying for an improved experience and thus more productivity.

> but ultimately not many people cared (despite the free tier) and the few paying users weren’t enough to make it

My guess?

Twitter only exist because of network effects.

Everything else somebody else does better.

Oh, and most orinary people are more interested in one-to-one group chat than in many-to-many.

WhatsApp managed to break through even though they were announcing loud and clear that they were going to cost money. Why? I guess because they promised to provide something valuable: a versatile, reliable, trustworthy messaging solution for friends, families and small groups.

Telegram has managed to break through as well. AFAIK Signal is climbing steadily and Matrix is getting more and more mindshare.

Based my experience a lot people here will probably disagree with what I say about twitter. If you are one of those, feel free to tell me what technical detail twitter does better than everyone else.

Yep totally agree, Twitter only exists because of network effects.

> even though they were announcing loud and clear that they were going to cost money

On the other hand, WhatsApp announced this back when scammers aka mobile networks still charged a ton for texts, and WhatsApp was the only cheaper option.

Now the masses are trained to expect everything for free, so it's really hard to get them to pay for something. It's not even about social networks, even getting someone to buy a stand-alone app is hard enough - outside of the tech circle, most people consider it weird that I pay for apps. Even getting them to pay for the likes of Spotify is hard (most non-tech people I know are still on the free tier and get their music fix on Youtube, with ads and everything).

This is why I don't think a paid service is unlikely to succeed and become mainstream. A social network needs to be free, at least until it reaches a critical mass, at which point it can switch to a "freemium" model with a free (possibly ad-supported) base tier and a premium, ad-free tier.

I think I can kind of understand those users: I like paying for apps but a number of them will still try to abuse me by adding all kinds of monthly payments etc, start adding ads, remove the features I bought it for etc.

I also acknowledge there is a problem where authors should be able to secure an income to do security updates etc and I've no "one-size-fits-all solution" to it but I think between

- feature keys for new features

- optional extra services (support, server side ocr etc)

- reasonable fees for special cases (for example I loved to pay USD 1 a year for a user friendly, secure messaging service that promised not to spam me, datamine my data or leak it voluntarily.)

- etc

there should be room for better alternatives than many of the ones we have today.

> reasonable fees for special cases

Who'd like to pay for a stackoverflow clone without the most annoying stackoverflow rules?

I mean, - a site not focused on curating knowledge but on asking and answering questions?

That answer could of course be: have you looked at this <embed or link>?

But nobody should be punished for asking again and certainly nobody should be punished for answering a duplicate.

You nailed it. When a company lives or dies by advertising revenue, they are going to optimize for maximum advertising revenue.

The pattern with TV is a good case study. I wonder if there will eventually be guidelines about how many ads you can show users and now much screen real estate is allowed to be covered by ads.

> I wonder if there will eventually be guidelines about how many ads you can show users and now much screen real estate is allowed to be covered by ads.

Isn't that kind of what "acceptable ads" programs in adblocks are? (Or, at least, should be?) There are a lot of Adblock Plus users there that basically decided (or got tricked into thinking) that ads that are appropriate are okay to view.

The issue with "acceptable" ads is that 1) there is no such thing as "acceptable" shit, and 2) the "acceptable" ads are just as bad (if not worse) than everyone else as far as privacy & stalking are concerned.

Even if they limit to a reasonable 10% ads or whatever, they'll still manipulate ad sacrifice content quality to maximize total engagement time/content.

Not only that, but when they can insert other media from friends/people you follow, it can potentially start a new chain of exploration. Even if you know that you've caught up, now you want to see what those other threads are about.

It works until it doesn't, when people start unfollowing or leaving IG.

I imagine the motivation as

1. There's an inherent belief that they know what you want better than you do (this is inherit in any recommendation engine that gives limited ability to manipulate it)

2. Most users, given the option, won't bother changing it anyways

3. You'll use the service regardless of whether they offer it, because network effects

> There's an inherent belief that they know what you want better than you do

I think it's actually that they believe they can create an algorithm that will keep you coming back. It clearly works. People want to be entertained, often by their friends, often by things that interest then. This algorithm based on _your input_.

You can absolutely manipulate the feed. Search for 12 yoga based people, click on 12 yoga based hashtags, and then watch your feed become yoga-ified for the next few days as the algo adjusts to keep you coming back.

>This algorithm based on _your input_. You can absolutely manipulate the feed.

You can incidentally manipulate the feed. But its clearly not intentional behavior; the recommender system naturally operates on the belief that you're using the services without considering the recommender system. You can manipulate the system to something specific by breaking that assumption, but all you've done is essentially exploit a bug. Its clearly not a feature. And in a perfect world, the recommendation system would realize you're lying to it, and discard that behavior, and serve you what you really want.

your input is expected to be natural consumption of the service, with limited, abstracted forms of feedback, but otherwise without knowledge of the recommender system judging your every move. You don't edit the results, you don't play with ordering, you ideally don't even search too much out of your defined subject range (lest it skew horribly and fuck up all your recommendations, like your yoga example).

>I think it's actually that they believe they can create an algorithm that will keep you coming back. It clearly works.

But this I agree with; it's a better understanding of what I was trying to get at.

1 and maximizing time spend in the app (see sibling comment) seem like a good explaination for this. I'm not sure about 2 but it might be true aswell.

> 3. You'll use the service regardless of whether they offer it, because network effects

Actually I stopped using Facebook for that reason. The feed just didnt provide any value to me anymore. Also I never really got into Twitter because of it. I don't see the value in following people if I cannot see their posts without hours of scrolling through retweets.

Twitter does allow choosing this though. Untick "Show the best Tweets first" in settings and you will get reverse chronological order back.

What's the deal with devs not going to sleep? That's unhealthy.

In the app, if I start by watching a bunch of stories... it'll refresh the main feed when I go out of stories, but I haven't even scrolled down to see those posts yet.

With a "random" feed, it makes it real hard to find out what you just missed.

This is the best they could come up?

This reminds me of Facebook's claim, when it introduced its algorithmic feed, that it was all for the benefit of the average person, who would benefit because (they said) a person with 300 friends would be having trouble managing 1500 posts a day. This was a lie manufactured by Facebook to sell the algorithmic feed to journalists, who passed it on as truth to their readers.

First of all it was misleading: people with 300 or even 500 friends have nowhere newer 1500 posts a day. Sure, some friends post 10 memes a day, but most people do not post anything at all. In any case, I did not hear anyone complaining that they had to deal with too many posts, and managing the feed wasn't an issue. No-one was complaining about the chronological feed before,

What was an issue, however, was that Facebook (and now Instagram - also owned by Facebook) needed to make more money. So they created an algorithmic feed that pushed posts that engaged people more to keep people on the platform. I suspect this is exactly the same tactic, by the same people, backed up by the same compliant media, that we have seen before.

Wow, the article actually closes itself if you scroll too far down? What kind of horrendous design is this?

And you can't even scroll back, potentially being baffled why the website malfunctioned and loaded a new page. It's like a bad UX pattern parody.

Not only that, but it only closes itself if you scroll too fast, thinking that you're not interested in that specific article you've clicked on.

Best example of trying to over-optimize a site I can think of.

The new Techcrunch design is horrible. I stopped reading it entirely because of this design. It is flaky and jumps around leaving you confused if you aren't paying attention.

A huge part of why they introduced an algorithmic feed is probably to sell user engagement instead of giving it away for free.

Prioritising actual people means that they de-prioritise businesses, who now have to pay to get to previous levels of user engagement.

Learned that lesson from Facebook.

Tried to pay a marketing company to grow my website. Instead they flaked after turning my page into a business page.

I get 100 people instead of 400 people seeing my posts.

And then I stopped using facebook completely. If facebook is only going to show my political garbage shares and they wont show my website? Theres 0 value from facebook.

> Shadowbanning is not a real thing, and Instagram says it doesn’t hide people’s content for posting too many hashtags or taking other actions.

Would they admit it exists in either case?

I was wondering how they know it is not a real thing: they use machine learning, they feed the content with all metadata in and then let some trained AI logic filter/prioritize it. If the learning algorithm has found that most people don't 'engage' with content from certain people, or with a certain type of content, it might filter it without being intentionally being programmed in, right?

Thats not really the same as shadowbanning though is it? What you've described isn't much different than you posting with few to no followers; its just being further enforced by the recommendation engine. But ultimately, hopefully, its still a reflection of the community's taste, and not the administration.

Shadowbanning is more of an explicit denial of the community taken by someone who doesn't represent the community, but rather the hosting service itself. It might align with community taste, but also might not.

I think what people have said about shadowbanning in the past was not that your post were hidden from your followers feed but more than your post would not be visible within the hashtag section. So let's say you keep spamming your posts with 30 unrelevant hashtags, instagram would stop showing your post within the hashtag page. Instagram didn't explicitly say it was not happening and they could make a case that it isn't the true definition of shadow banning. But it seems that many people have observed this behavior in the past.

> not that your post were hidden from your followers feed but more than your post would not be visible within the hashtag section

I had an account that was shadowbanned, or at least seemed to be, and I was able to confirm that my posts didn't show up under hashtags or in followers feeds. In order to see my posts you had to manually navigate to my page, which caused my likes to drop to less than 5 per post

This has happened to me. I had an account at one point that used to get 100+ likes on the majority of posts. After the introduction of the feed algorithm it went down to about 40-60 likes regularly, then at some point it dropped to <5 likes per post. I created a separate account on a computer that had never been used to login to an Instagram account before while at starbucks, and confirmed that none of the posts from my primary account were showing up under any of the hashtags I use regularly. I then followed my primary account from the temp account, as well as a few other random accounts, and the posts from my primary account never showed up in my feed. After a couple weeks of trying to resolve this with no luck I chose to stop using the account. I checked every couple weeks, and it took 4+ months for the posts to start appearing under hashtags and on users feeds again. I ultimately decided to just delete my account

There are lots of videos on YouTube where people discuss being shadow banned. Basically their engagement plummets and they stop showing up for hashtags and (sometimes) the explore page. They may also be showing up less for their followers too. Maybe Instagram doesn't think of it as banning and thinks of it more as penalties. Your content isn't hidden, but it's not promoted as much.

I'm pretty sure this has been my experience on Instagram. Two years ago, every post I made had between 400 and 800 likes. Now everything gets about 130.

Based on the comments on TC's piece, the article is bullshit.

If you do any type of real work on Instagram, most of this article is completely inaccurate. Everyone has reported getting far less engagement after the changes. I've more than doubled followers since recent changes, and my engagement is roughly half what it was. Reach is WAY down, and it's directly affecting it. This is a niche localized account, with real followers. No garbage.

I do work on instagram and yes engagement has been down. But they just said in the article that they prioritize posts with people you DM, people you comment and people you have been tagged with (ie: your actual real life friends). So this would explain the drop in engagement. Also have to keep in mind the platform has already more than doubled in users since the algo changes, that means people follow more people and this also dilutes your importance in someone's feed if they now follow twice as many people.

I think this means they prioritize people within your close circle above brand that try to engage with thousand of followers.

See quote on this article: Instagram’s main goal is to help you see content from your “friends and family,” and with the algorithm they say that people now see 90% of posts from their friends and family, instead of 50% when it was the chronological feed. https://later.com/blog/instagram-algorithm-update/

That means that if you're not friend or family, you are being seen less.

Everyone? So total image views or likes are decreasing, site wide?

Pretty much nothing in here is "news", every social media manager knows the same things. I guess now it's just official.

There is no news or even facts in this techcrunch article.

But you get to experience their horrendous design! Isn't it great how the final paragraph is harder to read because the page removes the contrast? Or how, when you scroll a little too far, the old page disappears completely?

  if(hasClickedOneImageOfASwimsuitModel) {



There are companies that approach instagrammers and promise increased exposure of posts within that account's followers - and it seems to work. Anyone know how they do it?


I saw some model with 1.1m followers. Went to her website page + looked it up in alexa.siteinfo

1.1M followers, top 11m website, most traffic coming from India.

So basically a fake instagram. I think this is extremely true and there are a few real users that make it all worth it.

In this case the instagrammer didn't get any new followers, somehow he just got more exposure among his own, real followers. The instagrammer did not know now how they did it, he said he just posted a normal photo.

Thanks for mentioning alexa siteinfo, didn't know they track individual ig profiles.

Now fix the explore page

its nothing but stupid prank videos now

They don't even seem to change. It's all just like a small group of "influencers" posting the same shit constantly. This is why social media is slowly making itself irrelevant. I just wish we humans had the self control to stop doing it once we realized it's not adding value.

We humans do. I wish you humans luck.

I was using that as a general population. You can be as proud as you want but it doesn't stop the general population from letting these services walk all over us.

Same as reddit's new front page, hah

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact