Social media algorithms. The concept of them is so unpopular that some alternative social platforms actually trade on the fact that they (supposedly) don’t have them. But in this post I just might change the detractors’ minds. I’ll explore the reality of algorithms, explode some of the myths that surround them, and find out whether they really do us a favour or a disservice.
WHAT IS AN ALGORITHM OR ALGO?
In the context of social media, an algorithm – algo for short – is simply a sorting mechanism. A system of priority for the display of content or profiles.
The most basic social media algorithm is a reverse chronological display. That’s newest posts always at the top of the timeline, with no filtering – i.e. nothing removed. Despite the order being very simple and basic, this is still an algorithm. And by default, any social app you can find on an app store will filter out some sensitive matter. They won’t be accepted by the app stores if they don’t. So essentially, all social networks with timelines use some type of algorithm – even those who claim not to. You can read some more network-specific details in Finding The Best Alternative Social Media Platform.
Most social platforms, however, additionally manipulate their basic, reverse chronological timeline. They’ll filter out more than they’re required to filter out by the app stores – most often to avoid endless repetition and valueless spam. Then they’ll start to interfere with the chronological order, so that popular or more personally-relevant posts are placed at the top of the timeline, even when they’re not the most recent.
This can get extremely complicated – especially since the ordering of the content often depends on individual users’ preferences and/or known social connections. If you and I are following exactly the same people, we may still, with our settings at default, see different timelines. Why? Because I Liked posts you didn’t Like, and you Liked posts I didn’t Like. Or otherwise positively engaged with them. Social media wants us to stay logged in and attentive, and the way it achieves that is by showing us posts from people in whom we’ve previously taken an interest. Or people in whom the people in whom we’ve taken an interest have taken an interest. Or… No, let’s just leave it there.
Complicating things further still, you and I may have changed our settings in different ways, and that might affect what we see too.
Some social networks may give us a choice of algorithm. Twitter, for instance, lets us select either a chronologically-ordered follow timeline, or one prioritised on personal relevance and quality. And we can easily toggle between the two. The same type of selection options are available with Twitter search. I like having that choice, and I do use both.
Finally, social sites with an ad-based revenue model will insert ads into timelines, so the ads are integrated into the algorithms too. One ad in twelve organic posts? One in ten? One in eight? The algorithm is what decides.
ARE ALGORITHMS FAIR?
The answer to the above question really depends on who you are. The only people to whom algorithms have real allegiance, are the social platforms themselves. For everyone else the outcome of algo’s is subjective.
Some people say that unmanipulated, reverse chronological display is the fairest type of timeline. That might at first seem true when it comes to a self-assembled timeline, like the output of the people you chose to follow. Provided you only follow on merit, that is – which a lot of users don’t.
But in search timelines, you wouldn’t want to see a completely unfiltered display, because the volume of repetition and useless spam on social media dwarfs the volume of useful, new content. A totally unfiltered search would be excruciating to browse, and I don’t think there’s a person on Earth who could sit looking at it for ten minutes and disagree.
So some filtering is always going to be necessary to avoid torturing readers. It’s just a question of where the line is drawn.
It can also be said that unmanipulated chronological timelines benefit publishers who post low quality content at high frequency. Indeed, those timelines encourage the posting of low quality content at high frequency, because they make the most frequent posters the most visible. And unless the most frequent posters are stealing, they’re almost inevitably not going to be producing the best posts. Empty vessels make the most noise.
Viewed like that, straight reverse chronology is not only unfair to people who invest time in what they create – it’s also unfair to readers, whose chances of seeing the most valuable material are reduced by the litter of borderline spammers.
I don’t really think there’s a valid, broad argument that algorithms which prioritise on quality are unfair. The problem starts when a platform allows its quest for revenue to override the organic priority of its best content. That is, it artificially makes lower quality content more visible, because it makes the platform more money.
Interestingly, ad-driven networks tend to avoid this problem. Yes, they thrust ads into their timelines, but in order to get people to read those timelines and see the ads, it’s still in the platforms’ interests to make the best organic posts most visible.
But on non-ad-driven platforms, there can be a motivation for the algo’s to prioritise poorer content which is monetised, over and above better content which isn’t.
Nowhere is this more evident than on Medium.com. I’m not recommending you join Medium – unless you wanna be charged to read smug, annoying, patronising, aspirational dronefeed that is broadly delivered in the words and style of a Team Leader at a call centre staff meeting.
But if you do join, you’ll see that the great majority of article titles the site pushes onto the homepage feeds of free users, are paid-access content. As a non-paying user, you can access three of these articles per month, after which you have to subscribe to access more. You can still access all of the free content. If you can find it. Which is difficult, because the overwhelming majority of what Medium puts in your feed, is paid-access.
Given that most of the content on the site is free, there’s no explanation for the overwhelming prevalence of paid-access content – other than that Medium deliberately engineers the algorithm to display paybait. And that means it’s disproportionately crippling free content. Even when it’s better. Craig Phillips has illustrated the prevalence concisely in this post, which is definitely not dronefeed, and which I found via a search engine, not through Medium itself.
If you’re thinking that the work of people who write for money is probably better than that of people who write for other reasons, I can fairly confidently tell you that’s not the case. Medium‘s paybait algo-feed is vastly more riddled with clickbait than feeds on equivalent platforms that don’t have a pay-to-read system (the platform you’re on now, for example). And the worst clickbait offenders on Medium, in my own experience, have been authors of paid-access content.
The fact that a post comprises a thousand words doesn’t mean it’s not clickbait. If the title is only tenuously related to what’s in the post, it’s a grift. And of course paid Medium authors are after clicks at any grift. How could they not be?
Medium’s algorithm has been highly detrimental both to the user experience and to the visibility of the most conscientiously written content. It’s put such critical importance on the visibility and clickability of ‘premium’ post titles, that the post title has become 99% of an article’s value. The idea that only the ad revenue model encourages clickbait is pure bullshit, and some of Medium’s paid authors are categorically painful writers. The priority awarded to their work is an example of algorithms doing the majority a disservice, for the benefit of the few.
WHAT IS DEBOOSTING?
Okay, so we’ve seen how algorithms work, and started to touch on their positive and negative implications. But content priority is a rapidly evolving world, and one of the most recent buzzwords which has appeared with regard to algo evolution, is deboosting. The word has raised a lot of questions.
The meaning of the word “deboosting” is still fairly cloudy. It came into the public domain after an unauthorised leak regarding internal Facebook practices. I’m not going to link to the Facebook leak because the expose has an overwhelming political bias and the site blocks Tor – the privacy-advocate‘s browser. What can be said, is that deboosting is not the same as shadowbanning. Shadowbanning is the outright hiding of content or a profile, whereas deboosting is an algorithmic adjustment, designed to ‘unprivilege’ a user with privileged status.
There are other definitions of deboosting, especially in relation to Twitter. The word has, for example, been used by shadowban.eu to describe Twitter’s ‘clickwalling’ of some replies in a reply thread. However, I wouldn’t describe that practice as deboosting, and neither do I agree that reply-clickwalling is always the result of a ‘misdemeanour’.
For me, the best interpretation of the meaning of deboost, is: to remove a user’s algorithmic priority, for a special reason.
So what is algorithmic priority? Well, for example, let’s say you and I both have a reach of about 1,000 impressions within our own follower-base. From those 1,000 views, your posts normally get around 75 engagements each, but mine only get about 5. With all else equal, a typical social platform’s algorithms would logically prioritise your posts over mine – or boost them – on the basis that more people probably want to see them. You have algorithmic priority over me.
This is where it gets complicated. Some social sites have a user-initiated distribution-boost system. On Twitter it’s called Retweet, on Tumblr it’s Reblog, etc. These are official tools, hardwired into the sites, and they put the general public in charge of inflating a given user’s reach. If I see your tweet, and I hit Retweet, my audience will see it as well as yours. Importantly, the decision to increase your reach is mine, not Twitter’s.
But Twitter also artificially, algorithmically, boosts the reach of users who have a good track record with engagement. So the visibility boost has a compound effect. The more Retweets you get, the more likely Twitter is to algorithmically boost your visibility. And the more Twitter algorithmically boosts your visibility, the more Retweets you’re likely to get. The result is a snowball effect.
There’s a tipping point at which this ‘snowball’ becomes too big to stop, and that’s when we see what we recognise as a viral tweet. Users who have artificial algorithmic boosts are a lot more likely to go viral on Twitter than those who don’t have them. So those people really have a substantial privilege.
I’ve felt the effects of that privilege from both sides. I have more than one Twitter account with between 1,100 to 1,500 followers, but virtually no reach. If tweets from those accounts get 100 impressions, they’re having a good day. But I also have one account that, more by luck than anything else, crawled onto the engagement ladder. That account currently has about 1,200 followers – so at the lower end of what the others have. But as long as I tweet within the core audience’s interest-base, a tweet from that account will get a lot more views than the profile has followers. Probably in the 4,000 to 6,000 range. Same number of followers, but around fifty times the reach.
Some of that is down to the effect of Retweets, and some is down to the fact that the followers are much more attentive than those on the other accounts. But there’s another factor. Tweets from the better-perfoming profile won’t lie down and die if people don’t pick up on them straight away. Twitter keeps feeding them onto timelines and waiting for a bite – even the next day. Some tweets are still getting Likes two or three days after posting. That doesn’t naturally happen with the other profiles. It’s an algorithmic boost.
Other social networks, such as Facebook, don’t have user-initiated distribution-boost mechanisms. Friend volume aside, it’s entirely down to Facebook to determine how far a piece of content can reach. So artificial, algorithmic boosting is more critical to the success of Facebook publishers than it is to Twitter publishers.
The fact nevertheless remains that the algorithmic boost is a privilege. Not a right. Before you complain that you’ve been deboosted, you should recognise that the majority of users are never boosted in the first place. And you can’t argue “but MY TALENT tho!“. I’m one person. One of my profiles has a boost. The others don’t. If talent were the criterium, either all my profiles would have boosts, or all would not. The boosts are not based on quality of content. They’re a product of factors like luck, persistence, personality type, etc. And someone who has high status in the offline world is many times more likely to be boosted than someone who doesn’t. Social media amplifies real world privilege. It’s not the great leveller it likes to think it is.
ARE THERE ALGORITHMIC BIASES AGAINST CERTAIN GROUPS OF PEOPLE?
Ironically, complaints about algorithmic victimisation tend to come most frequently from people who have way above average status on social media. Part of the reason for this is that they’re used to having privilege, so when they perceive that their privilege has been reduced they interpret it as unfair. Even though they’re still far better off than the vast majority of people publishing on the site.
The real advantages and handicaps in algorithmic distribution are deeply rooted in people’s real life status, and their personality. But there are additional factors, such as the overlap between a user’s role in life, and a given platform’s prohibitions.
Sex work is the biggest and most obvious example of people’s actual earning prospects being either allegedly or actually inhibited by social media, as compared with people in other professions. But algorithms don’t play as big a part in that as some imagine.
Some platforms, such as Instagram, have specific terms clearly stating that the marketing of sexual services is not allowed. So there’s no algorithmic tailoring involved. People who try to sell sexual services on Instagram are simply banned upon recognition. And if they return, once recognised they’re banned again – regardless of whether they breach another rule. Because Instagram bans are permanent. It’s forever. Not “until such time as you should wish to try your luck with a new account“. Technically, sex workers can’t use Instagram to market their services at all. Whether that’s right or wrong is a matter of opinion. But it’s a documented policy. Not a sly algo trick.
On Twitter, the marketing of sexual services is permissable, and the rules only apply to the methods of promotion. If adult workers can promote on Twitter without posting sensitive media, without spamming, and without being abusive, they’re unlikely to suffer algorithmic penalties. And because Twitter’s distribution boosts are in large part user-initiated (i.e. fans can retweet), building a sex work profile on Twitter is blatantly easier than building, say, a tiling store profile. Blokes are gonna RT pictures of women in bikinis – especially if those women are pretending to be single and specifically ask for RTs. No one’s gonna RT pictures of bathroom tiles. So I wouldn’t say that sex workers are currently disadvantaged on Twitter.
Political voices can also fall foul of algorithmic manipulation, but generally it’s not politics per se causing the problem. It’s stuff like racism, hate speech and abusive conduct. And again, the more likely consequence of racism, hate speech or abuse is not an algorithmic deboost. It’s suspension. Algorithms are primarily about best serving the platform’s need to make money. Not so much punishing rulebreakers – although they do have a role in that as well.
Social media algorithms are blamed for a lot, but on ad-driven networks, they serve the average reader pretty well. They have to. Because no one is going to sit around looking at ads if they’re not getting something worth having in between.
It’s exactly like watching a commercial TV channel or listening to commercial radio. The ads get annoying at times, but you tolerate them because you like the programme. If those commercial channels started serving us poorer programmes, we wouldn’t watch or listen to them, wouldn’t see/hear their ads, and therefore advertisers wouldn’t pay to advertise. I would never suspect that Channel 4 had deliberately rejected affordable programmes it knew would be the most popular with its typical audience, because why ever would it?
So likewise, I don’t suspect that Twitter is deliberately hiding its best tweets. People accuse the platform of having a political agenda, but in truth the only agenda any business has is making money.
And there’s one glaring contradiction with all the visible claims of an algorithmic gagging conspiracy… I CAN STILL SEE YOUR TWEETS. Even when you call the management rude names. Still seeing it. When you “expose the site’s corruption” with a hotlinked YouTube vid. Yep, still seeing that. When you eulogise the “enemies of Twitter”, promote alternative networks, and post a 25-tweet thread documenting the uncharted depth of your current shadowban… I can still see your tweets. All of them. So can all the people who RT’d you, Liked you, Followed you…
Algorithms are designed to best serve a platform’s need to make money. That’s it in a nutshell. That really is it.