We all know how social media algorithms work, right?
Most people think they know how social media algorithms work, but they don’t. This article will shed some light on how algorithms use iterative testing—not sophisticated technology.
As a digital strategist, I’ve closely monitored social media algorithms for two decades. I’ve learnt that algorithms aren’t your friends—and we must manage them.
Let’s dive right in:
- We Need Sorting and Labelling
- Social Algorithms are Not "Personal"
- Our Lack of Consistency
- Why Social Media Algorithms Aren't Better
- How Social Networks Have Evolved
- Virality Through Real-Time Testing
- Underestimating the Effects of Filters
- The Power of Perception Management
- The First Rule of Social Media Algorithms
We Need Sorting and Labelling
You interact with social media, and the platform owner collects your user data to serve you more content to keep you engaged and thus increase your exposure to third-party advertising.
“So what’s the Original Sin of the Internet? Nearly all business models it supports require spying on consumers and monetising them.”
— Bob Sullivan
Order is necessary, of course, since there’s a lot of content to structure:
Unfortunately, the actual inner workings of a social media algorithm have a much darker side. And yes, “darkness” is a reasonable analogy because these algorithms are being kept secret for many reasons.
And behind these curtains of secrecy, we don’t find myriads of complex computing layers but manufactured filters designed by real people with personal agendas.
We are constantly shaping, suggesting, nudging, and presenting.
Social Algorithms are Not “Personal”
As users, we have a general idea of how algorithms work, but only a handful of people know. To prevent industrial espionage, we can safely assume that most social networks are making sure that no one developer has full access to the entirety of an algorithm.
And even if you’re a Facebook programmer, how would you know exactly how Google’s algorithm works?
One might assume that you have a personal Facebook algorithm stored on a server somewhere. An algorithm that tracks you personally and learns about you and your behaviour. And the more it knows about you, the better it understands you. But this is not exactly how it works—for a good reason.
However complex, applying various types of machine learning to learn about users and their interactions on the individual level would be both slow and expensive. Few social media users would be patient enough to endure such a lengthy process through trial and error.
Anyone familiar with data mining will know that more advanced scraping techniques, like sentiment analysis from social media monitoring, will require large data sets. Hence the term “big data”. 1Batrinca, B., Treleaven, PC Social media analytics: a survey of techniques, tools and platforms. AI & Soc, 89–116 (2015).
A social media algorithm gets immense power primarily from harvesting data from large users simultaneously and over time, not from creating billions of self-contained algorithms.
Put another way: Algorithms are figuring out humanity at scale, not your behaviour.
Our Lack of Consistency
No one argues the fact that you are an individual. But is your behaviour consistent from interaction to interaction? The answer is probably no. How you act and react are likely to be much more contextual and situational — at least concerning your perceived uniqueness and self-identification.
Social networks are limiting your options to tweak “your algorithm” yourself. I would be all over the possibility of adjusting Facebook’s newsfeed or Google’s search results in fine detail using various boolean rulesets, but that would probably only teach the master social algorithm a bit more about human pretensions—and little else.
Accurate personal algorithms that follow us from service to service could outperform all other algorithms. I would probably ask my algorithm only to show me serious articles written in peer-reviewed publications or published by well-educated authors with proven track records. But if the algorithm doesn’t take it upon itself to show me some funny cat memes or weird Youtube clips now and then, I would probably be bored quite quickly.
When we understand that the social media algorithms aren’t trying to figure you out, it follows to ask how well they’re doing in figuring out humanity.
Why Social Media Algorithms Aren’t Better
Social media algorithms are progressing—painstakingly. Understanding human behaviour at the macro level is not a task to be underestimated.
Google is struggling to show relevant search results, and it still isn’t uncommon that users have to search quite a bit before finding the information they seek.
Facebook is struggling with users complaining about what they’re being shown in the newsfeeds.
Spotify is struggling to suggest new music and often misses the mark by a mile.
LinkedIn is struggling to be business-relevant while at the same time personally engaging (i.e. not boring).
Instagram is struggling not to make people feel bad about themselves.
Pinterest is struggling with interpreting personal visual taste and intent.
Netflix is struggling with suggesting what to watch (“Why on Earth would I want to see Jumanji 2?”).
Amazon is struggling with telling us what to buy (“Please stop, I regret clicking on those purple bath towels by mistake a year ago!”).
The practical engineering approach to these struggles is straightforward:
Take the guesswork out of the equation.
How Social Networks Have Evolved
The Silent Switch
Not too long ago, social media algorithms would deliver organic reach like this:
Today, social media algorithms deliver organic reach more like this:
It’s the silent switch where social networks have demoted the publisher’s authority and reputation and promoted single content performance instead.
This algorithmic change has likely had profound and severe media implications for society, such as trivialization, sensationalization, and polarization.
Our job as PR professionals is to help organisations navigate the media landscape and to communicate more efficiently—especially in times of change.
Read also: The Silent Switch
Virality Through Real-Time Testing
A dominating feature of today’s social media algorithms is real-time testing.
If you publish anything, the algorithm will use its data to test your content on a small statistical subset of users. If their reactions are favourable, the algorithm will show your published content to a slightly larger subset — and then test again. And so on.
Suppose your published content has viral potential, and your track record as a publisher has granted you enough platform authority to surpass critical mass. In that case, your content will spread like rings on water throughout more extensive subsets of users.
This testing algorithm isn’t as mathematically complex as one might think from a programming standpoint. The most robust approach to increasing virality is reducing cycle times.
YouTube’s algorithm arguably does well in cycle times, but the real star of online virality is the Chinese platform TikTok.
And this is where it gets pitch black. Because the complexity and gatekeeping prowess of today’s social media algorithms don’t primarily stem from their creative use of big data and high-end artificial intelligence — it stems from the blunt use of synthetic filters.
The social media algorithms could be much more complex using machine learning, natural language processing, and artificial intelligence combined with human psychology neural network models. Especially if we allow these protocols to be individual across services, we will enable them to lie to us just a bit.
But, no. Instead, the social media algorithms of today are surprisingly straightforward and based on real-time iterative testing. Today, virality is controlled mainly via the use of added filters.
Underestimating the Effects of Filters
The algorithmic complexity is primarily derived from humans manually adding filters to algorithms in their control. These filters are tested on smaller subsets before rolling out on larger scales.
Most of us have heard creators on Instagram, TikTok, and sometimes YouTube complain about being “shadowbanned” when their reach suddenly dwindles from one day to the next—for no apparent reason. Sometimes this might be due to changes to the master algorithm, but most creators are probably affected by newly added filters.
Please make no mistake about it: Filters are powerful. No matter how well a piece of content would negotiate the master algorithm — if a piece of content gets stuck in a filter, it’s going nowhere. And these filters aren’t the output of some ultra-smart algorithm; humans add them with corporate or ideological agendas.
“There is no information overload, only filter failure.”
— Clay Shirky
TikTok serves as one of the darkest examples of algorithmic abuse.
Leaked internal documents revealed how TikTok added filters to limit content by people classified as non-attractive or poor.
And yes, this is where the darkness comes into full effect—it’s when human agendas get added into the algorithmic mix.
“One document goes so far as to instruct moderators to scan uploads for cracked walls and “disreputable decorations” in users’ own homes—then to effectively punish these poorer TikTok users by artificially narrowing their audiences.”
Source: Invisible Censorship — TikTok Told Moderators to Suppress Posts by “Ugly” People and the Poor
The grim irony here is that adding filters is relatively straightforward from a programming perspective.
We often think of algorithms as advanced black superhuman code boxes that operate almost above human comprehension. But with reasonably exact algorithms, it is artificial filters we need to watch out for and consider.
The Power of Perception Management
“Since we cannot change reality, let us change the eyes which see reality.”
— Nikos Kazantzakis
What would happen if Google and Facebook filtered away a specific day? Everything that refers to that day wouldn’t pass any iterative tests anymore. Any content from that day would be shadowbanned. And search engine results pages would deflect anything related to that particular day.
To paraphrase a popular TikTok meme, “How would you know?”
No one makes decisions based on the actual reality; we all make decisions based on our limited understanding of that reality. Hence, if you control some parts of that reality, you indirectly control what people do, say, or even think. 2Lippmann, Walter. 1960. Public Opinion (1922). New York: Macmillan.
No one is basing their attitudes and behaviours on reality; we’re basing them on our perceptions of reality.
Walter Lippmann (1889–1974) proposed that our perceptions of reality differ from the actual reality. The reality is too vast and too complex for anyone to process. 3Lippmann, Walter. 1960. Public Opinion (1922). New York: Macmillan.
Those who can manage the perceptions of publics can control their attitudes and behaviours.
The research on perception management is focused on how organisations can create a desired reputation:
“The OPM [Organizational Perception Management] field focuses on the range of activities that help organisations establish and/or maintain a desired reputation (Staw et al., 1983). More specifically, OPM research has primarily focused on two interrelated factors: (1) the timing and goals of perception management activities and (2) specific perception management tactics (Elsbach, 2006).”
Source: Organizational Perception Management 4Hargis, M. & Watt, John. (2010). Organizational perception management: A framework to overcome crisis events. Organization Development Journal. 28. 73-87.
“We are all captives of the picture in our head—our belief that the world we have experienced is the world that really exists.”
— Walter Lippmann
This is the truth about how social media algorithms are controlling our lives:
Social media algorithms and filters influence our perceptions of reality and, by extension, our attitudes and behaviours.
The First Rule of Social Media Algorithms
Social networks don’t want us talking and asking questions about their algorithms despite being at the core of their businesses. Because 1) they need to keep them secret, 2) they are blunter than we might think, 3) their complexity is manifested primarily by artificial filters, and 4) they don’t want to direct our attention at how much gatekeeping power they yield.
And both journalists and legislators aren’t exactly hard at work exposing these apparent democratic weaknesses; journalists want their lost gatekeeping power back and legislators because they see ideological opportunities to gain control over these filters.
Any PR professional knows that the news media has an agenda. And that we must manage that agenda. Otherwise, it might spin out of control.
A social media algorithm can be successfully negotiated and sometimes work for you or your organisation. But an algorithm with its filters will never be your friend.
Social networks are “good” in the same way the news media is “objective” or politicians are “altruistic”. As public relations professionals, we should act accordingly and manage the social media algorithms—just as we manage journalists and legislators.
Read also: Social Media: The Good, The Bad, The Ugly
|Batrinca, B., Treleaven, PC Social media analytics: a survey of techniques, tools and platforms. AI & Soc, 89–116 (2015).|
|Lippmann, Walter. 1960. Public Opinion (1922). New York: Macmillan.|
|Hargis, M. & Watt, John. (2010). Organizational perception management: A framework to overcome crisis events. Organization Development Journal. 28. 73-87.|