Use Me if You Dare, Hack Me if You Can
TikTok’s billion-dollar algorithm recommends videos that even you don’t know you’ll like. But wait a second…
TikTok is probably the most popular social media platform that I have never used. Maybe I’m too old for it, maybe I’m simply too scared to become addicted. There is no lack of blog posts trying to decode the success formula of this mobile app – it fascinates many older millennials like me how TikTok has carved out so much space, in what is seemingly an already saturated world of social platforms, with a format that some had long tried and failed (RIP, Vine).
For those less familiar, TikTok is a short-form video social platform that provides many powerful video editing tools to empower just about anyone to be funny, get creative, or even go viral. The main appeal of TikTok, in addition to or unlike the trio (Facebook, Instagram and Twitter), lies in:
- Content is recommended to users in a frictionless way – a stream of videos are shown to you before you even sign up for an account, much less follow your friends and celebrities.
- TikTok’s content is known to be fun, surprising, relatable, and impressive. The novelty of the “familiar” and the ability for normal people to go viral is refreshing. For once, the extra cellulite (rather than perfectly toned abs) or the ordinary living room background (rather than a luxury exotic resort) is permissible.
- Somehow, TikTok’s content has remained relatively positive, light-hearted, and apolitical – a break much needed by many, from today’s physical and digital world.
You may now see just how omnipotent the TikTok’s algorithm is, particularly the one behind the “For You” section. This is TikTok’s main discovery channel and – unlike Instagram – is where users spend most of their time. Arguably TikTok’s algorithm to “read” and distribute video content is its most crucial success factor. A high performing recommendation algorithm retains its users with endless and ever-so-creative content, and motivates creators with wider distribution and validation. The flywheel thus gets flying (or wheeling?). 
Not surprisingly, then, understanding TikTok’s main algorithm is hugely beneficial to all content creators. In fact, there are start-ups that solely focus on hacking this algorithm, because the “hack” equates a recipe for going viral, which in turn equates profits.
There are many plausible hypotheses for what attributes are weighted by the algorithm – some of them almost surely true: basic quality of the video, creator account quality, and viewers’ reaction. The rest of the algorithm still remains largely elusive, probably for everyone’s benefit.
In the video ingestion and review process, each submission is likely reviewed first on its basic quality – resolution, volume, and whether there’s any suspected violation of content policies – and then tagged by both the algorithm and potentially humans to build a “flat” understanding of what the video is about and who it may cater to. Then the video is pushed to and tested on small batches of user groups. It allows the algorithm to learn in what user group this video fares best, based on reaction such as comments, hearts, and rewatches. The video may then get distributed to an increasingly bigger audience but with a clearer “persona,” or it may simply flop.
There is only so much we can hypothesize about what goes viral, as virality by definition is elusive. Those “hackers” often end up finding insights that are boring and disappointing (and reassuring to the rest of us): apparently originality, creativity, and quality are still the best predictors for success. Consistency in content and style across the videos from the same account also helps. Given the musical DNA that TikTok started out with, great tunes and dance moves might get a leg up too. Essentially, to increase your chance of becoming a TikTok influencer, you might want to consider… becoming really good at your craft.
As TikTok itself goes viral, many problems come knock on its door. Similar to the issues Facebook and Instagram face, TikTok’s content policy and algorithm have attracted a lot of scrutiny. There is an impossible balance for these new digital publishers to navigate, between complete freedom of speech and complete censorship or paternalism. In addition, in a meta sense, I always wonder if the algorithm knows, or rather decides, who we are. Am I truly a giant dog person, or did Instagram make me one? Do I really like Parisian chic, or did Alibaba have a hand in this? Are the algorithms from those platforms helping me find who I truly am, or shaping me into who they think I look like?
Despite my mind-wandering in the alarming meta world, I somehow remain a technocrat. The optimist in me hope that TikTok will use its power well, and that large tech companies will stand by their original missions – almost always positive and irreverent, like TikTok’s content – in times of uncertainty and even darkness. For TikTok, it may mean more transparency and openness to change, and a strong discipline to grow responsibly. With each additional view, comment, heart, and rewatch, comes with additional responsibility. It is in TikTok’s long term interest, to not think of user engagement as only additional cents to its topline, but a social, cultural, and even moral puzzle that it is tasked to solve.
Great post. Like you, I too view TikTok as the biggest platform I’ve yet to try out. When it first made waves in the States, the majority of users were teens/kids. I wonder if that demo has anything to do with the dearth of political posts that we see in other platforms. And perhaps another driver of the platform being more creative (and less of a soapbox) is due to its core based in music/music videos. Regardless, interesting post that’s relevant to a lot of these black box recommendation algorithm platforms.
Very interesting post. I haven’t used Tik Tok, and to be honest, I’m not a big fan of any type of social media. I wonder if the success of Tik Tok is just being appealing for the next generation of teens/kids as ka said. We’ve seen in the past the trend of different apps being “cool” for a specific generation and then moving on. First, it was Facebook, then Instagram, then Snapchat, now is Tik Tok. Are we going to see a new one for the next generation of kids that want to have their own thing? Where they can be creative and need not to explain anything to their older siblings and parents that may be missing the new fad?
I totally see your point. In fact I didn’t catch the Snapchat trend, and initially “dismissed” TikTok as just another fad like the way I see Snapchat. But I’m now thinking it may be more than that. I think the key difference is TikTok is more than just users being social by sharing life updates; it is more of a Youtube with a much easier creative process and an automated distribution system.
Very well written post – I have TikTok on my phone and I haven’t created an account yet, but I still go there once in a while to test and see if the videos are relevant for me. It is amazing to see how quickly it adopts and somehow “knows” what you like. My one comment here could be if this could just bucket you in one “persona” that will keep getting reinforced and make it difficult to change or even look at the other “persona” suited content. So some spikes to throw off things that are totally unrelated and seeing how they fare might be a good experiment – and probably something that they already do?
Totally! I think the algorithm does build in some surprise factor – and they don’t always work.. Sometimes that random video in the middle of the stream is really just a bad production or a terrible fit for your interest. This may also have something to do with the platform trying to keep/lure back in new content creators, by giving their early/mediocre videos more views.
I am very excited about the potential of TikTok for advertisers given TikTok’s nature of user generated contents. TikTok is able to collect much more data than traditional TV or long-video sites because it has a much more frequent feedback from users every 15 seconds. I think there are two ways to utilize TikTok’s user data for advertisers: 1) tailor contents to users’ personal tastes, and 2) guide and switch users to advertisers own app/websites.
Jack, a really well-written post! My 12 year old uses TikTok constantly and because of you, I know what he is doing for 23 hours of each day. The part that caught my interest is the algorithm hacking – humans learning about the algorithm’s decisions and monetizing on this advantage. I wonder if it is truly just about originality, creativity, and quality, or is there a way to manipulate it given that 0s and 1s eventually make the decision for what gets shown to viewers. I know that YouTube does have algorithm hackers, and when YouTube changes the algorithm, it sends shockwaves and protests from influencers that must re-learn how to manipulate the system. And would this honestly be in the interest of TikTok to avoid? Algorithm hackers effectively consolidate the content that entertains viewers and makes the platform efficient. In other words, do viewers want to see the content that predictably follows their preferences, or would they rather see raw diversity?
Very interesting point Nicholas. Thank you. I agree that algo hackers and content creators are an important part of the ecosystem. However, from the perspective of a selfish and cautious social media user, I’d rather the platform remains fun, irreverent, and less salesy (TikTok is known to deprioritize business/brand accounts to some extent). But as a business school student I bet this likely won’t last!
Interesting post.
You mention the social media platform Vine early on in your post. Vine used to be tremendously popular and like TikTok used crowd-sourced short, catchy videos. Why do you think Vine failed and is there a risk that TikTok suffers a similar fate?
I’m very interested to see where TikTok goes as a company. I have actively resisted the urge to download it or create an account, but friends send me links to videos at increasing frequencies and I find some of the content so creative! I do have some concerns about the potential impact on mental health though. There have been a few studies linking increased social media use with negative self-perception and increased mental health disorders. In particular, those that facilitate more social comparisons among peers (e.g. Instagram, Snapchat) seem to have the most significant effect. I wonder if this is something TikTok is aware of and if in the process of trying to create a stickier product they might be exacerbating these issues.
Interesting post and very relevant to today’s social media craze. I am curious to hear what you think the true value that Tik Tok is creating and whether you believe this to be a sustainable business model in such a crowded space? Likewise, I am curious to know what differentiates this platform from that of Vine and if the AI recommendation algorithm is really positioned in a defensible manner to not be pushed to the wayside in the long term. In addition, I think they have a unique opportunity during this pandemic to continue to drive adoption and long term retention through ‘addictive use’ while several impressionable individuals are spending more and more times at home surfing the web as opposed to being in school, work, etc.
Very interesting article. Thanks!
I am intrigued by this “hacking” effort. Since the algorithm, I presume, is evolving constantly using machine learning technology, no one can tell how exactly the algorithm decides and recommends in real-time, not even by Tiktok itself. It is virtually impossible for a human to decipher AI’s decision process. What Tiktok can do is that sets some initial parameters and add more inputs to influence the decisions. In other words, Tiktok cannot directly control the output. In this sense, hacking and predicting recommendation algorithms sound more of a marketing scheme rather than science.
Very interesting – thank you! I think another aspect that I find fascinating about TikTok is the frequency at which its users post, generating much more data than other social media platforms, which can in turn be analyzed to further improve the recommendation algorithm. While most people might only post weekly on Instagram and barely at all on other social media platforms, TikTok users will post multiple videos a day every day, across a wide range of topics. This data generation frequency is unprecedented among social media platforms from my knowledge, and another unique competitive edge for TikTok’s recommender system.
I find this post very relevant in Covid times, I believe Tiktok is one of the biggest winners of social isolation and I wonder how the backend of this moment looks like, and how older people joining the platform out of boredom has impacted the community and its content, like when parents and aunts joined Facebook and then the content and engagement shifted.
About its algorithm I wonder if they see themselves as a data business, rather than a content business? That is how I would understand their sustainability, because I don’t think the content is valuable for the long term. If this is the case, I believe they need to address all the security concerns around them.