How short-form algorithms actually decide what to push
No platform publishes its formula, but the broad logic behind Reels, Shorts and TikTok recommendations is easier to read than you think.
By Ananya Kapoor
Nobody has the formula, and that is fine
Every few weeks a new “algorithm hack” does the rounds: post at 7.13 pm, use exactly five hashtags, never edit a caption after publishing. Most of these are folklore. Instagram, YouTube and TikTok do not publish the exact signals or weights their recommendation systems use, and those systems change constantly. Anyone who claims to know the precise recipe is guessing, or selling something.
What the platforms have said publicly, in help pages, creator events and interviews, is surprisingly consistent. Their goal is to show each viewer the videos that viewer is most likely to enjoy and keep watching. Once you accept that, most of the mystery falls away. You are not trying to please a machine. You are trying to please a person, and the machine is watching how that person reacts.
The test-and-expand pattern
Short-form feeds broadly seem to work in waves. A new video is shown to a small group of viewers, often a mix of people who follow you and people who have watched similar content. The system watches what those viewers do. If the response is strong relative to other videos competing for the same slot, it shows the video to a larger group, then a larger one. If the response is weak, distribution slows and the video settles.
This explains a few things creators notice all the time:
- A video can sit quietly for a day and then suddenly pick up, because it found the right audience in a later wave.
- Follower count matters less than it used to; a small account can reach far beyond its followers if the first viewers respond well.
- One flop does not “shadowban” you. It just did not earn the next wave.
The signals that probably matter most
Again, the exact weights are private, but the categories of signal are not hard to infer from what platforms say they optimise for:
- Whether people stop scrolling at all. The first second or two decides if your video even gets a chance.
- How much of the video people watch, and whether they rewatch. A 20-second clip watched to the end tells the system more than a 90-second clip abandoned at 15 seconds.
- Active reactions: shares, saves, comments, follows from the video. Shares and saves in particular suggest the video was worth something beyond a passing glance.
- Negative signals: quick swipes away, “not interested” taps, hiding or reporting.
Notice that almost every one of these is a viewer behaviour. Hashtags, posting time and caption length are, at best, minor inputs that help the system understand what your video is about and who to test it on.
The feed does not reward effort. It rewards attention that viewers chose to give.
Why topic clarity helps
Recommendation systems need to know who to test your video on. If your account posts about street food on Monday, crypto on Wednesday and your cat on Friday, the system has a harder time finding the right first audience for each video. Creators with a clear set of topics tend to get more reliable early tests, because the platform has learned which viewers respond to them.
This does not mean you can never post outside your lane. It means your core content should be recognisable enough that a stranger could describe your account in one sentence.
What you can actually control
You cannot see inside the system, but you can shape the behaviour it measures. A practical checklist before posting:
- Does the first frame make it obvious what the video is about and why someone should stay?
- Is there any dead air you can cut? Every pause is an invitation to swipe.
- Is the payoff worth waiting for, and does it arrive before people lose patience?
- Would someone send this to a friend? If not, what would make them want to?
- Is the on-screen text readable without sound, since many people scroll muted?
Stop chasing tricks, start studying viewers
The healthiest way to think about any short-form algorithm is as a very fast, very large focus group. It is constantly asking real people whether your video was worth their time. When the answer is no, the fix is almost never a hashtag. It is a sharper hook, a tighter edit or a clearer idea.
Treat each upload as a small experiment. Change one thing at a time, watch how retention and shares move, and keep notes. Over a few months you will build your own understanding of what your audience responds to, and that knowledge is worth far more than any leaked “formula” because it is specific to you.
The platforms will keep tweaking their systems. Viewers, on the other hand, change slowly. Build for them and you will be building for whatever version of the algorithm comes next.