#Tagly
Fundamentals

How recommendation algorithms actually decide what to show

Every platform’s algorithm is different in detail and broadly similar in structure. Understanding the structure explains most of what otherwise looks arbitrary.

7 min readUpdated 29 August 2026By Tagly

"The algorithm" gets discussed as an adversary with moods. It is more usefully understood as a prediction system with one job: estimate how likely a given person is to engage with a given piece of content, and rank accordingly. Almost everything else follows from that.

The basic sequence

  1. Classification. The platform works out what your content is about, using the caption, hashtags, transcribed audio, on-screen text, visual content, and the history of your account.
  2. A small test audience. Your post is shown to a limited group — some followers, some non-followers with matching interests. This is the phase that decides most outcomes.
  3. Measurement. The system observes what that group does: watch time, rewatches, saves, shares, comments, and how quickly they scroll away.
  4. Expansion or stop. Strong response means a larger audience and another round. Weak response means distribution effectively ends.
  5. Repeat. Each successful round widens the audience, which is why viral growth looks stepped rather than smooth.

This structure explains several things that otherwise seem strange — including why a post can sit flat for hours and then take off, and why two similar posts can perform completely differently. The test-audience draw is not identical each time.

The signals, roughly ranked

SignalWeightWhy
Watch time / completionHighest on videoHardest to fake, closest proxy for genuine interest
RewatchesVery highAlmost unambiguous — nobody rewatches by accident
Shares and sendsVery highCosts the sharer something; strong quality signal
SavesHighIndicates lasting value rather than passing interest
CommentsHighEffortful, and generates dwell time
Scroll-away speedHigh, negativeA fast skip is a strong negative judgement
LikesLowCheap, ambiguous, easily habitual
Follows from postHighIndicates you reached the right audience

The pattern is that signals are weighted roughly by how much effort they cost. That is a sensible design: effortful actions are harder to game and better correlated with genuine interest.

Why the first minutes matter

Because distribution expands in rounds, early performance compounds. A post that does well in its first test round gets a larger second round, and so on. A post that does poorly early does not get a second chance, regardless of how good it would have looked to a wider audience.

This is the mechanism behind advice about posting when your audience is active, and it is one of the few pieces of timing advice with a real basis. It is also why the opening seconds of a video matter disproportionately — they determine the early watch time that decides whether there is a second round.

Where hashtags sit in this

Hashtags contribute to step one — classification. They help the system decide what your content is about, which affects which non-followers are in your test audience. That is genuinely useful, and it is also a much smaller role than hashtag advice implies. They influence who gets tested, not whether the test succeeds.

Which explains a common frustration: better hashtags produce a modest improvement, not a transformation. They are aiming the shot, not powering it.

Things commonly attributed to the algorithm that are not

What actually follows from all this

  1. Make the opening strong. It determines the test round, and the test round determines everything downstream.
  2. Optimise for effortful actions. Content worth saving or sending beats content worth liking, by a large margin in weighting terms.
  3. Be classifiable. A clear, consistent subject makes it easy to find your audience. An account covering five unrelated topics is genuinely harder to distribute.
  4. Post when your audience is around, because early engagement compounds.
  5. Ignore single-post variance. The test-audience draw varies; some good posts get unlucky rounds.
  6. Do not chase mechanics you cannot verify. Most algorithm folklore describes symptoms rather than mechanisms.
Put this into practiceTagly generates ranked, platform-tuned hashtags and keywords for any topic — free, no login, results in seconds.Open the generator →

Frequently asked questions

How do social media algorithms decide what to show?

They classify what your content is about, show it to a small test audience of followers and matching non-followers, measure watch time, saves, shares and scroll-away, then expand distribution in rounds if the response is strong.

Which engagement signals carry the most weight?

Watch time and completion on video, rewatches, shares, and saves. Signals are weighted roughly by how much effort they cost, which makes them harder to game and better correlated with genuine interest.

Why do the first minutes after posting matter?

Distribution expands in rounds, so early performance compounds. A post that does poorly in its first test round does not get a second, regardless of how a wider audience might have received it.

How much do hashtags affect the algorithm?

They contribute to classification, which affects which non-followers appear in your test audience. That influences who gets tested rather than whether the test succeeds — a real but modest role.

Is my reach down because of a penalty?

Usually not. Most unexplained drops are ordinary variance or content that performed less well. Genuine penalties are rare and mostly tied to policy violations.

Does posting about many different topics hurt reach?

It can, because it makes your account harder to classify. A clear consistent subject makes it easier for the system to identify and reach your audience.

Keep reading