How Social Media Algorithms Work to Curate Your Feed

Before you scroll any further, open your phone. TikTok, WhatsApp, Xiaohongshu. Pick one. The first post you see is not there because it is important or accurate. It is there because the ranking system predicted it would stop your thumb.

Every major social media service in Singapore optimises for one metric: time on screen. The longer you watch, tap, forward, or comment, the more valuable your attention becomes to advertisers. The system has no concept of truth. It has a concept of watch time, dwell time, and click-through rate. That is all. This guide shows you the specific signals each service uses to build your feed, and gives you a concrete way to see the ranking system at work on your own phone today.

Every recommendation system you encounter follows the same pipeline. First, the service builds an inventory of all available posts. Then it scores each piece against thousands of signals per user: who you follow, how long you watched the last video, what you searched for, what device you use. Finally, it ranks the inventory by predicted interaction and serves you the top-scoring items. Meta's 2023 documentation describes this as inventory, signals, predictions, and a relevance score. TikTok's 2020 description uses user interactions, video information (captions, sounds, hashtags), and device and account settings.

The result is a personalised feed unique to you. Two friends in the same Singapore neighbourhood can open TikTok's For You page at the same moment and see entirely different material. That is the ranking system doing its job: it has learned what keeps each of you watching.

TikTok's recommendation system, described in its 2020 documentation, starts with three categories of input: user interactions (what you like, share, follow, and how long you watch each video), video information (captions, sounds, hashtags, and trending topics), and device and account settings (language preference, country setting, device type). The system does not care whether a video is true. It cares whether you watch it to the end.

Dwell time and watch time are the dominant signals. A video that holds a viewer for 30 seconds outranks one that loses them at 5 seconds, regardless of accuracy. Emotionally charged material, surprising claims, and fast cuts perform well because they trigger the response the system rewards. Users under 18 in Singapore, since 2023, get an additional safety layer applied to their default feed, but the core optimisation remains watch time.

How To See The Ranking System At Work On TikTok

Open your TikTok For You page and note the first five videos. Ask a friend to do the same on their own phone. The overlap will be small or zero. That is not a glitch. It is the system showing each of you the posts it predicts will keep you watching. To reset this personalisation, use TikTok's recommendation reset feature, introduced in 2023. It restarts your For You feed from scratch.

WhatsApp and Telegram do not have visible recommendation feeds like TikTok. But they have a curation system that is equally powerful: the forward chain. In Singapore, 83.7% of adults use WhatsApp, and 44.5% use Telegram, making them the most-used messaging services in the country. Material spreads through broadcast lists, group chats, and channels. The mechanism that decides what you see is the social graph: who is in your group, who forwards you messages, and how many degrees of separation a piece of material has travelled.

Meta's 2021 and 2023 studies on amplification found that posts reshared in rapid succession get amplified, but reshare chains of two or more degrees now receive a distribution penalty. The service actively demotes material that has been forwarded multiple times, which is exactly the behaviour pattern of viral misinformation. The forwarded-as-received label on WhatsApp is a structural feature that indicates the message did not originate with the sender. That label is your first signal to apply lateral reading before believing or sharing.

What The System Cannot Tell You

WhatsApp and Telegram do not disclose the full ranking signals for broadcast lists or channel recommendations. Research is limited. What is known is that channel subscriptions are the primary signal: you see posts from channels you joined. The curation happens at the point of discovery, where the service may suggest channels based on your existing subscriptions, location, or behavioural data. Telegram's mechanism for channel discovery is opaque.

Xiaohongshu, or Little Red Book, is popular in Singapore for lifestyle, travel, and product recommendations. Its explore feed uses a content graph rather than a pure social graph. The system maps relationships between pieces of material based on keywords, image tags, user behaviour, and interaction patterns. If you search for a specific restaurant in Singapore, the system does not just show you that restaurant. It shows you related posts: nearby cafes, similar cuisines, travel tips from the same neighbourhood.

The signals Xiaohongshu uses include dwell time on each post, saves, shares, comments, and the specific keywords you search. The service's documentation is sparse, and independent research is limited. What is clear is that the system optimises for interaction through the content graph, creating a filter bubble that narrows your exposure over time. A user who searches for one type of material will gradually see less of anything else, because the system predicts that the content graph around your existing interests will keep you scrolling.

How To See The Xiaohongshu System

Search for a specific, narrow topic, such as a particular dish at a particular hawker centre. Scroll your explore feed for the next ten posts. Count how many are directly about that dish versus how many are about related but different topics. The system is showing you the content graph: it has learned the connections between your search terms and other material that similar users engaged with. If you see only the same narrow set of recommendations, the filter bubble is active.

Every system described above serves the same business model. The service sells access to user attention to advertisers. Meta, TikTok, and YouTube all generate the majority of their revenue from advertising. The ranking system is the factory that packages attention into predictable, measurable units. It optimises for interaction because interaction is what advertisers pay for.

This is why the system does not care about accuracy. It cares about click-through rate, watch time, and dwell time. A false claim that keeps you watching for 30 seconds is more valuable to the service than a true claim that you scroll past in 2 seconds. The consequence is social sorting, filter bubbles, and echo chambers. The system feeds you material that confirms what you already believe, because confirmation bias keeps you engaged. A 2021 Meta study showed that posts reshared in rapid succession get amplified; a 2023 update added a distribution penalty for reshare chains of two or more degrees, but the underlying optimisation remains interaction.

What This Means For You In Singapore

When you see a viral claim on TikTok, a forwarded message on WhatsApp, or a recommended post on Xiaohongshu, ask yourself one question: does this keep me watching or scrolling? If the answer is yes, that is exactly why the system showed it to you. The system has no opinion on whether the material is true. It only knows that it held your attention.

The failure case is emotional override. A forwarded message on WhatsApp triggers anger, fear, or hope, and you share it before verifying. The system has already optimised for that emotional trigger: it showed you material designed to provoke a strong response because strong responses produce interaction. Pause before sharing. Open a new tab. Search the claim. Check Factually, CheckMate, or Sure Anot. Apply lateral reading. The system will keep showing you what holds your attention. Only you can decide whether to verify before you forward.