How ranking actually works on TikTok, Instagram, YouTube, LinkedIn, X, and Facebook — the top signals, how new content gets distributed, and what suppresses reach. Official platform statements are separated from third-party analysis, with every claim sourced.
The single ranking signal that matters most, per platform: TikTok — watch-time / completion rate; Instagram — watch time, likes-per-reach, and sends (DM shares); YouTube — viewer satisfaction (surveys + retention), not just watch time; LinkedIn — dwell time and substantive comments; X — replies and retweets (a retweet outweighs a like ~2×); Facebook — how likely the post is to spark a conversation among friends. Across every platform, original content beats reposted/watermarked content, and negative signals (hides, blocks, "not interested") suppress reach faster than positive signals lift it.
Most "algorithm explained" posts state third-party guesses as fact. This index tags every claim. TikTok, Instagram, YouTube, and Facebook publish official (if high-level) explanations of their ranking systems — those statements are marked OFFICIAL below. X open-sourced its ranking code in 2023, so its weights are real but dated (X is mid-migration to a Grok-based ranker). LinkedIn officially documents only dwell time; the rest of its model is third-party analysis. Where a mechanic is commonly repeated but not officially confirmed (TikTok’s test-cohort loop, the Instagram repost cap, the "80 billion signals" figure), it is flagged as third-party rather than presented as fact.
| Platform | Top signal | What suppresses reach |
|---|---|---|
| TikTok (For You feed) | In TikTok’s stated order: (1) user interactions — likes, shares, comments, and especially watch time / whether you finish a video; (2) content information — sounds, hashtags, captions; (3) device and account settings (weakest). | Content ineligible for the For You feed (e. |
| Instagram (Reels + Feed) | Instagram ranks each surface separately, but Adam Mosseri named three cross-surface signals (Jan 2025): watch time, likes-per-reach, and sends-per-reach (DM shares). | Instagram officially makes reels "less visible" when they are watermarked (e. |
| YouTube (long-form + Shorts) | Two official categories: Viewer Personalization (your watch history, search history, subscriptions, likes) and Content Performance (whether viewers click, watch, and positively engage). | "Not interested" / "don’t recommend this channel" feedback, dislikes, low satisfaction scores, misleading metadata/clickbait, and reused content all reduce recommendation. |
| LinkedIn officially documents one feed signal in depth: dwell time (how long a post holds attention). | Engagement bait ("Comment YES if you agree!"), and — per widespread third-party reporting, not official confirmation — outbound external links in the post body and low-dwell content. | |
| X (Twitter, For You) | From X’s open-sourced 2023 ranking code: replies and retweets carry the highest positive weight, then profile clicks / bookmarks / link clicks, then likes (lowest). | Blocks, reports, and mutes downrank hardest (several times the weight of a like). |
| Facebook (Feed) | Meta lists the inputs as: who created the post and how you’ve interacted with them before, the post format (photo/video/link), and how many of your friends engaged. | Engagement bait, clickbait, and links to low-quality/ad-farm sites are downranked (Meta integrity disclosures + third-party). |
Top ranking signals: In TikTok’s stated order: (1) user interactions — likes, shares, comments, and especially watch time / whether you finish a video; (2) content information — sounds, hashtags, captions; (3) device and account settings (weakest). TikTok says interaction signals "are generally weighted more heavily than others," and completion of a longer video carries more weight than a weak signal like shared location.
How new content gets distributed: The widely-repeated "new videos get tested on a small audience, then expand if completion is strong" loop is third-party characterization (Buffer, Hootsuite) — TikTok’s official support page does not describe a fixed test-cohort mechanic. Treat the specifics as inference, the direction as sound.
What suppresses reach: Content ineligible for the For You feed (e.g. extreme fitness/dieting), Community Guidelines violations, and "not interested" feedback. Third-party sources add watermarked/reposted and duplicate uploads.
The actionable takeaway: Optimize for completion rate above all. Hook in the first second and keep the video tight enough that viewers finish it — TikTok explicitly weights a full watch over almost everything else.
Sourcing: OFFICIAL: TikTok Newsroom + Support ("How TikTok recommends content"). The test-cohort loop is third-party (Buffer, Hootsuite).
Top ranking signals: Instagram ranks each surface separately, but Adam Mosseri named three cross-surface signals (Jan 2025): watch time, likes-per-reach, and sends-per-reach (DM shares). Likes-per-reach weighs more for reaching existing followers; sends-per-reach weighs more for reaching non-followers. Feed and Reels each layer your activity + your history with the poster on top.
How new content gets distributed: Instagram splits distribution into Connected Reach (followers) and Unconnected Reach (recommendations to non-followers). Reels expand into recommendation surfaces based on early watch time and sends. The "small test then expand" framing is third-party (Later, Buffer); the connected/unconnected split is official.
What suppresses reach: Instagram officially makes reels "less visible" when they are watermarked (e.g. a TikTok logo) or have already been posted on Instagram (recycled/aggregated). Original, unwatermarked content is favored. The specific "10+ reposts in 30 days = excluded" number is a third-party claim, not official.
The actionable takeaway: Post original, unwatermarked content; optimize the first ~3 seconds for watch-through; and engineer DM shares (sends) — that is the signal that unlocks reach to people who don’t follow you yet.
Sourcing: OFFICIAL: "Instagram Ranking Explained" + Mosseri (Jan 2025). The repost-cap number is third-party (Dataslayer).
Top ranking signals: Two official categories: Viewer Personalization (your watch history, search history, subscriptions, likes) and Content Performance (whether viewers click, watch, and positively engage). Crucially, YouTube officially weights satisfaction-survey feedback — not just raw watch time — so a video people rate highly can outperform one with more passive minutes.
How new content gets distributed: The homepage leans on watch history; "watch next" uses the current video as the primary signal. For Shorts, third-party analysis (Hootsuite, Shortimize) reports swipe-through rate, loop/replay rate, and shares replace click-through-rate as the core signals — you swipe, not click. The Shorts-specific mechanics are not in YouTube’s official Help text.
What suppresses reach: "Not interested" / "don’t recommend this channel" feedback, dislikes, low satisfaction scores, misleading metadata/clickbait, and reused content all reduce recommendation.
The actionable takeaway: Build genuine satisfaction (retention + positive feedback), not just clicks — YouTube uses survey feedback as a direct input. For Shorts, maximize loop/replay and swipe-through. Publish 2–3 hours before peak so the system can index and test before traffic arrives.
Sourcing: OFFICIAL: YouTube Help ("How YouTube recommendations work"). Shorts swipe/loop specifics and the "80 billion signals/day" figure are third-party (Hootsuite/Shortimize), not on official pages.
Top ranking signals: LinkedIn officially documents one feed signal in depth: dwell time (how long a post holds attention). Third-party analysis adds a three-stage flow — quality/spam filter → small-sample engagement test → network-relevance ranking — with thoughtful comments and relationships ranking high and likes ranking weak.
How new content gets distributed: A post is shown to a small slice of your network; strong first-hour engagement (especially comments and long dwell) expands it to 2nd- and 3rd-degree connections. This staged model is third-party (Hootsuite, AuthoredUp), not officially documented beyond the dwell-time post.
What suppresses reach: Engagement bait ("Comment YES if you agree!"), and — per widespread third-party reporting, not official confirmation — outbound external links in the post body and low-dwell content.
The actionable takeaway: Write posts that earn long dwell time and substantive comments. AuthoredUp’s dataset found posts with 61+ seconds of dwell hit ~15.6% engagement vs ~1.2% for 0–3 second dwell — the strongest reach lever on the platform.
Sourcing: OFFICIAL: only "Understanding feed dwell time" (LinkedIn Engineering). The 3-stage flow + 15.6% dwell stat are third-party (AuthoredUp/Hootsuite).
Top ranking signals: From X’s open-sourced 2023 ranking code: replies and retweets carry the highest positive weight, then profile clicks / bookmarks / link clicks, then likes (lowest). One widely-cited formula from the code weights a retweet at 1.0 and a like at only 0.5. Negative actions dwarf positives — a block scores roughly −3.0, a report ≈ −1.5.
How new content gets distributed: Roughly 500M daily posts are narrowed to ~1,500 candidates per user, drawn from in-network (people you follow) and out-of-network (interest-matched) pools. The exact 50/50 in/out split is third-party; the GitHub README states the search index supplies "~50% of posts."
What suppresses reach: Blocks, reports, and mutes downrank hardest (several times the weight of a like). Visibility filters can hard-filter or downrank. Third-party sources report external-link posts are deprioritized and X Premium gives a discovery boost (not officially quantified).
The actionable takeaway: Optimize for replies and retweets, not likes — and never post anything that triggers blocks/reports/mutes, since those negative weights overwhelm any positive engagement.
Sourcing: SOURCE: X open-source repo (twitter/the-algorithm) + code analysis — the official engineering blog now 403s, and X is migrating to a Grok-based ranker, so these legacy weights may be stale.
Top ranking signals: Meta lists the inputs as: who created the post and how you’ve interacted with them before, the post format (photo/video/link), and how many of your friends engaged. The system predicts how likely you are to comment, how likely your friends are to comment, and whether the post will spark a conversation.
How new content gets distributed: A four-step official process: gather all eligible posts → a lightweight model selects ~500 of the most relevant → heavier models score relevance/value → final ranking balances content types so formats don’t repeat. Meta stresses the models "are dynamic and change frequently."
What suppresses reach: Engagement bait, clickbait, and links to low-quality/ad-farm sites are downranked (Meta integrity disclosures + third-party). Outbound-link posts historically reach less than native content (third-party).
The actionable takeaway: Post content likely to spark genuine comments and conversation among friends — Meta explicitly predicts "how likely the post is to spark a conversation" as a ranking input.
Sourcing: OFFICIAL: Meta Transparency Center (Facebook Feed ranking). Link-reach penalties are third-party.
Primary sources are each platform’s own creator/engineering documentation, with corroborating analysis from Hootsuite, Buffer, Sprout Social, and AuthoredUp where the official record is thin:
Algorithms change constantly and platforms disclose only partial detail; this index is versioned for 2026 and re-audited quarterly. Where a claim is third-party, it is labeled so you can weight it accordingly.
Strip away the per-platform specifics and the same three rules hold everywhere: (1) the algorithm optimizes for completed, satisfying consumption, not raw exposure — finish rate, dwell time, and satisfaction beat impressions; (2) original, native, unwatermarked content is favored over reposts and cross-platform leftovers; (3) negative signals (hides, blocks, "not interested," low satisfaction) cut reach faster than positive engagement builds it. Build for the first rule, avoid the third, and post natively to each platform — which is exactly what repurposing per-platform (not cross-posting one asset) is for.
Read the autonomous content methodology → · Try Kompozy free · More research