How Does the TikTok Algorithm Decide Who Sees a Video? (Reels and Shorts Too)
How Does the TikTok Algorithm Decide Who Sees a Video?
TL;DR: TikTok’s current help page says For You ranks on your interactions (what you like, share, comment on, and “watch in full or skip”), content information (sounds, hashtags, view count, the country it was published in) and your device and location settings, with interactions such as watch time “generally weighted more heavily”. It doesn’t list follower count for For You, and in 2020 TikTok said follower count is not a “direct factor”. Instagram’s Reels card shortlists about 100 reels and ranks them on predictions such as whether you’ll watch under three seconds, watch longer than most viewers, reshare or follow. A peer-reviewed data-donation study found TikTok personalizes 30–50% of a new user’s first 1,000 recommendations and explores with the rest, which is the slot an unknown account can win.
Primores’ organic reach posts on schedule across TikTok, Instagram Reels and YouTube Shorts, answers the comments, and measures saves, shares, follows and comments per piece: organic reach.
⚠️ Verification note. TikTok’s help page “How TikTok recommends content” (undated) and Meta’s Instagram Reels Chaining system card (updated 11 Nov 2025) were read in a browser on 2026-11-10. TikTok’s 2020 Newsroom post and Instagram’s 2023 ranking post were read in full on 2026-11-09. No official YouTube text on Shorts ranking signals was found. Audit findings decay quickly; see the section on studies.
What each platform says officially
| Platform | Signals named | Follower count | Weighting disclosed | Source date |
|---|---|---|---|---|
| TikTok (current help page) | For You: user interactions (“Content you like, share, comment on, and watch in full or skip”, accounts of followers you follow back); content information (“Sounds, hashtags, number of video views, and the country in which the content was published”); user information (device settings, language preference, location, time zone and day, device type) | Not listed for For You. Listed only for LIVE (“number of followers of that creator”) and account recommendations | ”For most users, user interactions, which may include the time spent watching a video, are generally weighted more heavily than others”; the weight of each factor “can change over time” | Undated, read 2026-11-10 |
| TikTok (2020) | Interactions (likes, shares, follows, comments, content created), video information (captions, sounds, hashtags), device and account settings | ”neither follower count nor whether the account has had previous high-performing videos are direct factors” | Finishing a longer video “would receive greater weight than a weak indicator”; device and account settings get lower weight | 18 Jun 2020 |
| Instagram Reels | In rough order: your activity (reels liked, saved, reshared, commented on), your history with the poster, information about the reel (audio, visuals, popularity), information about the poster | Used: popularity signals such as “number of followers or level of engagement” | Order only, no weights | 31 May 2023 |
| Instagram Reels Chaining (system card) | A lightweight model selects “approximately 100 of the most relevant reels”, then ranks them by predicted value. Predictions include: watching less than three seconds, watching more of a reel “than 95% of users who watched reels of the same length”, commenting, resharing, sharing off Instagram, using the audio, following the author. Signals include reel length, time people spent viewing, skips within two seconds, external shares, and time spent on the reel by “viewers from your country” | One signal: “The amount of followers the author received through the reel” | Predictions and signals named, no weights | Updated 11 Nov 2025 |
| YouTube Shorts | No official ranking text found | Not stated | Not stated | n/a |
Instagram also says it makes reels less visible when they are low-resolution, watermarked, or “have already been posted on Instagram”. Reuse rules are covered on marketing/platform-originality-and-multi-account-rules.
TikTok’s help page also says it diversifies feeds with new creators, “typically” doesn’t recommend content you’ve already seen, and has safety teams review some content “as it rises in popularity”. Neither TikTok’s nor Instagram’s text gives a size for a first test audience, or says how long a video keeps being distributed.
What independent studies found
| Study | Method | Finding | Caveat |
|---|---|---|---|
| Vombatkere et al., “TikTok and the Art of Personalization” (WWW ‘24) | 347 users donated their data (9.2M recommendations); 220 users with 1,000+ recommendations analysed | TikTok “exploits users’ interests in between 30% and 50%” of the first 1,000 videos, 50% or more for real users (bots 31%, random baseline 20%); “liking and following are the primary drivers of personalization” | Measures the viewer’s feed, not a video’s views |
| Boeker & Urman, personalization factors on TikTok (WWW ‘22) | Sock-puppet audit | Following was the strongest factor, then view rate, then likes; viewer location strongly shaped the feed, more than language | 2022, web client |
| Mosnar et al., “Revisiting Algorithmic Audits of TikTok” (SIGIR 2025) | Re-ran earlier audits | Findings reproduce poorly and “often hold only in the short term” | Abstract only |
Our read: if roughly half of a settled user’s feed is exploitation and the rest exploration, exploration is how an account with no followers reaches anyone. That fits TikTok’s statement on follower count. No study measures how much of the exploration slot goes to unknown creators, or the view trajectory of a new video over time.
What this means for a new account
Our read, from the sources above:
- On TikTok, a new account can compete, because followers are not a direct factor and exploration is a large share of every feed. Total views still come from rare outliers (see marketing/short-form-view-distribution), so plan in batches of posts, not single videos.
- On Reels, followers help. Instagram names author popularity as an input, so a fresh Reels account starts behind a TikTok one.
- Watch time is what both platforms name. TikTok weights interactions such as watch time and “watch in full or skip” most; Instagram predicts whether you’ll watch under three seconds or longer than most viewers of reels the same length. Keep the hook and captions clear of the interface so nothing gives a viewer a reason to swipe early (see marketing/short-form-video-safe-zones).
- Location shapes the viewer’s feed, which matters for country targets (see marketing/organic-view-geography).
- Treat any audit number as dated. The SIGIR 2025 replication is the reason.
FAQ
Does follower count matter on TikTok?
TikTok said in 2020 that follower count and past high-performing videos are not direct factors in recommendations. Its current help page doesn’t list follower count among the For You factors; it lists it only for LIVE and for account recommendations.
Does follower count matter on Instagram Reels?
Yes, as one input. Instagram lists popularity signals including “number of followers or level of engagement” about the poster, while saying most of what people see in Reels comes from accounts they don’t follow.
Does TikTok test every video on 300–500 people first?
No official TikTok source names a test-batch size. The figure traces to uncited explainers.
What signals does TikTok weigh most?
Your interactions. TikTok’s help page says interactions, “which may include the time spent watching a video, are generally weighted more heavily than others”, and lists watching in full or skipping among them. In 2020 it said finishing a longer video counts for more than a weak signal. Studies of real users find liking and following drive personalization.
Is there a seed audience for YouTube Shorts?
Not in any YouTube text we found. The claim appears only in third-party summaries.
How long do algorithm audit findings stay valid?
Not long, per a 2025 replication study, which found TikTok audit findings reproduce poorly and often hold only in the short term.
Do not cite
| Claim | Why not |
|---|---|
| ”TikTok shows every new video to 300–500 (or 200–800) people first” | No official TikTok source names a test-batch size |
| ”YouTube tests each Short on a small seed audience” | Found only in vidIQ and blog summaries, not in YouTube’s words |
| ”Follower count doesn’t matter on any short-form platform” | True only of TikTok; Instagram lists follower count as a Reels input |
| WSJ 2021 and other 2021–22 audit numbers as current behaviour | Audit findings decay quickly (SIGIR 2025) |
| Ranking “formulas” with percentage weights | No platform publishes weights |
Key Takeaways
- TikTok (current help page): interactions, including watch time and watching in full or skipping, are generally weighted most; the publishing country is a content signal; follower count isn’t listed for For You.
- Instagram Reels card (Nov 2025): ~100-reel shortlist ranked on predictions such as sub-3-second watches, longer-than-typical watches, reshares and follows.
- Instagram (2023): Reels ranks on your activity, your history with the poster, reel information and poster popularity, including followers.
- No official YouTube Shorts ranking text was found.
- TikTok personalizes 30–50% of a new user’s first 1,000 recommendations; the rest is exploration.
- Audit findings age fast; cite them with their date.
Related
- marketing/short-form-view-distribution
- marketing/organic-view-geography
- marketing/organic-content-strategy
- marketing/short-form-video-safe-zones
- marketing/platform-originality-and-multi-account-rules
- marketing/discovery-before-scale
Sources
- TikTok Support, How TikTok recommends content — current For You, Following, LIVE and search factors; interactions weighted most; diversification. Undated. Tier 1, Primary read (browser, 2026-11-10).
- Meta Transparency Center, Instagram Reels Chaining AI system (updated 11 Nov 2025) — ~100-reel shortlist, prediction targets and signals. Tier 1, Primary read (browser, 2026-11-10).
- TikTok Newsroom, How TikTok recommends videos #ForYou (18 Jun 2020) — signals, completion weighting, follower count not a direct factor. Tier 1, Primary read.
- Instagram, Instagram Ranking Explained (31 May 2023) — Reels signal order, poster popularity incl. followers, most Reels from unfollowed accounts, demotion of reposts. Tier 1, Primary read.
- Vombatkere, Mousavi, Zannettou, Roesner, Gummadi, “TikTok and the Art of Personalization”, WWW ‘24 — 30–50% exploitation in the first 1,000 videos; liking and following drive personalization. Tier 1, Primary read.
- Boeker & Urman, “An Empirical Investigation of Personalization Factors on TikTok”, WWW ‘22 — following, view rate, likes; location stronger than language. Tier 1, Primary read.
- Mosnar, Skurla, Pecher et al., “Revisiting Algorithmic Audits of TikTok”, SIGIR 2025 — poor reproducibility, short-term validity. Tier 1, Partial (abstract).