1. The Rule Change Every Douyin Team Is Talking About
In September 2026, Douyin executed its broadest ecological update of the year. The changes span short-video creation, algorithmic distribution, livestream governance, MCN agency contracting, and Douyin Shop e-commerce rules — five tracks moving at once.
The headline item, and the one that has dominated creator discussion groups, is a traffic-weight reset that penalizes the very tactic the entire short-video training industry taught for years: the aggressive three-second hook.
For advertisers and agencies running organic or paid Douyin campaigns, this is not a platform-policy curiosity. Douyin remains the largest single attention pool in China. Its distribution formula determines organic reach, which in turn determines paid-media efficiency, creator selection, and content production budgets.
This article separates what is verifiable from what is circulating as industry estimate, and translates the change into actionable guidance.
Data-provenance note: The specific weighting figures widely cited for this update (completion rate 70% / interaction rate 20% / conversion rate 10%) originate from Douyin Creator Center announcement language as interpreted by industry analysts and MCN training content. Douyin has not published a formal weighting document confirming these exact percentages. Treat them as directional industry estimates. The directionally verifiable official facts are listed in Section 8.
2. The Weighting Shift, Plainly Stated
| Signal | Reported Weight | Notes |
|---|---|---|
| Completion rate (完播率) | 70% | "Overall completion," not 3-second or 5-second completion |
| Interaction rate (互动率) | 20% | Quality-weighted; saves and long comments rank above likes |
| Commerce conversion rate (带货转化率) | 10% | Lowest of the three — platform is not optimizing for GMV first |
A separate advancement threshold is widely reported: content whose overall completion rate exceeds 40% is more likely to be promoted into a higher traffic tier.
Two structural points matter as much as the numbers:
- "Completion" is being redefined. The metric is overall completion — the proportion of the video a user actually watched — not whether the first three seconds retained them. A video with a spectacular opening and a weak body no longer converts its early retention into distribution.
- Ranking inside the interaction bucket has an order. Industry interpretation consistently places deeper behaviors — saves, profile revisits, follower ("iron fan") interaction — above shallow ones such as a five-second hold. Likes sit at the bottom.
What this means for advertisers:
- Content briefs that mandate a shock-open or "you won't believe this" hook should be rewritten. The hook still has to avoid losing viewers, but it no longer earns distribution on its own.
- Creative teams should be briefed on overall watch-time proportion as the primary production KPI.
- A strong opening paired with a padded middle is now the highest-risk creative structure on the platform.
3. Precise-Interest-First Distribution: The New Cold Start
The cold-start mechanism has changed in a way that is arguably more consequential than the weights themselves.
| Stage | Previous Logic | New Logic |
|---|---|---|
| Initial push | Broad interest-tag pool | 100–300 vertical, precisely-matched users |
| What decides expansion | Early completion rate across broad reach | Completion and interaction within that small vertical cohort |
| Weight of casual passers-by | Moderate | Heavily suppressed |
The operating principle can be summarized as: traffic is a test quota, not a reward. The platform is no longer gambling broad reach on a potential hit. It first tests whether a narrow, knowledgeable audience accepts the content — and only then opens the tap.
This is the change most likely to hurt under-positioned accounts. If the system cannot determine who your content is for, it cannot assemble a precise test cohort — and the video stalls at a few hundred views regardless of production quality.
What this means for advertisers:
- Account positioning clarity is now a distribution prerequisite, not a branding nicety. Blurry, multi-topic brand accounts will find organic reach structurally harder.
- Vertical authority content outperforms broad-appeal content in cold start, even if its absolute ceiling looks lower.
- Paid amplification should be used to accelerate proven vertical content, not to rescue content the algorithm could not classify.
4. Deeper Behaviors Beat Shallow Ones: Why Saves Outrank Views
The strategic logic behind the weighting shift is worth examining, because it explains why the hook era is ending.
| Behavior | Depth | Can it be faked? | What it signals |
|---|---|---|---|
| Completion | Shallow | Yes — pacing tricks, withheld payoff | Attention captured |
| Like | Shallow | Yes — emotional reflex | Momentary approval |
| Save | Deep | Difficult | "This will be useful later" |
| Profile revisit | Deep | Difficult | "I trust this account, not just this video" |
| Substantive comment | Deep | Difficult | Content provoked real thought |
A completion rate is a shallow signal — easy to manufacture with reversals, clickbait titles, or by withholding the substance until the end. The viewer finishes and leaves, and never returns. Saves, revisits, and profile visits are costly behaviors that users only perform when they judge the content to be genuinely valuable. Deep behaviors are proxy signals for retention and total time-on-platform, which is ultimately what the platform monetizes.
What this means for advertisers:
- Content designed to be kept — tutorials, checklists, comparison frameworks, data references — gains a structural advantage over content designed to be watched once.
- Post level KPIs should shift from completion rate to average watch-time proportion.
- Comment-section sentiment is now a distribution variable. A comment section dominated by complaints or accusations of fraud can push a video into broad, low-intent audiences and erode account positioning.
5. Livestream Governance: From Single-Session Gambling to a 7-Day Assessment
Douyin's livestream assessment model has shifted from single-session performance to a continuous seven-day rolling evaluation, covering viewer dwell time, retention, content quality, negative feedback, and interaction validity.
A livestream health score governs access to platform privileges:
| Health Score | Consequence |
|---|---|
| 100 (initial); up to 120 for established large accounts | Normal operation, full recommendation access |
| Below 40 | Recommendation-list placement restricted |
| Below 20 | Gifting function disabled |
What this means for advertisers:
- Brands running brand-owned livestream rooms should manage to a seven-day rolling standard, not to a single hero session.
- Stable, predictable livestream scheduling is now rewarded over burst-campaign gambling. Rotation planning and host training should reflect this.
- Negative-feedback management (returns, complaints, moderation) is a direct input into the score, not a downstream customer-service concern.
- Because the health score also caps recommendation access, a degraded score raises the effective cost of paid traffic into that room.
6. AI Labeling and Compliance Red Lines
The September update tightened content governance across several fronts that directly affect advertisers using AI production tools.
The compliance red lines most relevant to marketing teams:
- AI content must disclose its identity on-screen. AI-generated imagery, editing, face-swapping, digital-human presenters, and AI-produced short dramas must carry a visible in-frame label indicating AI creation. Digital-human accounts must complete real-name registration. AI-produced short dramas require a label on every episode. Bulk, low-effort AI content produced at volume is identified and de-weighted.
- Off-platform traffic diversion faces zero tolerance. Directing users to WeChat, third-party links, or off-platform transactions — in videos, livestreams, comments, or direct messages — is a high-risk violation. Lead-generation funnels for financial products, part-time work, and AI courses are specifically named as priority enforcement targets. Confirmed violations can freeze commissions or permanently revoke commerce permissions.
- Dramatized fiction must be labeled in-frame. A bio-only disclosure does not satisfy the requirement. Enterprise accounts running scripted story content must now reserve label space within the frame itself — a change to content layout, not merely a line of copy.
- Homogenized and template-driven content is de-weighted. Matrix accounts built on bulk template output face direct penalties.
- Audit scope has expanded. Inspection now covers video frames, subtitles, captions, account bio, and comment sections — not just the video body.
Additionally, on September 15, 2026, Douyin begins cleaning up legacy personal yellow-V verification badges. This does not affect posting, livestreaming, or commerce permissions, and accounts can reapply after a badge lapses.
What this means for advertisers:
- AI production workflows must build label placement into the template, not into post-production review.
- Any existing funnel that routes Douyin traffic to WeChat groups or external landing pages should be audited immediately.
- Paid advertising landing pages are distinct from organic diversion — this rule set targets organic content and profile-based routing.
- KOL agreements should include clauses requiring AI-disclosure compliance and prohibiting off-platform diversion.
7. Market Context: Douyin's Attention Share Keeps Growing
The rule change lands against a backdrop of Douyin's accelerating dominance in Chinese consumer attention.
| Metric | Value | Source |
|---|---|---|
| Douyin share of total mobile internet time | 19.4% | QuestMobile, cited in a Nomura research note |
| WeChat share of total mobile internet time | 18.8% | Same |
| Douyin daily active users | 714 million | Same |
| WeChat daily active users | 931 million | Same |
| ByteDance Top-50 apps' combined share of user time | 40.9% | Same |
| Tencent Top-50 apps' combined share of user time | 29.1% | Same |
| Douyin year-over-year usage-time growth | ~29% | Same |
| WeChat year-over-year usage-time growth | ~3% | Same |
| Search-linked share of Douyin traffic | 22% | Platform disclosures |
| Creators reporting traffic gains from search | 76% | Platform disclosures |
| Monthly Douyin users who search | 80% | Platform disclosures |
| Average daily short-video usage time per user | 129 minutes (+28.4% YoY) | 《中国网络视听发展研究报告(2026)》 |
In July 2026, Douyin's total daily usage time surpassed WeChat's for the first time. WeChat retains its role as national infrastructure by daily active users, but Douyin has become the largest single attention inlet in the country.
What this means for advertisers:
- Douyin is no longer an optional channel for China-market brands; it is the primary organic attention pool.
- The growth of search-linked traffic means Douyin content increasingly functions as a discovery surface with a longer tail than the feed alone implies.
- Platforms tightening rules from a position of strength is a normal pattern: compliance cost rises, and unprofessional operators exit.
8. The Official Algorithm Direction vs. the Circulating Version
Because a substantial volume of "Douyin new rule" content circulating online is unaudited — and in some cases demonstrably fabricated — it is worth separating the verifiable official direction from the industry interpretation.
Officially verifiable:
- Douyin's stated north-star is long-term user value. Distribution is not decided by any single metric such as completion rate.
- The platform runs multi-objective modeling that weighs watch time, likes, comments, saves, shares, follows, long-term content value, content diversity, and creator experience together. Douyin has published a ten-episode algorithm explainer series jointly with Chinese Academy of Sciences researchers covering recall, ranking, multi-objective optimization, cold start, large models, evaluation, and governance.
- Multi-objective modeling explicitly exists to capture content whose completion rate is unremarkable but which generates saves, re-watches, and follow-up searches. In official explanation, a knowledge video with a low completion rate may still be high-value.
- Douyin's dual-tower model separates a user tower from a content tower, moving beyond simple tag matching. It has been deployed in Feed ranking.
- ByteDance's RankMixer architecture, published in July 2025, scaled parameters roughly 70-fold while holding inference cost flat, and is deployed across Douyin Feed ranking traffic.
- The evaluation window has been extended from a 24-hour sprint toward 7-day and multi-cycle assessment.
Caution on circulating claims: Widely forwarded "Douyin new rules" posts frequently cite contradictory effective dates, lack any corresponding official document, and are often published by parties selling courses or matrix-account tooling. Claims that "paid placement is required to avoid throttling" have been explicitly denied by Douyin's official clarification channel. Any weighting number should be treated as industry estimate unless reproduced in Douyin Creator Center or Douyin Shop Learning Center announcements.
9. Why This Matters Most for International Brands Running TikTok Playbooks
The commercial significance of this update extends well beyond domestic Chinese operators.
International brand teams often run TikTok on a content model built around the three-second hook, aggressive watch-time engineering, and completion-rate optimization. That playbook is now actively counterproductive on Douyin, even though both platforms share a corporate origin.
| Dimension | TikTok (international) | Douyin (September 2026) |
|---|---|---|
| Opening hook | Still rewarded; core retention mechanic | Neutral to negative; does not earn distribution alone |
| Primary optimization target | Watch time / completion | Overall completion, then deep behaviors (saves, revisits) |
| Cold start | Broad interest targeting | 100–300 precise vertical users |
| Success horizon | First hours | 7-day rolling window |
| In-frame AI labeling | Voluntary / platform-dependent | Mandatory for AI-generated content |
| Livestream assessment | Per-session | 7-day rolling health score |
Teams that port TikTok creative briefs directly into Douyin campaigns are optimizing for a metric the Douyin algorithm has stopped rewarding, and are exposed to compliance requirements that have no TikTok equivalent.
What this means for advertisers:
- Local creative teams should be authorized to deviate from global TikTok templates.
- Global brand-safety and AI-disclosure policies must be reconciled with Douyin's stricter in-frame labeling mandate.
- Agency and creator briefs should specify overall watch-time proportion and save-based objectives, not hook performance.
10. Action Plan for Advertisers
### Immediate (First 2 Weeks)
| Priority | Action |
|---|---|
| 🔴 Critical | Audit all AI-generated or AI-assisted content for in-frame labeling compliance |
| 🔴 Critical | Remove off-platform contact routing from videos, livestreams, comments, and bios |
| 🔴 Critical | Re-brief creative teams: the shock-open hook is no longer a distribution strategy |
| 🟡 High | Switch the primary content KPI from completion rate to average watch-time proportion |
| 🟡 High | Establish 7-day performance review cycles; stop judging content at hour 3 |
| 🟡 High | Verify digital-human account real-name registration status where applicable |
### Structural (Q4 2026)
| Priority | Action |
|---|---|
| 🟢 Medium | Rebuild content templates to reserve in-frame label space by default |
| 🟢 Medium | Refocus account positioning on vertical authority rather than broad topical range |
| 🟢 Medium | Redesign livestream operations around a 7-day rolling script and staffing model |
| 🟢 Medium | Shift creator selection criteria toward save rate, profile revisit rate, and follower quality |
| 🟢 Medium | Build search-oriented content into the organic plan, given search's growing share of traffic |
| 🟢 Medium | Reconcile global TikTok creative guidelines with Douyin's stricter local rules |
11. Key Takeaways
- The three-second hook has lost its structural advantage. Completion is still measured, but distribution is earned by what viewers do after watching, not by whether they were captured.
- Completion rate is re-emphasized at 70% weight, but "completion" now means overall completion, and the figure is an industry estimate rather than a published official number. The 40% advancement threshold is reported, not confirmed.
- Cold start now runs on 100–300 precise vertical users. Traffic is a test quota. Positioning clarity has become a distribution prerequisite.
- Deep behaviors outperform shallow ones. Saves, profile revisits, and substantive comments rank above likes and forced watch-through.
- Livestreams are assessed over a rolling 7 days, with a health score below 40 restricting recommendation and below 20 disabling gifting.
- AI disclosure must appear in-frame. Bio-only labeling is insufficient, digital-human accounts require real-name registration, and AI short dramas need per-episode labels.
- Off-platform diversion is a zero-tolerance red line, with finance, part-time-work, and AI-course lead generation named as priority targets.
- Douyin's share of Chinese attention is still expanding — 19.4% of total mobile internet time, ahead of WeChat's 18.8%, with 29% year-over-year growth.
- The TikTok playbook does not transfer. International brands must localize creative strategy and compliance posture rather than port global templates.