一、流量池赛马分层算法（自然推荐核心算法）

1、 Traffic Pool Racing Hierarchical Algorithm (Natural Recommendation Core Algorithm)

这是大众非常熟悉，也是抖音全域流量的根基。视频完成审核之后，系统首先投放初始种子流量池（200-500 曝光），定向推送给少量潜在用户，收集真实行为数据。

This is the most familiar to the public, and it is also the foundation of Tiktok's global traffic. After the video is reviewed, the system first releases an initial seed traffic pool (200-500 exposures) and targets a small number of potential users to collect real behavioral data.

系统根据完播率、停留时长、点赞、评论、收藏、转发综合打分，开启残酷的同级赛马。数据优于同赛道同层级作品，视频晋级更高一级流量池，数据持续疲软，推荐量逐步收缩，自然流量进入停滞状态。

The system scores based on completion rate, dwell time, likes, comments, favorites, and shares to start a brutal race of the same level. The data is better than works of the same track and level, and the video has advanced to a higher level of traffic pool. The data continues to be weak, and the recommendation volume gradually shrinks, resulting in a stagnant state of natural traffic.

流量池大致层级：初始种子池→初级流量池（3000-5000 播放）→中级流量池（1 万 – 10 万）→出彩流量池（10 万 – 100 万）→爆款全域池（100 万以上）

Roughly speaking, the hierarchy of traffic pools is as follows: initial seed pool → primary traffic pool (3000-5000 plays) → intermediate traffic pool (10000-100000) → advanced traffic pool (10000-100000) → popular all domain pool (over 1 million)

运营要点冷启动阶段优先保住前 3-5 秒留存，中长视频不必盲目追求全程完播，核心提升中段停留、收藏与转发，不要频繁删除低数据视频，持续稳定输出优良内容，提升账号整体权重。

Operation points: During the cold start phase, priority should be given to maintaining the retention of the first 3-5 seconds. Medium and long videos do not need to blindly pursue complete broadcasting throughout the entire process. The focus should be on improving mid stream retention, bookmarking, and forwarding. Do not frequently delete low data videos, continuously and stably output high-quality content, and enhance the overall weight of the account.

二、标签兴趣匹配算法（合理垂直流量来源）

2、 Tag interest matching algorithm (precise vertical traffic source)

赛马算法决定视频能不能放大，标签匹配算法决定推给谁。系统通过 AI 识别视频画面、字幕、音频、标题、话题，提取内容特征，生成内容向量标签。

The horse racing algorithm determines whether the video can be enlarged, while the label matching algorithm determines who to push it to. The system uses AI to recognize video frames, subtitles, audio, titles, and topics, extract content features, and generate content vector labels.

同时持续抓取用户浏览、点赞、搜索、下单行为，生成用户兴趣标签。算法持续计算内容与用户的匹配相似度，把视频推送给兴趣高度契合的人群。

Simultaneously, continuously capture user browsing, liking, searching, and ordering behaviors to generate user interest tags. The algorithm continuously calculates the similarity between the content and the user, and pushes the video to a group of people whose interests are highly aligned.

2026 年算法持续升级，已经从 “静态标签匹配” 转向动态行为预测。不再单纯依靠关键词打标签，更看重用户刷到视频后的实时反应。很多创作者账号越做越泛、流量杂乱，根源就是内容方向摇摆，算法无法稳定锁定合理受众。

The algorithm will continue to upgrade in 2026, shifting from "static label matching" to dynamic behavior prediction. We no longer rely solely on keywords for tagging, but place greater emphasis on the real-time response of users when they come across a video. Many creator accounts are becoming more and more ubiquitous, with chaotic traffic, and the root cause is the swinging direction of content, with algorithms unable to stably target precise audiences.

运营要点账号内容保持赛道统一，标题、文案自然植入垂直关键词，避免一条美食、一条汽车、一条育儿交叉发布，打乱账号标签。减少无关热门话题堆砌。

The key points of operation are to keep the content of the account consistent with the track, naturally embed vertical keywords in the title and copy, and avoid cross posting of one food, one car, and one parenting, which will disrupt the account tags. Reduce the accumulation of irrelevant hot topics.

三、地域 & 时空推荐算法（本地号、时效内容核心流量）

3、 Regional and spatiotemporal recommendation algorithm (local account, core traffic of time sensitive content)

这套算法包含两大维度：地域定向、时间时效匹配。地域维度：基于用户定位信息，优先推送同城、同区域内容，是本地餐饮、探店、房产、同城服务账号的流量支柱。

This algorithm includes two major dimensions: geographical orientation and temporal matching. Regional dimension: Based on user positioning information, priority is given to pushing local and regional content, which is the traffic pillar of local catering, store exploration, real estate, and local service accounts.

时间维度：算法会根据全天不同时段用户偏好、节日节点、热点周期调整推荐倾向。深夜更容易推送情感、美食内容，工作日上午偏向职场、知识干货；节假日优先出游、探店内容。

Time dimension: The algorithm will adjust the recommendation tendency based on user preferences, holiday nodes, and hot periods at different times of the day. Late night is more likely to push emotional and food content, while weekday mornings tend to focus on the workplace and knowledge; Priority travel and store exploration during holidays.

不少创作者发现，同一条视频，工作日早上发布没流量，周末晚间发布轻松起量，本质就是时空算法在起作用。

Many creators have found that when the same video is released on weekday mornings with no traffic and on weekend evenings with ease, it is essentially due to the effectiveness of spatiotemporal algorithms.

运营要点本地商家视频带上同城定位，根据赛道人群作息选择发布时间，节日、热点节点提前布局内容，抓住时效流量红利。

Key points of operation: Local merchants should include local positioning in their videos, choose release times based on the daily routines of the track audience, and plan content in advance for holidays and hot spots to seize the benefits of timely traffic.

四、铁粉优先推送算法（账号长效稳定流量底盘）

4、 Iron powder priority push algorithm (account long-term stable traffic chassis)

铁粉算法是近几年抖音核心强化的分发机制，也是区分 “短期爆款” 和 “可持续账号” 的关键。视频发布之后，第一批流量优先分配给铁粉。铁粉定义：长期关注账号、经常看完视频、主动评论互动、频繁进入主页的粉丝。

The iron powder algorithm is a distribution mechanism that Tiktok has focused on strengthening in recent years, and it is also the key to distinguish between "short-term blockbusters" and "sustainable accounts". After the video is released, the first batch of traffic will be prioritized for fans. Definition of fan base: fans who follow their account for a long time, watch videos frequently, actively comment and interact, and frequently visit the homepage.

铁粉群体的数据表现，会直接影响视频能否进入更大的公域流量池。铁粉完播、互动数据越好，公域放大的概率越高。单纯依靠泛粉、路人流量爆火的视频，长尾流量通常很短；铁粉基础扎实的账号，更容易持续产出稳定流量。

The data performance of the fan group will directly affect whether the video can enter a larger public domain traffic pool. The better the completion and interactive data of iron fans, the higher the probability of public domain amplification. Relying solely on popular videos with fan base and pedestrian traffic, long tail traffic is usually very short; Accounts with a solid foundation of loyal fans are more likely to consistently generate stable traffic.

很多博主粉丝几十万，播放依旧低迷，核心问题就是粉丝量大，但铁粉占比出彩低，粉丝只是 “躺关注”，几乎没有互动。

Many bloggers have hundreds of thousands of followers, but their views are still sluggish. The core problem is that they have a large number of followers, but the proportion of fans is extremely low. Fans only "lie down and follow", with almost no interaction.

运营要点视频结尾引导老粉丝持续互动，固定更新时段，培养粉丝观看习惯，多回复评论，提升粉丝粘性，适度做粉丝向内容，巩固铁粉基本盘。

At the end of the operation key video, guide old fans to continue interacting, fix the update period, cultivate fans' viewing habits, reply to comments more frequently, enhance fan stickiness, moderately create fan oriented content, and consolidate the basic base of loyal fans.

五、搜索推荐算法（长效免费流量，被严重低估）

5、 Search recommendation algorithm (long-term free traffic, severely underestimated)

绝大多数创作者只关注首页推荐流，忽视搜索流量。搜索算法属于被动需求分发。用户主动在抖音搜索关键词，算法检索全网匹配视频，按照相关性、内容质量、更新时间排序展示。首页推荐流量波动出彩大，热点褪去流量快速下滑；优良搜索内容可以持续数月持续供应获得曝光，形成稳定的 “流量复利”，非常适合知识科普、教程、测评、行业干货账号。

The vast majority of creators only focus on homepage recommendation streams and ignore search traffic. Search algorithms belong to passive demand distribution. Users actively search for keywords in Tiktok, and algorithms retrieve matching videos across the network, which are displayed in order of relevance, content quality, and update time. The recommended traffic on the homepage fluctuates greatly, with hotspots fading and traffic rapidly declining; High quality search content can continuously gain exposure for several months, forming a stable "traffic compound interest", which is very suitable for knowledge popularization, tutorials, evaluations, and industry knowledge accounts.

运营要点标题、字幕、文案布局用户高频搜索关键词；回答赛道内用户常见问题；内容紧扣大众搜索需求，提升视频在搜索结果里的排名。

Key operational points: Title, subtitle, and copywriting layout. Users frequently search for keywords; Answer common questions from users on the track; The content closely follows the search needs of the public and improves the ranking of videos in search results.

六、付费助推推荐算法（付费流量撬动自然流量）

6、 Paid recommendation algorithm (paid traffic leverages natural traffic)

DOU+、千川付费流量背后，运行独立的付费推荐算法。付费算法可以人为设定人群标签、年龄、地域、兴趣，定向投放。

DOU+、 Behind Qianchuan's paid traffic, an independent paid recommendation algorithm runs. The payment algorithm can manually set audience tags, age, region, interests, and targeted advertising.

付费不只是单纯买播放，合理投放能够给视频注入初始数据，反向触发自然赛马算法，撬动免费自然流量。

Paying is not just about buying and playing, reasonable advertising can inject initial data into videos, trigger natural racing algorithms in reverse, and leverage free natural traffic.

但付费算法和自然算法相互独立，如果视频本身内容承载力不足，付费带来的数据很差，无法带动自然推流，只会单纯消耗预算。

However, paid algorithms and natural algorithms are independent of each other. If the content carrying capacity of the video itself is insufficient and the data brought by payment is poor, it will not be able to drive natural streaming and will only simply consume the budget.

运营要点冷启动数据合格之后再投放付费；不要盲目给所有视频投流；付费投放优先选择合理人群，避免泛流量造成无效消耗。

Operation points: After the cold start data is qualified, the payment can be released; Don't blindly stream all videos; Priority should be given to selecting targeted audiences for paid advertising to avoid ineffective consumption caused by excessive traffic.

六套算法联动逻辑，看懂流量全貌,一条视频完整传播链路：视频发布 → 系统推送铁粉（铁粉算法）+ 初始自然流量（赛马算法）→ 根据标签匹配推送给潜在兴趣人群（兴趣标签算法），同步开放同城流量（时空地域算法）→ 视频产生数据，达标则持续放大自然推荐→ 用户主动搜索，持续获取搜索长尾流量→ 可选付费投放，助推视频突破流量层级

Six sets of algorithm linkage logic, understanding the full picture of traffic, a complete video dissemination chain: video release → system push of iron fans (iron fan algorithm)+initial natural traffic (horse racing algorithm) → push to potential interest groups based on tag matching (interest tag algorithm), synchronously open local traffic (spatiotemporal regional algorithm) → video generates data, and if it meets the standard, it continues to amplify natural recommendations → users actively search and continuously obtain search long tail traffic → optional paid advertising, helping videos break through the traffic hierarchy

没有明显一条爆款，只依靠单一算法。大爆款往往同时激活多套推荐体系；而很多视频只有短暂流量，仅仅激活了浅层赛马流量，没有打通搜索、铁粉流量。

There are no best-selling products, relying solely on a single algorithm. Popular products often activate multiple recommendation systems simultaneously; However, many videos only have short-term traffic and activate shallow horse racing traffic, without connecting search and fan traffic.