Does “you” in a title,
really get more plays?标题里写「你」，
播放真的会更高吗？

Anyone who makes videos has heard this kind of advice: talk to the viewer in the title, use “you”. It sounds right; “you” makes a video feel personal. ViraLens was written to test advice like this. It pulls a creator’s public video data and checks whether a claim holds up in real videos, not in a single good example.做视频的人大概都听过这类建议：标题要跟观众说话，多用「你」。听起来很有道理，一个「你」字，让观众觉得这条视频是讲给自己听的。ViraLens 就是为了检验这类建议而写的：取回创作者的公开视频数据，看一个说法放到真实的作品里能不能成立，而不是只看某一个成功的例子。

The first test looked like a win.第一次检验，结果看起来很漂亮。

The retracted claim was not deleted. It stays in the report, marked “overturned”, next to the findings that survived. It was the third hypothesis the project has overturned.被撤回的结论没有删掉。它留在报告里，标着「推翻」，和站得住的结论放在一起。这是项目推翻的第三个假设。

So what does hold?那么，什么是站得住的？

Sort each creator’s videos by plays and compare the same person’s top 5 with their bottom 5. The gap runs from 3.7× to 109×.³ Where the gap is widest, the creator has almost always stepped outside their signature form: a sponsored read, a vlog, an interview, a livestream re-upload.把每位创作者的视频按播放排序，比较同一个人的头部 5 条和尾部 5 条，差距在 3.7 倍到 109 倍之间。³差距最大的地方，几乎都是创作者离开了自己的招牌形式：商单、vlog、访谈、直播回放。

The pattern showed up first in the knowledge zone. The identical test was then run in tech, daily life, food and gaming, and it held in all five. In May 2026 it was re-run on four English YouTube entertainment creators: gaps of 3.5× to 8.8×, with no counter-case.这个规律最早出现在知识区。随后用完全相同的检验在数码、生活、美食、游戏区重做，五个分区全部成立。2026 年 5 月，又在 4 位英文 YouTube 娱乐区创作者身上复测，差距为 3.5 到 8.8 倍，没有反例。

Two exceptions make the rule sharper. A tech creator’s sponsored video reached 9.8 million plays, because the ad was made to the same standard as the channel’s usual work. A knowledge creator’s livestream re-upload got 17.6 thousand; the topic was fine, the production was not.两个例外，反而把规律说得更清楚。一位数码区创作者的商单拿到 980 万播放，因为那条广告按他平时的水准制作；一位知识区创作者的直播回放只有 1.76 万播放，选题没有问题，制作没有达到招牌水准。

The ceiling depends on whether the work is finished to your own standard.上限取决于作品是否按自己的水准完成。

And the other tricks? Across 14 creators, almost every title or cover rule turned out to belong to individual creators.其他技巧呢？扩到 14 位创作者后，标题和封面上的规则几乎都只对个别创作者有效。

One cover measure, visual busyness, even pointed in opposite directions for two creators: −0.55 for one, +0.52 for the other. Same ruler, opposite sign.封面「繁杂度」这一项，在两位创作者身上方向完全相反：一位是 −0.55，另一位是 +0.52。同一把尺子，正负号相反。

Try it on the creators you care about用你关心的创作者试一试

List any Bilibili or YouTube creators in one file and ViraLens pulls their public data into a CSV and JSON you can open in Excel. Add the report option and it runs the full analysis above. Everything stays on your machine.在一个文件里列出任意 B 站或 YouTube 创作者，ViraLens 就会取回他们的公开数据，整理成可以用 Excel 打开的 CSV 和 JSON；加上报告选项，就会跑完上面这些分析。所有数据都留在你自己的电脑上。

• Desktop app桌面版Windows, macOS and Linux. Download, unzip, double-click; no Python needed.Windows、macOS、Linux。下载、解压、双击即可，不需要安装 Python。Latest release最新版本→
• Local web console本地网页控制台Paste keys, fetch, and open any video for a breakdown against that creator’s own hits.粘贴密钥、取数，点开任一视频，查看它与该创作者自身爆款的逐项对比。
  python scripts/app.pyCopy复制Copied已复制
• Command line命令行Just the data, or the data plus the report.只取数据，或连同报告一起生成。
  python scripts/viralens.py --reportCopy复制Copied已复制

Notes and limits注释与局限

1. Bilibili sample: 14 creators across five zones (knowledge, tech, daily life, food, gaming), 549 videos, public metadata only. Each zone has 2 to 6 creators, so these are signals, not laws.B 站样本：14 位创作者，覆盖知识、数码、生活、美食、游戏五个分区，共 549 条视频，只使用公开元数据。每个分区只有 2 到 6 位创作者，这些结果是信号，算不上规律。
2. Values read from second_person_falsified.png in the repository. That chart also shows two English YouTube creators, left out here to keep to the Bilibili sample.数值读自仓库中的 second_person_falsified.png。原图另有两位英文 YouTube 创作者，这里只保留 B 站样本。
3. Medians of each creator’s 5 most and 5 least played videos. “Off-format” is tagged by generic keywords (sponsored, vlog, interview, livestream), which mislabels a creator whose signature is vlogs.取每位创作者播放最高与最低各 5 条视频的中位数。「离开招牌形式」靠通用关键词标注（商单、vlog、访谈、直播），对以 vlog 为招牌的创作者会误标。
4. Public metadata cannot show causation: it cannot tell whether the algorithm rewards form or the audience simply expects it. Trends use videos at least 30 days old; old videos keep collecting views, so some bias remains.公开元数据无法说明因果，分不清是推荐算法偏好某种形式，还是观众本身有这种期待。趋势分析只用发布满 30 天的视频；老视频会持续积累播放，这一偏差无法完全消除。