Gmail正在预取图片，导致打开率略微升高

ǞǞǞ recent changes with 苹果邮件隐私保护 had us wondering – where else is prefetching happening? While false opens come as no huge surprise, we have additional details around the limited set of circumstances in which Gmail is prefetching images in emails sent to Gmail users. 

Gmail预取的打开方式在以下情况下发生。

1. 一个Gmail收件人登录到Gmail应用程序（网页或移动应用程序），并有一个活动会话打开。

2. 当Gmail收件人的会话处于活动/开放状态时，会向其发送一封电子邮件。

3. Gmail会在用户界面显示邮件之前立即预取所有图片。

4. This image prefetch is in addition to (and different from) 谷歌图片缓存 opens, which occurs when the user opens the email.

ǞǞǞ image prefetch only occurs when the user is logged into the Gmail application, comes from a Google IP address, and is requested using the following user-agent string:

Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/42.0.2311.135 Safari/537.36 Edge/12.246 Mozilla/5.0

在调查数十亿次的打开事件中，我们可以自信地说，这些打开事件是虚假的打开，并不表示实际的用户打开事件。这些打开事件是独立的，不同于由谷歌图像缓存触发的用户发起的打开事件。

Gmail预取的影响

What are the implications of the false opens at Gmail? Thankfully, they are minor and nowhere near the Apple Mail 隐私 Protection scale. 

In looking at over 9.8B Gmail recipient open events in December 2021, for most senders, we saw that false opens accounted for between 1-6% of open events. What this means is your open rate may be inflated by up to 2 percentage points. For example: If you currently have an overall open rate of 20% at Gmail, your correct open rate would be closer to 18%.  

Your specific false open rate may be significantly higher or lower than what we are reporting above. Because, the false opens are triggered based upon when users are using the Gmail application, your specific audience’s behavior and use-cases are the primary factors in how much you will be impacted by this anomaly.  

如何检测和忽略Gmail预设的打开方式

For SparkPost senders, we have you covered. We have already updated our events API and event webhooks to automatically identify these Gmail Prefetch events using the 新引进的 is_prefetched 旗帜. We are also actively working to add the ability to distinguish prefetched and proxy opens in our Analytics Report UI and Metrics API. Stay tuned for future updates regarding the Report UI enhancements.

对于其他人来说，检测Gmail的预取打开事件仍然是比较直接的。对于每个打开事件，你要忽略（或唯一标记）任何符合以下用户代理字符串的打开事件。

Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/42.0.2311.135 Safari/537.36 Edge/12.246 Mozilla/5.0

We have been able to confirm that this string is unique to Google’s Prefetch Bot. 

Gmail Prefetch详细分析

As detailed above, Gmail prefetching only occurs in a limited set of circumstances. Prefetching does not occur with other mail clients. Instead, this behavior is specific to when a Gmail user has the Gmail app open within their web browser or is actively using the mobile app. Our best guess is that it’s a security scan before showing the email 到 user in their browser.

图片请求的完整请求标题详见下文。你会注意到几件事。

• The referer is set to http://mail.google.com. Interestingly, even though the user is on https://, Gmail still sets the referer 到 http:// protocol when making the request. 

• 该请求来自Gmail的服务器，而不是用户的浏览器。客户端IP总是解析到谷歌拥有的IP空间。

• 与谷歌图像缓存不同的是，用户代理字符串并不识别请求来自谷歌的一个机器人。相反，该用户代理字符串看起来像一个实际的用户图像请求。然而，我们已经证实，这个用户代理字符串确实识别了谷歌预取机器人。

• 开放请求发生在电子邮件交付的几秒钟内。此外，该请求发生在电子邮件出现在用户的Gmail界面之前。这种行为使我们相信该请求是出于安全目的。

• 预取似乎只发生在每个未读的Gmail邮件线程中。在我们广泛的测试中，一旦用户阅读了一个邮件，任何未来进入该线程组的邮件都不会启动预取请求。

• 这种预取是独立于谷歌图像缓存的。我们的测试表明，即使在图片被预取后，当用户打开邮件时，也会有一个单独的谷歌图片缓存请求。

• 如果用户打开了Gmail移动应用，预取将在短时间内继续发生，甚至在关闭移动应用之后。

下面是一个例子，当从Google Prefetch Bot请求一张图片时，请求标题会是什么样子。

  headers: {

    host: ‘{redacted}.m.pipedream.net’,

    ‘x-amzn-trace-id’: ‘Root={redacted}’,

    ‘accept-language’: ‘en-US’,

    referer: ‘http://mail.google.com/’,

    accept: ‘image/avif,image/webp,image/apng,image/svg+xml,image/*,*/*;q=0.8’,

    from: ”,

    ‘user-agent’: ‘Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/42.0.2311.135 Safari/537.36 Edge/12.246 Mozilla/5.0’,

    ‘accept-encoding’: ‘gzip, deflate, br’

   },

   body: {},

   inferred_body_type: ‘FORM’,

   method: ‘GET’,

   url: ‘https://{redacted}.m.pipedream.net/header-1641a1.gif’,

   client_ip: ‘66.249.92.1’,

   query: {}

正如苹果公司的邮件隐私保护一样，发件人应谨慎对待所有打开事件。打开只是众多参与度指标中的一个，而且往往不是最好的一个，发件人在确定用户参与度时应该监测并包括在内。