We recently noticed inflated open rates across campaigns, and upon a detailed examination, we found that Gmail has been triggering false opens on emails since May of 2018.
Specifically, Gmail has been triggering opens via a bot as soon as an email is received by a Gmail user, and Gmail triggers the open from one of many Google IP addresses with this specific 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
By analyzing all the IP addresses that downloaded our open tracking pixel with this specific User Agent, we are confident that all the opens from this User Agent are false opens.
Additionally, opens from this User Agent happen within seconds after sending an email. We recently added a time designation to the “open” notifications that GMass users receive in the GMass Reports –> Opens Label. The Subject line of the notification shows how much time has elapsed between the email sending and the email opening. All of the “false” opens are recorded in mere seconds after the email is sent:
In GMass’s lifetime, as of today, approximately 307 million total email opens have been recorded across all campaigns across all users. Today’s analysis shows that this User Agent first started appearing in our open tracking logs in May of 2018 at a fairly infrequent rate of about 500,000 incidences per 20 million recorded opens, or a false open rate of 2.5%. This steadily increased over time to an incidence rate of 1.3 million per the 20 million most recently recorded opens, or a false open rate of 6.5%. That means that at most your open rates on past campaigns may go down by 6.5%, but likely less than 6.5%.
How good are you at percentage math? Assuming your open rate goes down by 5%, this does NOT mean that a campaign that previously had a 20% open rate will now have a 15% open rate. It means that a campaign that previously had a 20% open rate will now have a 19% open rate (19% is 5% less than 20%).
The Google Image Proxy
Several years ago, Google introduced the Google Image Proxy, which is a system by which images are downloaded in Gmail not from the source server, but via a proxy operated by Google. The whole concept of “open tracking” in email campaigns relies on a unique one-pixel image being downloaded.
The purpose of inserting the proxy between the end recipient and the email marketing service’s server was to obfuscate the location of the download, preventing companies like GMass and MailChimp from knowing the details of the “open”, such as the location and device used to open the email. The proxy still allowed ESPs to know IF an email was opened, however.
This new behavior is a separate issue from the Google Image Proxy, and Litmus has also found this to be the case.
The Google Image proxy uses a variety of User Agents, such as:
Mozilla/5.0 (Windows NT 5.1; rv:11.0) Gecko Firefox/11.0 (via ggpht.com GoogleImageProxy)
but always includes this bit:
(via ggpht.com GoogleImageProxy)
An open triggered by the Google Image Proxy does seem to be a real human opening an email, however an open triggered by the previously mentioned User Agent seems to be an open triggered by a Google bot.
The irony is therefore, that the User Agent that looks like a bot is actually a human, while the User Agent that looks like a human is actually the bot.
We have deleted millions of opens
In order to accurately convey open statistics to our users, we have deleted the approximately 4.5 million false open records from our system, and updated campaign reports will reflect this. If you notice your open rates drop on a campaign, it’s because we have deleted these false opens.
Going forward, we won’t count these as real opens
As of today, opens that originate from this User Agent will be ignored by GMass and not counted in your campaign reports.
Why did it take so long for us to notice?
First, wishful thinking. You see high open rates, and you want to believe it’s true. Why investigate something that’s making you look like an email marketing rockstar?
Secondly, the User Agent looks like a regular, normal User Agent. It looks like a real browser being used by a real human. Shame on Google for doing this in such a sneaky way.
Thirdly, it seems that not many other email marketing services have noticed either. We belong to a number of email marketing industry associations and discussion groups, and this issue hasn’t been mentioned at all. In fact, the only other mention of this issue I could find was in this Litmus discussion.
Other sources of false opens
Unfortunately, the mysterious Google bot isn’t the only source of false opens. It’s possible for you, the user to trigger a false open on behalf of your recipient as well, as we explain in this post on why your open tracking stats may be inaccurate.
The good news is that we now have a solution to this problem — such that if you open an email from your Sent Mail, or you open a bounce notification, an “open” will not be triggered.
Unrelated, but interesting SEO find
I’ve reported before how Google doesn’t rank its own content over others. This is the case with the Google Image Proxy. If you Google “Google Image Proxy”, you’ll see articles from x, y, and z, but the official Google announcement on the introduction of the proxy is all the way on page X! Go figure!