Gmail filtering in a nutshell

Gmail’s approach to filtering; as described by one of the old timers. This person was dealing with network abuse back when I was still slinging DNA around as my job and just reading headers as a hobby.

Gmail uses a 10+ year old neural network that analyzes thousands of factors, related to email, IP, and web, integrated with all Google products, and with 99.9%+ accuracy for identifying certain types of messages, combined with an email-specific domain-based reputation system that combines IP reputation, content, read rates, reputation of other senders with similar content.

This excerpt was shared with a bunch of delivery experts and every one of them agreed. The Gmail filters are incredibly complex and they measure thousands of different things about email. Yes, sometimes you can remove a link or a URL and get mail to the inbox for a while. That doesn’t mean the block was against the URL, simply that changing the URL changed the score enough for the mail to go to the inbox.
This is part of what makes Gmail delivery issues so difficult to troubleshoot. There isn’t one thing, it’s all the things that contribute to where an email ends up. We, as senders and deliverability experts, don’t have access to the Gmail data. The poster goes on to say:

Trying to fix this using only inaccurate proxy data where there is no mediation pathway in a matter of weeks is complex.  We consume data from a multitude of sources, compile and analyze the data, determine which of the hundreds of factors we can influence should be adjusted, come up with the easiest plan to address the most influential factors, and explain that to the customer using the clearest language possible to individuals who are not educated on the definition of a complaint.

We do our best, with limited data and try and tell you how to fix things.
One of the biggest challenges with Gmail delivery is I am convinced they look at your profile of recipients. They can map someone who is collecting addresses through third parties, or buying lists based on the specific Gmail accounts targeted by a mailing. Gmail has publicly stated and has on their website that they don’t think co-reg or purchased lists are opt-in. They have the technology and ability to track that. I think it’s one reason senders trying to use email for acquisition have such a challenge getting into the inbox and Gmail. I think it’s a feature, not a bug for them.
 

Related Posts

Thoughts on Gmail and the inbox

Over the last few months more and more marketers are finding their primary delivery challenge is the Gmail inbox. I’ve been thinking about why Gmail might be such a challenge for marketers. Certainly I have gotten a lot of calls from people struggling to figure out how to get into the Gmail inbox. I’ve also seen aggressive domain based filtering from Gmail, where any mention of a particular domain results in mail going to the bulk folder.
It’s one of those things that’s a challenge, because in most of these cases there isn’t one cause for bulk foldering. Instead there’s a whole host of things that are individually very small but taken together convince Gmail that the mail doesn’t need to be in the inbox.
A pattern that I’m starting to see is that Gmail is taking a more holistic look at all the mail from a sender. If the mail is connected to an organization, all that mail is measured as part of their delivery decision making. This is hurting some ESPs and bulk senders. I’ve had multiple ESPs contact me in the last 6 months looking for help because all their customer emails are going to bulk folder.
Gmail’s filtering is extremely aggressive. From my perspective it always has been. I did get an invite for a Gmail account way back in the day. I moved a couple mailing lists over to that account to test it with some volume and discussion lists. I gave up not long after because no matter what I did I couldn’t get gmail to put all the mail from that list into the tag I had set up for it. Inevitably some mail from some certain people would end up in my spam folder.
Gmail has gotten better, now they will let you override their filters but give you a big warning that the message would have been delivered to spam otherwise.
Gmail_NotSpam
What are mailers to do? Right now I don’t have a good answer. Sending mail people want is still good advice for individual senders. But I am not sure what can be done about this ESP wide filtering that I’m starting to see. It’s possible Gmail is monitoring all the mail from a particular sender or ESP and applying a “source network” score. Networks letting customers send mail Gmail doesn’t like (such as affiliate mail or payday mail, things they mentioned specifically at M3AAWG) are having all their customers affected.
I suspect this means that ESPs seeing problems across their customer base are going to have to work harder to police their customers and remove problematic mail streams completely. Hopefully, ESPs that can get on the Gmail FBL can identify the problem customers faster before those customers tank mail for all their senders.

Read More

July 2015: The Month in Email

Once again, we reviewed some of the ways brands are trying (or might try) to improve engagement with customers. LinkedIn, who frequently top lists of unwanted-but-legitimate email, announced that they’ll be sending less mail. Josh wrote about giving subscribers options for both the type and frequency of messages, and about setting expectations for new subscribers. In each case, it’s about respecting that customers really want to engage with brands in the email channel, but don’t want the permission they’ve granted to be abused. I also wrote a brief post following up on our June discussion on purchased lists, and as you’d predict, I continue to discourage companies from mailing to these recipients.

Read More

Thoughts on Gmail filtering

Gmail has some extremely complex filters. They’re machine learning based and measure hundreds of things about incoming mail. The filters are continually adjusting to changes and updating how they treat specific mail.
One consequence of continually adjusting machine learning filters is that filtering is not static. What passes to the inbox now, may not pass in a couple hours.
One of the other challenges with Gmail filters is that they look at all the mail mentioning a particular domain and so affiliate mail and 3rd party mail can affect delivery of corporate mail.
The good news is that continually adjusting filters adapt to positive changes as well as negative ones. In fact, I recently made a segmentation suggestion to a client and they saw a significant increase in inbox delivery at Gmail the next day.
Gmail can be a challenge for delivery, but send mail users want and mail does go to the inbox.

Read More