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

Delivering to Gmail

Gmail is a challenge for even the best senders these days.
With the recent Gmail changes there isn’t any clear fix to getting open rates or inbox delivery back up. Some of it depends on what is causing Gmail to filter the mail. Changing subject lines, from name, from address may get mail back to the inbox in the short term, but it only works until the filters catch up.
What I am seeing, across a number of clients, is that Gmail is doing a lot of content reputation and that content reputation gets spread across senders of that content.  That means you want to look at who is sending any mail on your behalf (mentioning your domain or pointing at your website) and their practices. If they have poor practices, then it can reflect badly on you and result in filtering.
From what I’ve seen, these are very deliberate filtering decisions by Google. And it’s making mail a lot harder for many, many senders. But I think it is, unfortunately, the new reality.

Read More

ISPs speak at M3AAWG

Last week at M3AAWG representatives from AOL, Yahoo, Gmail and Outlook spoke about their anti-spam technologies and what the organizations were looking for in email.
This session was question and answers, with the moderator asking the majority of the questions. These answers are paraphrased from my notes or the MAAWG twitter stream from the session.
What are your biggest frustrations?
AOL: When senders complain they can’t get mail in and we go look at their stats and complaints are high. Users just don’t love that mail. If complaints are high look at what you may have done differently, content does have an effect on complaints.
Outlook: When we tightened down filters 8 years ago we had to do it. Half of the mail in our users inbox was spam and we were losing a steady number of customers. The filter changes disrupted a lot of senders and caused a lot of pain. But these days only 0.5% of mail in the inbox is spam.  Things happen so fast, though, that the stress can frustrate the team.
Gmail: Good senders do email badly sometimes and their mail gets bulked. Senders have to get the basic email hygiene practices right. Love your users and they’ll love you back.
What’s your philosophy and approach towards mail?
AOL: There is a balance that needs to be struck between good and bad mail. The postmaster team reminds the blocking team that not all mail is bad or malicious. They are the sender advocates inside AOL. But the blocking team deals with so much bad mail, they sometimes forget that some mail is good.
Yahoo: User experience. The user always comes first. We strive to protect them from malicious mail and provide them with the emails they want to see. Everything else is secondary.
Gmail: The faster we stop spam the less spam that gets sent overall. We have highly adaptive filters that can react extremely quickly to spam. This frustrates the spammers and they will give up.
Outlook: The core customer is the mailbox user and they are a priority. We think we have most of the hardcore spam under control, and now we’re focused on personalizing the inbox for each user. Everyone online should hold partners accountable and they should expect to be held accountable in turn. This isn’t just a sender / ESP thing, ISPs block each other if there are spam problems.
What are some of your most outrageous requests?
We’ve been threatened with lawsuits because senders just don’t want to do the work to fix things. Some senders try to extort us. Other senders go to the advertising execs and get the execs to yell at the filtering team.
Coming to MAAWG and getting cornered to talk about a particular sender problem. Some senders have even offered money just to get mail to the spam folder.
Senders who escalate through the wrong channels. We spent all this money and time creating channels where you can contact us, and then senders don’t use them.
Confusing business interests with product interests. These are separate things and we can’t change the product to match your business interest.
What are your recommendations for changing behaviors?
Outlook: We provide lots of tools to let you see what your recipients are doing. USE THE TOOLS. Pay attention to your recipient interaction with mail. Re-opt-in recipients periodically. Think about that mail that is never opened. Monitor how people interact with your mail. When you have a problem, use our webpages and our forms. Standard delivery problems have a play book. We’re going to follow that playbook and if you try to get personal attention it’s going to slow things down. If there’s a process problem, we are reachable and can handle them personally. But use the postmaster page for most things.
Gmail: Get your hygiene right. If you get your hygiene right, deliverability just works. If you’re seeing blocking, that’s because users are marking your mail as spam. Pay attention to what the major receivers publish on their postmaster pages. Don’t just follow the letter of the law, follow the spirit as well. Our responsibility, as an ISP, is to detect spam and not spam. Good mailers make that harder on us because they do thinks that look like spammers. This doesn’t get spammer mail in more, it gets legitimate mail in less. Use a real opt-in system, don’t just rely on an implied opt-in because someone made a purchase or something.
Yahoo: ESPs are pretty good about screening their customers, so pay attention to what your ESPs are saying. Send mail people want. Verify that the email addresses given to you actually belong to people who want your mail. Have better sender practices.
What do you think about seed accounts?
The panel wasn’t very happy about the use of seed accounts. Seeds are not that useful any longer, as the ISPs move to more and more personalized delivery. Too much time and too many cycles are used debugging seed accounts. The dynamic delivery works all ways.
When things go wrong what should we do?
AOL: Open a ticket. We know we’ve been lax recently, but have worked out of our backlog and are caught up to date. Using the ticketing system also justifies us getting more headcount and makes everyone’s experience better. Also, don’t continue what you’re doing. Pausing sending while you’re troubleshooting the issue. We won’t adjust a rep for you, but we may be able to help you.
Gmail: Do not jump the gun and open a ticket on the first mail to the spam folder. Our filters are so dynamic, they update every few minutes in some cases. Be sure there is a problem. If you are sure you’re following the spirit and letter of the sender guidelines you can submit a ticket. We don’t respond to tickets, but we work every single one. When you’re opening a ticket provide complete information and full headers, and use the headers from your own email address not headers from a seed account. Give us a clear and concise description of the problem. Also, use the gmail product forum, it is monitored by employees and it’s our preferred way of getting information to the anti-abuse team. Common issues lots of senders are having will get addressed faster.
Outlook: Dig in and do your own troubleshooting, don’t rely on us to tell you what to fix. The support teams don’t have a lot of resources so use our public information. If you make our job harder, then it takes longer to get things done. But tell us what changes you’ve made. If you’ve fixed something, and tell us, our process is different than if you’re just asking for a delisting or asking for information. When you’ve fixed things we will respond faster.
How fast should users expect filters to respond after making changes?
Filters update continually so they should start seeing delivery changes almost immediately. What we find is people tell us they’ve made changes, but they haven’t made enough or made the right ones. If the filters don’t update, then you’ve not fixed the problem.

Read More

IP reputation and email delivery

IP reputation is a measure of how much wanted mail a particular IP address sends.  This wanted mail is measured as a portion of the total email sent from that IP. Initially IP reputation was really the be all and end all of reputation, there was no real good way to authenticate a domain or a from address. Many ISPs built complex IP reputation models to evaluate mail based on the IP that sent the mail.
These IP reputation models were the best we had, but there were a lot of ways for spammers to game the system. Some spammers would create lots of accounts at ISPs and use them to open and interact with mail. Other spammers would trickle their mail out over hundreds or thousands of IPs in the hopes of diluting the badness enough to get to the inbox. Through it all they kept trying to get mail out through reputable ESPs, either by posing as legitimate customers or compromising servers.
These things worked for a while, but the ISPs started looking harder at the recipient pool in order to figure out if the interactions were real or not. They started looking at the total amount of identical mail coming from multiple IP addresses. The ISPs couldn’t rely on IP reputation so they started to dig down and get into content based filtering.
As the ISPs got better at identifying content and filtering on factors other than source IP, the importance of the IP address on inbox delivery changed. No longer was it good enough to have a high reputation IP sending mail.
These days your IP reputation dictates how fast you can send mail to a particular ISP. But a high reputation IP isn’t sufficient to get all the mail in the inbox. It’s really content that drives the inbox / bulk folder decisions these days.
 
Generally IPs that the ISP has not seen email traffic from before start out with a slight negative reputation. This is because most new IPs are actually infected machines. The negative reputation translates to rate limiting. The rate limiting minimizes people getting spam while the ISP works out if this is a real sender or a spammer.
Some ISPs put mail in the inbox and bulk foldering during the whitelisting process. In this case what they’re doing is seeing if your recipients care enough about your mail to look for it in the bulk folder. If they do, and they mark the mail as “not spam” then this feeds back to the sender reputation and the IP reputation.
If you’re seeing a lot of bulk foldering of mail, it’s unlikely there’s anything IP reputation based to do. Instead of worrying about IP reputation, focus instead on the content of the mail and see what you may need to do to improve the reputation of the domains and URLs (or landing pages) in the emails.

Read More