Increasingly over the last few months I’ve been seeing questions from folks struggling with reputation at Gmail and inbox delivery. It seems like everything exploded in the beginning for 2019 and everything changed. I’ve been avoiding blaming it all on TensorFlow, but maybe the addition of the new ML engine really did fundamentally change how things were working at gmail.
What folks are seeing
- Cutting back to engaged only users is not effectively improving reputation.
- Inboxing is no longer directly tracking with reputation on GPT (high reputation mail is going to bulk, low reputation mail is going to inbox).
- Recipients are complaining about mail being misfiltered.
What can we do?
Right now, we’re mostly falling back on the things that worked 6 months ago: cut back sending to the most engaged recipients and then gradually add back in other addresses once you’re back in the inbox. The challenge is we’re not seeing the improvements we’ve become accustomed to seeing when using this strategy.
With one of my clients their reputation, as reported on GPT, actually fell during the period of cutbacks. Based on consistently high open rates and various inbox tests, including my own and those by one of the commercial vendors, we determined that recipients were getting mail in the inbox despite the low reputation.
Other delivery experts have reported similar patterns. Horrible domain and IP reputation (sometime in the deepest, darkest red) but reaching the inbox and seeing open rates in the 20 – 30% range.
I’ve also seen the flip side, green IPs and domain rep, but the mail is going to bulk.
That’s frustrating.
Yup. Sorry. Wish I had better news. Right now we’re falling back to what worked 4 months ago, because it’s what we had and it did work.
One thing that is new information to me is that, according to folks inside Google, the domain and IP reputation showing on google postmaster tools includes all domains handled by gmail, including GSuite hosted domains. Maybe these are having a disproportional effect on reputation, I don’t know.
My current thoughts are:
- Pay attention to your engagement and open rates at Gmail
- Don’t worry about domain and IP rep too much, if your marketing is performing.
- Maybe we need to start including G Suite hosted domains in engagement restrictions, as painful as that’s going to be.
Anyone got any insight? Is there something we’re missing here?
RSS - Posts
One pattern we are starting to see as well:
Larger volume senders, that send consistent volume every day, tend to avoid the reputational swings more easily. Even with various audience expansions mixed in, reputation is not moving down.
Smaller volume senders, that ebb and flow volume, are moving reputation statuses frequently and are having a tough time rebounding out of a Low/Bad reputation when it happens.
At worst, we would see things move after 30 or 60 days of aggregation periods being pushed out. Not so much these days.
I’m noticing that consistency matters but only if you’re sending to engaged individuals.
Cutting to engaged individuals does help but it’s a deep cut, past three months. That does improve inbox placement and thus open rates. In about 1 month I’ve seen an improvement from 6% open rate to now 22% open rate and now we can add back other addresses to expand the universe slowly.
This is one of those things I have noticed. Smaller senders are really getting the short end of the stick. I blogged about it a little bit recently (https://wordtothewise.com/2019/04/email-filters-and-small-sends/). But I’ve been continuing to think about how to address and resolve this problem.
Part of the issue may be that the bad guys figured out smaller sends were better and so the filters NEED to address the small sends in ways they haven’t before. Take your average B2B “I bought a list of highly qualified leads” spammer. They send under 1000 emails a day, unless they’re using multiple sending accounts but they’re still in the <10K range. These messages regularly get delivered to the inbox, but they're highly annoying to those folks who are attractive targets. The spam runs lack a lot of the signals that filters have relied on in the past, so new things are being tested to see what works and what doesn't. Likewise, a lot of the criminal mail (spearphising, for instance) is going out in very low volumes.
Bulk filters don't cope with this kind of mail well, but this type of mail needs to be blocked to protect users. Basically, we're seeing a lot of things being tested and winnowed down to what will work. The only real question is how long will it take and what will the end result look like? (answer: as long as it takes and if we knew what would work we wouldn't have to test).
How recently have you seen this?
I think we started seeing issues late last year with Gmail (as well as a shift in AOL/Yahoo around November or December) but Gmail especially has changed a lot this year, I’d say around March was clients noticing and we started to see a definitive shift in February. We have folks currently dealing with it and the focus on actives has worked and improving their reputation with Gmail and increasing the open rates. But, it’s very slow taking months to fix when this sort of work would improve things in a week previously.
I have also noticed that smaller senders seem to be getting the short end too but we’ve stuck to our guns of using whatever is “active” no matter how big or small it is. Just stick to the plan and its been working.
But, with large and small senders, I don’t see a pattern that one is impacted more than the other. The inactive/active seems to be the biggest indicator with those who keep their lists “clean” not having issues and those sending to whomever are.
Seeing very similar things. The tried and true methods just aren’t working like they did according to the GPT reputations at least. That said, we haven’t seen huge drop offs in engagement that we would expect with the reputation drops.
There’s nothing you’re missing. This problem is being seen across the board with Gmail. We’re seeing this on various mail streams – newsletter, marketing, etc. This could be the new norm. We need a coalition of experts , industry groups such as ESPC, MAAWG, etc to bring this “overfiltering” to their attention. Short of that, signup pages will simply out “gmail” as a “bad” provider.
I’ve seen lately an open rate going from under 5% to 47% on the same segment from one day to another and with a bad domain rep…
Another thing to take a look at is split testing your content and email signature. I’ve actually seen inboxing issues occur because of specific flagged content such as someone’s phone number or title in their signature.
One other strange thing we noticed this week is that, very inconsistently, Gmail started to throw the following warning when clicking a link:
“Suspicious link
This link leads to an untrusted site. Are you sure you want to proceed to [tracking subdomain]?”
The issue suddenly popped up this week, though I’ve heard from another ESP they started to see this earlier this month.
A whole range of checks (SSL, TLS, domain reputation, ip reputation, web reputation,..) did not bring up anything, so I’m hoping my message to Google’s support will shine some light on this issue.
Is no one interested in or talking about the content of the email itself? Machine Learning makes analysis of the email content trivial, particularly with natural language processing – it’s trivially easy for Tensorflow to take into account the semantics of your message and then make decisions based on it.
We’re working on an experiment to test this idea. Will let you know what results we get.
I don’t think its necessarily that talking about content is not valuable, its just one of the more mysterious aspects of Gmail’s spam filter so its hard to really speak with authority one way or the other. It really is all about split testing when trying to get a handle on potential content issues.
Gmail is changing all the time
This article shows that Gmail recently changes algorithm in sorting the emails around the tabs.
Quote: “Gmail adjusted their spam filters during September and October, according to numerous data sources reviewed by our Email Deliverability Services team.”
Good source
https://blogs.oracle.com/marketingcloud/email-deliverability-quarterly%3a-gmail-spam-filter-changes%2c-sparkpost-acquires-edatasource%2c-and-more