[jwr]

> Google is pretty accurate in figuring out transactional versus marketing. They don't tell their heuristics, but you don't think engineers who build web crawlers cannot build email classifiers?

Yes, I definitely think that. The engineers can build anything, but where the company focuses matters.

I've seen transactional E-mails get sorted into people's spam/junk/newsletter folders too many times.

[ndriscoll]

I also get tons of spam to my inbox despite regularly marking it as such, so if they are classifying marketing emails, they're not doing anything with that information.

How hard is it to classify a message that literally contains the string "this is an advertisement"?

[ehutch79]

Turn off your spam filtering for a week.

You'll find out what the quality of the job they are doing is.

[jodrellblank]

Your comment literally contains the string "this is an advertisement" but I can't tell what you are advertising?

[ndriscoll]

It's also not an email. I've never seen a legitimate email with that string, and all of the illegitimate ones should be triggering other heuristics, such as the existence of an unsubscribe link, or things like "$x off".

In any case the false positive rate on that would likely be incredibly low, so it's a good heuristic considering how bad the false negative rate is right now.

[Macha]

Dear site admin,

This is as advertisement that appeared on your site yesterday that is a phishing scam pretending to be a bank.

<Screenshot>

Please prevent ads like this showing up on your site.

Regards,

Client XYZ

---

Maybe it's just the positions I've been in, but I've often seen variations of the above email, and I've never seen advertisement emails that flat out say "this is an advertisement"

In fact, what I have seen are advertisement emails of the form

"This is not an advertisement, we'd like to arrange a call to discuss ways to grow your business. Signed, Bob the XYZ product sales manager"

[ndriscoll]

Like others, I get people handing my email out all the time by mistake because I grabbed first-initial-last-name 20 years ago, so I get lots of corporate spam that others have signed me up for. If you look at corporate spam, it frequently contains a passage like this:

> This is an advertisement and outbound email only. Please do not respond to it. This email has been sent on behalf of Kia Motors America, Inc. (KMA). To opt-out of receiving marketing/promotion emails from or on behalf of KMA, please click here.

Or this one I got last week from J Crew:

> We want you to hear about what's just right for you. Update your email preferences here. This email may be considered an advertising or promotional message.

For a while I just ignored it, and this kind of thing never went to spam. Now I always mark it as spam, and it's starting to, but their default spam heuristics are apparently awful, and it seems like marking as spam just affects that one sender, so you have to do it all the time for new spammers. I still just got linkedin spam yesterday after I have marked thousands of their messages. It can't be that hard to come up with heuristics for this. The biggest signal is probably that it contains an unsubscribe link since it has to be there by law.

Your example is also a single message. I imagine they look at patterns, and a single sender sending thousands of emails which are 99% similar is probably also a strong signal that it is spam (yes there are transactional emails that are templated; that's why it's a signal). That combined with the "this is an advertisement" heuristic is probably pretty accurate.

The reality is--obviously--that they are not trying to stop corporate spam. They're an advertising company; they don't want to normalize the idea that advertisements are supposed to be filtered.