Hey Rasmus! I honestly wouldn't even bother running Inbox Vision tests. Here's what we typically say about them on the Deliverability Services team here at Braze:
This type of spam test by its nature is not terribly reliable. In the past, I've compared them to personality quizzes in Cosmo and Facebook. The tests do not indicate what the failure conditions are - because they don't actually know what they are. We are meant to assume, I think, that they are looking at content and formatting, though they explicitly avoid saying that, too.
Instead, they're using static seed accounts to see whether the mail appears or not. That's fine for what it is, but the method largely fails to account for recipient engagement, either by the individual recipients or by a preponderance of recipients at a given domain. Engagement has been the primary driver of inbox placement at the large free inbox providers (particularly Gmail) for a few years now. Seeds by their very design don't engage with anything. So those spam "test" data needs always to be taken well-salted.
A better approach is to do some multivariate content testing with small but statistically meaningful cohorts of random live recipients from the customers' lists. Strong open and click rates are a better, more reliable indicator of inbox placement
