It often depends on the presence of multiple hyper-links in a post by a relatively new user, or the usage of certain keywords. Bayesian filters (which the mod-attention filter almost definitely is) can only check for certain triggers, they can't distinguish between spam content and legitimate content.