The searching and using of links is a general purpose approach. We have classifiers that allow our models to determine if for example data that it needs or the question being asked references stuff that would need the latest and greatest information. But for things like pointing people to canonical resources, those are things that are more hard coded. And we have a repository of those resources. So canIvote.org as an example. Or in a different domain, if there are people asking about self-harm instructions, potentially pointing them to suicide hotlines and websites. That’s another example. So it’s a just a more specialized kind of classifier and then pointing to specific resources that we know exist and are the best ones to point people to.