[ricardobeat]

You are right about analytics, but that's an extremely easy problem to solve - simply disregard stats coming from the search engine bot. In fact both Chrome and Safari currently have a prefetch feature that I think is enabled by default, that's clearly not an issue.

For state, I don't see the connection to AMP here. RFC 7231 which defines the HTTP protocol specifically states that GET is idempotent, and should not cause any state changes in the server. The majority of servers conform to that, and if they don't, it's a bug with worse consequences than messing up prefetch.

[SquareWheel]

Unfortunately I think you're being a little too optimistic. The reality I've seen is that GET is used for all sorts of things, including state changes and irreversible consequences.

Analytics could check for useragent (and probably already does), but it just introduces a lot of noise.

Plus the other concern is that it'd be a huge privacy issue to have analytics being triggered just by performing related searches.

[ricardobeat]

So what’s your take on Chrome and Safari link preload which are in use by the millions at this exact moment?

[joshuamorton]

That it shouldn't be used on a search engine's results page.

Preload within a domain is reasonable. Preload across domains is, imo, a security concern, but maybe acceptable if the domains are trusted. Preloading untrusted, unknown content is a terrible idea.