From the post:<p><i>> In the end, though, Google is heavily invested in users landing on their result pages so they can be served ads. So it's hard to imagine they would voluntarily redirect people away, even if they are clearly searching for a specific page.</i><p>The sad part of this is that Google, in the "10 blue links" era, actually <i>did</i> optimize for users clicking through to a result (and not immediately coming back to click another, which indicates the page wasn't satisfactory). This extended well into the early AdWords period where optimization extended to showing highly relevant ads in the sidebar or at the top of the results.<p>Things seem to have started to go wrong first with a series of design changes that made it harder to distinguish between ads and search results, and accelerated with the introduction of info boxes sometimes scraped from external sources and integrating results from other Google services (such as maps).<p>Since those changes, Google seems to have been optimizing for two use cases: either the user does a search and then stops (indicating they received the answer they were looking for) or they click through to a Google property or an ad and don't come back to the search engine. Clicking through to a search result is a lower priority.<p>The problem, as I see it, is that the first case of searching and then stopping also indicates "these results are garbage, why even bother" which dilutes the signal. You can potentially distinguish the two cases of 'completely satisfied' vs. 'completely unsatisfied' with a bunch of other weaker signals (clicking through to the long tail of SERPs, trying a bunch of search variations one after the other), but most users today won't try further no matter what (Google, ironically, has all along been training users that the first result page of the first query they try satisfices).<p>In short, Google has been shooting themselves in the foot by muddying the strongest signal it had for search quality.<p>I imagine that Google's solution to this involves behavioral tracking/profiling and analysis as in "is the user in the 'easily frustrated' bucket" to amplify or discard the weaker signals appropriately, but clearly whatever they're doing isn't working as well as the unambiguous signal did.