It’s a bit more complex than that. If you read the more detailed post we link to from the video, you’ll see that top Google engineers have emphatically, clearly, and repeatedly stated that they do not use CTR as a direct ranking factor. They do, however, use it in limited test of search results quality. Then they take what they learned from those tests and refine the search algorithms. So over time, it is natural that on the whole, sites that draw higher CTR will rank higher, but it is not because a direct ranking factor is being applied to those sites. Does that make sense?

Thanks for the comment, John. You’ve pretty much listed the reasons that most of us used to believe that CTR “must” affect rankings. We have to remember, though, that search is always more complicated than our view of it. In Eric’s and my discussions with Google search people over the years (both public and private), we’ve been struck by how consistently they tell us two things:
1) What they see from their side is always more complex (and often less clear) than what we see on our side.
2) They have to ignore a lot of signals that might seem “useful” to us, either because they are messier than they appear or they just don’t have the resources (yes, even Google doesn’t have unlimited resources!)
As far as their ability to sniff out manipulated clicks, Gary has told us they actually do find a lot of that, but even with that the data from CTR is still too difficult to interpret to be useful. It’s one thing to sort out false clicks. It’s quite another to successfully discern why users click (or don’t click) a result.
Take the case of pogosticking, which we show in our video. Most SEOs have assumed that pogosticking (a user clicking one result, then in a short time coming back to the SERP and clicking another) is clearly indicating the first result clicked is unsatisfying. But there could be any number of scenarios in which it’s not (such as a user doing quick price comparisons, for one). So even this isn’t as clear a signal as you might assume.
Add to that the testing done by Rand Fishkin, in which I’ve participated. We did numerous tests in which a large number of users were asked to click a certain lower result for a query in a short period of time. Sometimes, especially in the early test, the ranking did jump up quickly. But even in those cases, it only worked for very localized results, and the jump fell off over a few days. In later tests, even those carefully hidden from Google, it became harder to produce even those results.
All that is not to say that Google doesn’t use click behavior at all. In addition to the limited tests mentioned by Paul Haahr, other Googlers have mentioned that it is sometimes used in personal results (i.e., a site you have clicked on in the past may rank higher for your personalized, logged in results). But these are limited cases where they feel they can be relatively sure of the meaning of the signal.