I need to preface this by saying that I have always been a huge fan of Charlie Cook. I've used his ratings in my academic work; I've used his ratings to identify close races for classroom assignments. He's been very good at what he does.
What he does, however, is rapidly becoming obsolete. Just as with The Hotline, the explosion of information that's free or nearly free makes what he does a whole lot less valuable than it was twenty years ago. I think that's the context in which to read Cook's complaints about polling, which are criticized by John Sides here.
Years ago, Cook and Stu Rothenberg discovered a market for insider information on elections, and basically were able to capture that market between the two of them. They collected information that was hard to come by, assessed it, and packaged it for people who would pay good money to know whether the North Dakota or South Dakota Senate contest was a good investment. That was an excellent business in the days in which information about elections was scarce. All information -- back in 1988, just getting a full list of candidates wasn't all that easy, let alone polling numbers, campaign finance data, or information about party support. A PAC trying to decide which candidates to support really had its work cut out for them, and it made lots of sense to pay an expert to tell them what was happening. That's just not true any more: the stuff that you needed an expert for in 1990 could be done by an intern in 2010.
As I said, in 1990 Cook was paid to collect, assess, and package information. The first two of those have been undermined. Data is, of course, far more readily available. Moreover, anyone can look at all that data and form their own opinions, and their judgment is apt to be just as good as that of Cook and Rothenberg, who after all have no special training or methodology.
What Cook and Rothenberg do have, however, that Nate Silver or a Silver wannabe don't have, is insider information. In such situations, its only natural for those with insider information to place particular emphasis on it. Thus it's not surprising that Cook emphasizes the value of partisan polling (which may or may not be public) over non-partisan polling (which is always public). I expect that Cook would feel the same way about the value of polling overall compared to the value of any less quantifiable, less public information -- just as sportwriters have a bias in favor of valuing "clubhouse influence" over what actually happens on the field.
It's to Charlie Cook's credit that he's been able to survive as long as he has, despite the declining value of what he does. And fortunately for him and Stu Rothenberg, once people decide that something's an oracle and start paying for it, they're likely to need plenty of evidence that it's not before they give up, and it's not as if what Cook does has become (as far as I know) any less accurate over the years; indeed, almost certainly the opposite is true. As I said, I think they do a good job at what they do. It's just that what they do is less valuable every year, and there's not much they can do about it.
(Personal note: I mentioned 1988 because during the 1988 election cycle, I worked for a Senator up for re-election, and one of my jobs was packaging information about his campaign and sending it to Charlie Cook's shop. The goal was always to convince Cook, and by extension potential donors, that my boss was absolutely certain to win, and therefore any contributions to his challenger would be a waste of money. Not to mention that anyone seeking access to a Senator should be giving to my boss, who was certain to be around for the next Senate. So I can say that my personal experience with the process certainly confirms what John Sides says about the selective use of candidate-commissioned polls).