Something that might make WolframAlpha look substantially more intelligent would be what the Eliza psychotherapist program did half a century ago, it used your own intelligence against you. When it was a little at a loss for words it would pick a couple of words you had just used and ask you something related to them.
Example: I just asked WolframAlpha for inflation adjusted federal debt last ten years. It fumbled around a bit and didn't seem to really understand what I was asking for, it knew federal and debt and maybe even last ten years, but nothing more. I waited and watched and waited. And it popped up one of the typical helpful windows asking me if I wanted to know what year George Orwell's 1984 was published. That really really communicates the impression of just how deeply retarded WolframAlpha is, probably even more than it really is. And these totally unrelated pop up questions successfully communicate that after every single query, especially the more times you try to reword a query, again don't get an answer, become even more frustrated and get another senseless pop up. But maybe that is what thirteen year olds want.
If it had, on the other hand, when it was estimating that it probably didn't really answer the question you asked, it popped up one or two example queries that directly related to the terms it didn't understand and perhaps connected to one or more of the terms it did understand, even when those were not answering the exact question you just asked, and showed you the notation needed to communicate that concept to it then you might have been far more impressed that there was some intelligence lurking in there.
It also might be nice if someone could hard code deep into WolframAlpha somewhere that the code "i.a." means inflation adjusted and this is to be given a high priority in any question involving money. Somehow, perhaps using one of those pop ups, communicating that code to users asking questions about money would make it look far more intelligent.