Let's say I have a Teakettle test which assigns a score to the temperature of the water in a teakettle based on only qualitative properties of a teakettle (is it whistling, is it rumbling, does touching it cause pain?). This test is probably good at assessing the extreme cases (water is 212 F, water is around 70F) but I'd guess it's a poor predictor of temperature. Now if you gave me a thermometer I could see how correct my test was.
The problem is we don't have a thermometer for intelligence (that I know of) and more important I don't know if there is any consensus that 'intelligence' can even be assigned a quantitative value the same way the vibration of molecules can.
Now maybe the IQ test, like the Teakettle test, is a good summation of qualitative properties. Now you can compare this quantitative value to others to show a useful correlation. The Teakettle test may be a really great way to predict time to tea-time. An IQ score can only be useful if it shows a strong correlation to other, quantitative things (SAT score, salary, life expectancy, number of hairs in eyebrows)
To say that a real IQ test is a decent measure of intelligence seems to be at best a tautology.
Intelligence is not a property like physical attributes. It is fuzzy and has a lot of aspects. Some forms of intelligence aren't tested with the IQ test (unless somehow done recently), the primary one that comes to mind is social nous.
The psychology of intelligence is a complex, varied, and interesting field, and those who naysay "you can't study / define / measure intelligence!" are submitting to pop science just like "take this pop IQ quiz!". I've lost track of it over the years, but there are reams of research into intelligence, what it is, how it's defined, how tests do or don't reflect intelligence.
The IQ test covers a lot of different aspects of intelligence - it gives a fairly comprehensive representation of many aspects, and hence why it's a decent measure of intelligence. It's not a cut-and-dried thermometer measure, yes - but this is because psychology never works that way, something that engineers have trouble getting their heads around. Nothing in psychology is concrete, and if you're going to dismiss based on not being 100% precise, well, then psychology can't exist, nor can any of the fields based from it (like sociology, criminology, so on and so forth)
Don't the shortened IQ tests correlate pretty strongly with the longer ones? When I read psychology papers, even ones dealing with children, they pretty frequently use speeded or abbreviated ones without apparent qualms.
I did a partial IQ test when I was studying intelligence. I was on track for getting around 140 according to the conductor of the test. But I can't say my IQ is 140 because I may have fared worse (or better!) on other parts of the test (for example, we didn't even touch on the verbal tests).
Everything else that calls itself an IQ test is really a form of pop science quiz and of course has flaws. Any test that you're supposed to administer yourself is not the IQ test - that's the easiest first step.