The algorithm cleans up the inputted text and then attempts to analyze the quality of the vocabulary used. It then maps that information to an IQ value, using a custom made function that was designed to fit data of texts and the IQs of the authors behind them.
As accurate as one would expect such an algorithm to be. As "proof" I have run the algorithm on a few examples, and you can see that the results aren't surprising (nor should they be).
The algorithm assumes that when writing the text, the author put thought behind every sentence he wrote. Furthermore, it assumes that the author didn't purposefully attempt to dumb himself down and is sufficiently fluent in English as to not be restricted in his choice of words. Ideally, you should paste in a paper, article or another document that you have written and are proud of. Note that the algorithm is meant to be equivalent to an IQ test for adults, so if you are under 18 the result will likely be significantly lower than your true IQ score.
If you don't possess a text fitting the requirements, you can attempt to answer the following prompt to create one: describe one way in which the world will be different 30 years in the future and explain your reasoning to the best of your ability.
Note that you can use a spell checker.