It doesn't measure whether the tweet is positive v negative, which surely has a large impact.
I think the 'reach' figure is reaching as well, because I am guessing they just counted the number of followers the person who tweeted has, but of course, there is no way to know that every follower read that tweet.
Funny you should mention it, I actually made an Android app that does basic sentiment analysis on tweets given a search term. Not very sophisticated, but it uses large corpuses assembled by text mining researchers.
I was under the impression tweets are near impossible to get right to a usable degree because of how short they are on top of the usual problems of sentiment analysis.
Its just measuring the buzz, not if it's negative or positive. In any case people talking about your ad is a good thing. Even if they say bad things, it increases your brand recall.
Reach is a reasonable proxy for comparison, if you assume that ratio of people who won't(or would) see the ad, to followers is pretty much the same across ads.
Negative buzz was unlikely the goal of any of these ads. Depending on the brand it may matter more or less. Godaddy for instance has all the negative buzz they need.
And once again the take-away message from this: Sex and fuzzy animals sell (see the H&M/David Beckham in his underwear and Coca-Cola Polar Bears and the tops spots).
I would say it's not very accurate on a per-tweet basis, but when you analyze a bunch and you look at the breakdown of positive:neutral:negative and the different keywords, it can be pretty telling.
The best results obviously come with things that people feel strongly about, because they are more likely to use strong words associated with their sentiment.