Computing statistical significance for qualitative data?

#1
In internet marketing there's something called pay per click marketing. You only pay if a user clicks on your ad. You can also test different ads to see which ad performs better.

People seem to believe that if they gather 100 or so clicks they'll be able to determine which one has the better conversion rate (conversion rate = number of sales / number of clicks).

I assume a sample size of 100 is pretty low and unless the conversion rates are worlds apart one would have to gather more data to get smaller confidence intervals?

I just confused myself:

1)if Im testing statistical significance do I have to use confidence intervals or could I do it without them?

2)the sample size for this IS 100 if 100 people click, right? NOT the number of people that saw the ad all in all or something?

Anyways: I couldn't compute a variance or a standard deviation..or could I? It's only "made a sale" and "did not make a sale".

How will I determine when the data sample is big enough so I have a say 95%significance that the conversion rate of one ad is better than that of another ad?