When I was looked I was unable to find the SNPs. This might be because I did it as soon as 23andMe reduced prices to $200 for the first time, before they'd cleaned up their website.
Turning "probability of alzheimers disease" into a single percentage is a meaningless project. Alzheimers has an age likelihood - so at age 50, some portion of the population will have the pre-clinical precursor, some portion will have mild symptoms, some moderate, some severe. By age 60, and 70, and 80 and 90, the proportions change again.
So turning it into a single percentage number, as 23andMe did, is unable to capture what should be captured. It reeks of a company who did just enough research to put up a percentage number (because that's how their UI worked and that's what their UX designers thought users wanted to see), but didn't have the integrity to think through whether it was meaningful.
Indeed, if you wanted a single numerical probability for whether you'll show clinical signs of Alzheimers before you die, you'd likely get a more accurate number by estimating when you'll die from other factors like cancer or heart disease -- will it be before the "numbers game" of alzheimers and old age catches up with you, or not?