Population of the US: 349 M, of which 250-300 M use Google services, multiplied by 1605 USD per user = from 401 B USD to 481 B USD, but in 2025 Alphabet did 403 B in total, from every service, in the whole world.
https://support.google.com/google-ads/answer/6366577?hl=en
This is the process for determining which ads get run. The bid is only one of many factors, so as their support document indicates, the price you pay is often quite lower than the bid, which reflects a ceiling rather than a real sale price.
https://support.google.com/google-ads/answer/2580383?sjid=17...
This is their guidance on demographic targeting. Note there is no category allowing you intentionally target children. This doesn't mean advertisers can't figure out some way to do it anyway, but it means Proton can only sample from adults. Presumably, some probably very large number of the people who "use Google services" in your estimate are children, which childstats.gov indicates represent about 22% of all Americans. That makes it more like 195-235M adult users of Google services.
https://support.google.com/google-ads/answer/2464960
As indicated here, you don't pay to place an ad. You pay for clicks, so regardless of what you bid and who you target, Google isn't getting revenue for the number of placements you bid on, which is what Proton is sampling here. Presumably of the 250 x 0.78 to 300 x 0.78 million adult users advertisers are placing those $1605 average bids on, quite a bit fewer than 100% actually click on at least one ad.
similar anti-fingerprinting tech can kill ad revenue as it makes users non distinguishable from bots (but likely doesn't matter here)
median is the sample in the middle of the distribution is treated as a sequence of samples ordered by their value, e.g. if you have sort(seq(dist))=[100$, 5$, 5$, 3$, 1$] the median is 5$
average is sum(dist)/count(dist), so avg * size(dist) => sum(dist) (in the example above that would be avg. 22.8, total 114, size 5
if you where to multiple the median by count you had 25$ which is very much very wrong
proton did 54,000 samples of US users and made an average of what advertisers are willing to pay to target, not what they actually did across the whole population
and plus this isn’t to inform you, it’s to sell you on another proton honeypot
- avg. $1_605
- but mean is $760, i.e. half the users generate $760 or less
I also wouldn't be surprised if the sampling distribution has two maxima even if smoothed (on around the mean and another at the lower end). Would be nice to have that plotted out properly.
Median*. Mean and median are both measures of averages, though colloquially average is taken to exclusively mean the mean.