Our analysis of Nightly permission prompt interaction data suggests that only a fraction of users reconsiders their decision to deny a notification permission prompt when asked a second time, to quote
Figure 1 shows that the first time a site prompts a user (during the study period), the user is more likely than not going to grant the site permission: the blue bar at x=0 shows a 52% accept rate. However, if the user does not grant permission and the site prompts the user again, then on each subsequent prompt, the user is less likely to grant permission. After only one denied or ignored prompt, the next prompt only has a 14% accept rate. After ten or more occasions, the accept rate dropped to less than 1% - yet there was a heavy tail of such prompts, as can be seen from the blue line in Figure 1.
So, given that a site has previously prompted a user, it is unlikely that any given future prompt is going to be accepted, and the likelihood drops as the number of previously not-accepted prompts grows. From this perspective, we probably should simply refuse to prompt the user more than once per site.
I have attached said figure to this bug.
I agree with the document that we should refuse to prompt more than once and we should start shipping this and evaluate if it improves our Telemetry numbers for this prompt