No one really has a handle on this, especially in the US. Much is made of the lack of a standard definition of “recovery,” and for good reason.

The data are so spotty, public-health authorities say they don’t know what the true count [of recoveries from a Wuhan Virus infection] is. …
The spottiness stems from the absence of both an agreed-upon definition for a coronavirus recovery and a standardized way to track the numbers of patients, the health experts say. What constitutes recovery is so nebulous that some states don’t even track it, and those that do probably undercount the true number.

However, even were there a standard definition, we still wouldn’t have any idea of the true statistic, nor any idea of the accuracy of an estimate for a recovery statistic.

This illustrates the key:

Such measurements [numbers of recovered] might indicate how many people who tested positive for the coronavirus didn’t die, but might miss those who never displayed symptoms and didn’t undergo testing.

The only recoveries, under any definition, being counted are those from confirmed—test-identified—cases. There is no estimate of the number who have been infected, never tested, and recovered.

Those recoveries are not included in guesses of the number of recoveries.