When considering the entire mitogenome (HVR1, HVR2, and the Coding Region), Family Tree DNA places the cutoff for "matching" at a combined genetic distance of 3. So if there are more than 3 differences found (excluding a couple of very common mutations) FTDNA will not list people as being possibly genealogically related.
But even that is considered by many researchers to be way too optimistic. An important paper by Mikkel Andersen and David Balding in 2018 (https://doi.org/10.1371/journal.pgen.1007774) determined that mutation rates for the entire mitogenome ranged up to 1 change per 70 generations. More detailed information in the same paper used the Översti method for a non-endogamous population and arrived at an approximate 87% chance of an exact full-sequence mtDNA match every 190 birth events.
If we assume the average female generational interval to be 22 years, even the 1 in 70 generations number places us back to around AD 480, so before the common genealogical timeframe. My personal opinion is that FTDNA is allowing an over-generous report of "matching" at genetic distances of 3. There are both marketing and scientific reasons they would do that, but I certainly agree that any separation of two tests by more than 3 differences indicates no common ancestor in a genealogical timeframe. At that point the common ancestor may already be over 4,000 years distant.