You are confusing two different DNA scenarios. Matching with descendants of common ancestors is accurate and indisputable above match lengths of about 16-20 cM. With match lengths less than that, the chances that the purported match is false increase significantly the lower the match length, so that at 6 cM (which is the lowest match length that most DNA testing companies will report) there may be around a 50% likelihood that the proposed match isn't a genuine match at all.
The reasons for that are complex and varied, but because match lengths to ancestors decrease with successive generations, and therefore the generational distance between current relatives, it is generally accepted that a genuine match can only be determined over 5 to 8 generations at an absolute maximum.
But ethnicity estimates are a different (pseudo) science. They do not attempt to match you with relatives who have descended from a common ancestor within the last 5 - 8 generations. Instead, the theory is that segments of your DNA are compared to segments which the testing companies believe may be commonly found in groups of people (whether more recently related or not) who originated from particular regions within the last several hundred to a few thousand years ago.
As previous replies have said, the definitions of particular regions differ between different testing companies. The reference populations that are used to determine those regions are incredibly small in comparison to the actual populations of those regions - often only a few hundred people and in some cases fewer than ten. Ethnicity estimations to continental level are considered to be more reliable than those to individual country level or even regions within countries, and again, lower percentages allocated to particular regions are likely to be more suspect, so a supposition that you have 45% English ethnicity is quite likely to have some bearing of truth, whereas an indication that you inherited 2% Swedish ethnicity may well be totally incorrect.
So the assertion by one testing company that my ethnicity (as determined from my transferred to them Ancestry DNA test) is 98% European is, I think, quite likely to be reasonably accurate, but is it at all useful? Most people could probably give me an estimate with a similar degree of accuracy just by looking at me
Whether you choose to believe and place any reliance on your ethnicity estimates is your choice. There may be a degree of expectation bias involved in people who claim that they are accurate. Personally, as already said, I choose to treat them as an informative but not necessarily accurate ploy to sell DNA tests to people who would otherwise have no interest in testing and the hard work involved in researching and identifying relatives descended from common ancestors.
I prefer to concentrate on using DNA matches with other people to further (and confirm or disprove) my research into my own ancestral lineage.
