For example, years ago there was a controversy in which the North American Gay Amateur Athletic Alliance (NAGAAA) refused to acknowledge the bisexual identities of some of their baseball players. They interrogated five bisexuals to determine whether they were gay or straight, and in the end they determined that the two white men were gay, while the three men of color were determined to be straight. (Read about this on page 12 of this report.) This is obviously an inappropriate and racist way to use a gaydar.
However, I would suggest that there are some innocent uses of a gaydar. For instance, if I think several people in a group are GLB, that makes me feel safer in that group, even if I may be wrong about some of the individuals.
Whether gaydars are used for good or for ill, it's a separate question whether a functional gaydar exists. There are actually many studies showing that gaydars really do work. Here's one. Based on faces alone, and no other cues, people were 65% accurate in differentiating lesbian women from straight women, and 57% accurate in differentiating gay men from straight men. Actually, those numbers are misleading*, and I feel it's better to look at the hit rate and false positive rate.
(Table 1. The image was cropped and some information removed to better fit in my margins. "Upright" means the faces were upright; the study also considered upside-down faces, but I removed that data from the table.)
In the experiment, there are roughly the same number of photos of gay/lesbian people and straight people. In real life, maybe 3% of people are gay/lesbian. With false positive rates of 22% or 30%, most "pings" on the gaydar will be false positives.**
And yet it seems like gaydars are more accurate than that. Nobody ever says, "I have a really good gaydar. My hit rate is 10% higher than my false positive rate!" Why is it that the gaydar seems so much more accurate in real life than in the studies?
I propose (without any real evidence) that in real life, the gaydar mostly works by confirmation bias, and by reading social cues. In other words, the gaydar is like cold reading. Cold reading is the method used by psychics to make seemingly-accurate predictions about a person. Cold reading also works by reading social cues and by confirmation bias.
Confirmation bias is about remembering the hits and ignoring the misses. Psychic readers tend to make lots of deniable predictions. When they get it right, it's amazing. When they get it wrong, they act like it wasn't really a concrete prediction after all. Gaydar also appears to work the same. We tend to remember the times our gaydar was accurate, and forget the times it was inaccurate.
Cold reading is also about reading social cues, such as the emotional reaction to the predictions. The gaydar works similarly. Because I'm familiar with gay culture, I can spot people who have been influenced by that culture. However, I'm not as familiar with lesbian culture, so it's harder to apply to women. And it doesn't work on closeted people, or on people of different cultures, or on straight people influenced by gay culture. It's good to know the limitations of one's gaydar lest it be abused.
In the scientific studies, all social cues are removed, and confirmation bias is removed. It's interesting that there still exists a "gaydar" under these conditions, but it seems to me that this isn't the gaydar that people talk about all the time. In real life, people are just talking about cold reading.
*The "accuracy" percentages reported are actually a statistical parameter called A'. Here is a paper that defines A'. A' does not really match the intuitive definition of accuracy in my opinion.
**Presumably in real life people would adjust the sensitivity of their gaydar downwards in order to get a smaller false positive rate, and also a smaller hit rate. Basically this means moving to a different point on the ROC curve. Signal detection theory is complicated.