On paper, at least, nearly everyone is against online harassment. But how that line is defined — and how consistently the rules are applied — can vary dramatically on platforms and social media. Does it require a direct or immediate threat of violence, or is it enough to say you “hope” someone dies? Does it count if a popular user targets someone for dogpiling by their followers, even if their own language isn’t abusive? Do the rules apply to the president of the United States when he celebrates violence against the media, or taunts volatile dictators with nuclear war on Twitter? (Update: no.)

According to a new Pew Research Center survey, defining online harassment is just as complicated for the average American user as it is for huge social media companies — and the line gets even more fuzzy when gender or race come into the picture. The survey polled 4,151 respondents on various scenarios and asked them whether each one crossed the threshold for online harassment. In one hypothetical, a private disagreement between a man and his friend David is forwarded to a third party and posted online, which escalates to David receiving “unkind” messages, “vulgar” messages, and eventually being doxxed and threatened.

When asked whether or not David was harassed, 89 percent of respondents agreed that he was. However, opinions on exactly when the harassment began varied widely: 5 percent considered it harassment when David offends his friend; 48 percent said it’s when the friend forwards the conversation; 54 percent said it’s when the conversation is shared publicly. Others agreed it crossed the line when David received the unkind messages (72 percent), the vulgar messages (82 percent), is doxxed (85 percent), and threatened (85 percent). There was little difference in responses by gender.

Questions regarding sexual harassment, perhaps unsurprisingly, are more divisive — especially between men and women. In a second example, a woman named Julie receives “vulgar messages” about her looks and sexual behavior after posting on social media about a controversial issue. Women were about three times more likely than men (24 percent vs. 9 percent) to label it online harassment when Julie’s post is shared by a popular blogger with thousands of followers. Fifty percent of women vs. 35 percent of men consider it harassment when Julie starts getting unkind messages. When it comes to vulgar messages, threats, or Julie’s photo being edited to include sexual imagery, 8 out of 10 men consider it harassment, as opposed to 9 out of 10 women.

That trend — of men being less likely to believe something is sexual harassment than women — echoes what women (as well as LGBT users and people of color) have been saying for years: that while they are more likely to be harassed, they are less likely to be believed.

There’s also a curious division between acknowledging something as harassment and believing that action should be taken by social media platforms. In the case of sexual harassment, for example, 43 percent of respondents considered the unkind messages harassment — yet only 20 percent thought the social media platform should intervene. In a scenario where a woman’s picture is edited to include sexual imagery, 84 percent called it harassment, but only 71 percent thought platforms should step in. The same can be said of an example involving racial harassment. Although 82 percent of respondents called messages with racial slurs and insults harassment, only 57 percent thought the platform should step in; the same goes for the person having their picture edited to include racially insensitive images (80 percent vs. 57 percent) and threats (82 percent vs. 67 percent). In both cases, respondents’ gender is not provided.

Pew’s survey points at a disheartening roadblock in addressing the issue of online harassment: that people often have trouble agreeing on what qualifies as harassment in the first place, especially when women or minorities are involved. It also paints a troubling picture where even when people do define behavior as harassment, many still hesitate to hold the offenders accountable for it. If we truly want to change the culture around online harassment, we need to be able to recognize it when it happens to people who don’t look like us, yes — but we also need to have the courage to take a real stand against it when we do.