Skip to content
Don't Feed the TrollsGet help

Counter-speech

Replying in public so that onlookers see a claim answered, rather than to persuade the person who made it — a narrower and rarer case than most advice about replying implies.

Stop reading this page if any of these is true

  • Someone is threatening you, following you, or contacting you against your wishes.Go here instead →
  • You're being targeted by a coordinated campaign rather than one person.Go here instead →
  • You are frightened of what replying might provoke, or unsafe in some other way.Go here instead →
  • You are the one being targeted and want to answer for your own sake, not for an audience.Go here instead →

None of what follows is written for those situations, and some of it can make them worse.

Works when
There is a real audience beyond the person you would be replying to, and something is being asserted in public that will otherwise stand unanswered.
Fails when
You are the one being targeted rather than a bystander speaking up, replying risks extending the exchange and handing the other person the engagement they were after, or there is no audience beyond the two of you for a reply to reach.

Don't feed the trolls ends with six questions about whether disengaging is the right call, and two of them lead nowhere from that page: are you the audience, or is someone else, and is anything being asserted that will stand if unanswered. This page is the answer to those two, and it starts by separating two things most advice about replying treats as one. Replying to change your harasser's mind and replying so that the people watching see a claim challenged are different acts. They are aimed at different audiences, and they succeed or fail for different reasons. Counter-speech, on this site, means the second one only.

What was actually tested

There is exactly one study behind this page, and it is worth being upfront about how little that is before going anywhere near what it found. The study is Kevin Munger's 2017 paper, "Tweetment Effects on the Tweeted." Only its abstract is archived here — the full text sits behind a paywall — so nothing beyond the abstract can be claimed for it. What it measured was narrow: the use of anti-black racist slurs by white men on Twitter. [1]

The sanctioning messages came from accounts the researcher controlled, not from anyone the subjects had actually harassed. [2] Those accounts varied along two lines: the identity they appeared to have, and how many followers they appeared to have. [3] One combination produced a measured effect. Subjects sanctioned by an account that looked like a high-follower white man significantly reduced their use of the slur. [4]

Read that again before you use it

That result describes strangers running an experiment. It does not describe you answering the person who came after you. The accounts doing the sanctioning were built and operated by the researcher for the purposes of the trial — nobody in the study was defending themselves. If you are deciding whether to answer someone who is targeting you, you are not in the situation this study tested, and nothing here is evidence that replying to your own harasser works. That is the single most important thing on this page, and everything below assumes you have read it.

It is also worth being precise about what the abstract does not say. It reports one condition producing a significant reduction. It does not report the other conditions as having done nothing — an abstract that states one result is not the same as one that rules the rest out, and treating "only the in-group, high-follower message worked" as a finding states an absence the source never states. What can honestly be said is narrower: this condition, run by this researcher, on this platform, in work published in 2017, produced a measured drop in the use of one slur.

How thin this is

One study, one platform, one slur, one demographic, published in 2017 — that is the entire evidence base, and it has not been the subject of a wider search for replication here. A single field experiment on one behaviour is a thin foundation for a general rule about when to answer a stranger. The honesty this site tries to bring to grey rocking and to declining to reply applies here in the same measure: a plausible mechanism and one significant result are worth something, and neither is the same thing as tested, general advice.

Who the study is closest to

The condition that produced an effect was a sanctioning account with a large following — one an onlooker would read as speaking with some standing behind it. That description fits an organisation, or an account with a genuine audience of its own, far better than it fits a person replying from a small personal profile. It is why this page treats institutions and accounts with reach as the closer match, while still being clear that a closer match is not the same thing as tested support for whatever your account actually looks like.

What it costs

Replying carries the ordinary costs of replying, and none of them were measured here. It can extend an exchange that would otherwise have ended. It can hand the other person exactly the reaction they wanted, and expose you to more of what you were trying to answer rather than less. None of that shows up in an abstract about slur frequency, and it is worth weighing before you draft anything.

And if you are the one being targeted, rather than someone deciding whether to speak up on someone else's behalf, the study offers you nothing at all — its entire measured effect came from a third party standing outside the exchange, not from the person being harassed answering for themselves.

Where that leaves you

Most of the time, for most people, this site still thinks the better bet is disengaging or going flat rather than replying. This page exists because two of the questions that advice ends on had nowhere to send a reader who answered them a particular way: a real audience watching, and a claim that will otherwise stand unanswered. Those are the specific circumstances where speaking up has anything behind it at all, and even then, what is behind it is one study on one platform, not a guarantee. If you do reply and want a record of what was said regardless of how it goes, keeping a contemporaneous account is a separate, and separately sourced, practice.

Sources

Last checked . Each quote below is verified against an archived copy of its source every time the site is built.

  1. [1] Kevin Munger. Tweetment Effects on the Tweeted: Experimentally Reducing Racist Harassment, Political Behavior (2017) [archived]
    i employ an intervention designed to reduce the use of anti-black racist slurs by white men on twitter
  2. [2] Kevin Munger. Tweetment Effects on the Tweeted: Experimentally Reducing Racist Harassment, Political Behavior (2017) [archived]
    i collect a sample of twitter users who have harassed other users and use accounts i control
  3. [3] Kevin Munger. Tweetment Effects on the Tweeted: Experimentally Reducing Racist Harassment, Political Behavior (2017) [archived]
    by varying the identity of the bots between in-group (white man) and out-group (black man)
    ; Kevin Munger. Tweetment Effects on the Tweeted: Experimentally Reducing Racist Harassment, Political Behavior (2017) [archived]
    by varying the identity of the bots between in-group (white man) and out-group (black man) and by varying the number of twitter followers each bot has
  4. [4] Kevin Munger. Tweetment Effects on the Tweeted: Experimentally Reducing Racist Harassment, Political Behavior (2017) [archived]
    subjects who were sanctioned by a high-follower white male significantly reduced their use of a racist slur