What clip farming is, and why you cannot prove it

In May 2025, Mouv's Culture Internet gave four minutes to a word that had turned into an accusation across French streaming: clip farming. The definition Radio France offers is narrow, and worth keeping as written: deliberately creating moments that are absurd or overplayed, built to go viral on social platforms.

The chronicle is only half about the practice. The other half is about what the accusation did to the audience: a public that began auditing streamers for sincerity, and a term that became a way of calling someone a fraud.

The sourceLa tendance clip farming : ou quand les spectateurs confondent sincérité et mise en scèneCulture Internet, by Tahzio · Mouv' (Radio France), 15 May 2025 · 4 min, in FrenchListen on radiofrance.fr

Why the practice exists at all

The chronicle opens with the reason, and it is not a moral one. TikTok and algorithmic feeds have become, in a handful of years, one of the most effective ways to bring new viewers to a Twitch channel. Short, simple, addictive content is exactly what that recommendation system rewards.

Read as a supply chain, the arrangement is unambiguous. The long broadcast is where the community and the revenue are. The short clip is where new people come from. Which means every minute of the long thing can be scored on whether it produces the short thing, and once a behaviour can be scored, it gets optimised.

Nobody had to plan this. The buy side already prices it: agencies pay clippers by the thousand views, which is why clips exist in volume in the first place (we went through those rate cards in how the paid clipping economy works). The sell side is the mirror image. If being clipped is free distribution, then being clippable is a property worth cultivating, and the person on camera is the one best placed to cultivate it.

The clip police

What Radio France documents is the backlash, and it escalated faster than the practice did.

For part of the Twitch audience, streaming was supposed to be spontaneous, natural, authentic; clip farming reads as a cynical way of maximising numbers at the expense of that original sincerity, and therefore as a betrayal. In France the criticism crystallised on one streamer, AmineMaTue, whose reactions were labelled clip farming and who drew an unusually aggressive campaign as a result. It began, the chronicle notes, as light mockery, and degenerated quickly.

The consequence is the part worth quoting. From seeing clip farming everywhere, including where there is none, "a sort of clip police has set itself up", and "the slightest slightly original or spontaneous moment on a live stream can now be suspected of manipulation." The risk Tahzio names is a climate in which any streamer is suspect of not being entirely sincere.

That is a strong claim about an audience. It also happens to be the predictable result of the format, and the reason has nothing to do with anyone's character.

The accusation cannot be settled from a clip

To prove clip farming you would have to prove intent: that the person did the thing because it would clip. The only evidence in circulation is the clip.

And a clip is precisely the artefact with the surrounding minutes removed. It is thirty seconds selected for standing alone, cut so that nothing before or after is needed to make sense of it. Whatever would tell you how the moment arose (what was being discussed, what someone had just said, whether the reaction was proportionate to something you cannot see) is the material the edit exists to remove. The clip is not weak evidence of intent. It is the removal of the evidence.

Worse, selection is a filter that makes both cases look the same. Of the thousands of moments in a week of broadcasting, the ones that get cut are the loud, sudden, self-contained ones. That is the whole selection criterion, and we spent a whole piece on it in what makes a moment clippable. A genuinely spontaneous outburst that survives that filter looks engineered, because looking engineered is what the filter selects for. A manufactured one looks identical. The clip cannot separate them, because it was produced by the step that made them indistinguishable.

So the accusation is unfalsifiable in the ordinary sense: no clip can confirm it and no clip can clear it. What circulates instead is a vibe, applied to whoever is currently unpopular. That is a bad test, and a bad test used at scale is how you get the climate the chronicle describes.

What "authentic" was ever measuring

Tahzio's conclusion is the uncomfortable one: the current ecosystem imposes this strategy on streamers, succeeding online means catching the eye first, and the audience itself now consumes mostly short clips. Clip farming, deliberate or not, answers a demand. Twitch remains entertainment; everyone stages themselves in front of a webcam; "demanding perfect authenticity is an illusion."

We would put it slightly differently. A live broadcast is not the absence of performance. It is a performance with the edit removed. Six hours of it is still someone doing a bit, for an audience, in a room they lit. The thing viewers actually valued was never the absence of intent. It was that the broadcast was not only aimed at the clip: that the hours between the highlights were doing something for the people watching them live.

Which suggests a better question than was this staged. Ask whether the clip represents the broadcast it came from. A stream that is entertaining for three hours and yields six good clips passes. A stream that is dead air punctuated by manufactured spikes fails, and it fails for the people who watched it live long before any clip police convenes: they sat through the dead air.

If you cut your own clips

None of this is a reason to make your material harder to clip. It is a reason to be clear about which of two things you are producing.

If you are paid per thousand views, the practice described in the chronicle is just the strategy working. Volume, arrest, and no particular attachment to any individual clip.

If the clip is an advertisement for your own source material, it carries a second obligation that a paid clip does not: it has to survive the viewer arriving at the full thing. A hook that a farmed moment cashes in the feed is spent the moment someone opens the stream and finds nothing around it. The practical version is the same test we keep landing on: would the person in the clip, and someone who watched the full broadcast, both recognise the excerpt as fair? It is a low bar, it takes five seconds to apply, and it is the only difference between a clip that earns you a viewer and one that earns a view.

The second thing to take from the chronicle is defensive. The clip police is real audience behaviour now, whatever you think of it, and the cost of a clip that merely looks farmed is paid by the person in it. Two seconds of context left at the head of a cut (the question that prompted the reaction, the line that set it up) costs you almost nothing in retention and removes the only reading of the clip you cannot argue with afterwards.

Sources

This piece is built on one four-minute chronicle, and it is worth separating what it reports from what we have added:

• Mouv' (Radio France), Culture Internet, « La tendance clip farming : ou quand les spectateurs confondent sincérité et mise en scène », by Tahzio, 15 May 2025: listen to it (in French). The definition of clip farming, TikTok as an acquisition channel for Twitch, the accusation of lost authenticity, the AmineMaTue case and the campaign against him, the "clip police" and the argument that the ecosystem imposes the strategy are all reported there. The three quoted passages are ours in translation; the French is on that page.
• The reading is ours: that the accusation is structurally unprovable because a clip is the removal of the evidence, that selection makes spontaneous and manufactured moments look alike, and the shift from was it staged to does the clip represent the broadcast. Radio France makes none of these claims.
• For the money underneath all of it, see how the paid clipping economy works; for how the moments get chosen, what makes a moment clippable.