Twitch, which made headlines (derogatory) in August with platform-wide Amazon AI training that’s opt-in by default, is intertwining AI ever more inextricably with livestreaming despite overwhelmingly negative reception. During a recent podcast appearance, CEO Dan Clancy said that because AI already transcribes everything streamers say on the site, it’s now also being used to evaluate whether they’re brand friendly—and, if they’re not, yank ads from their broadcasts.
"We have our streamers label their content based on certain criteria,” Clancy said in a response to a question from the Snarketing Podcast (via Zach Bussey) about advertising and brand safety in the age of viewbotting and other social media-borne issues. “We have content classification guidelines [so] that they have over 99 percent consistency in labeling it. So then, when you as an advertiser tell us what your brand considerations are, you can make sure that it doesn't appear [next to] stuff that doesn't match your brand. And we also lately have been using AI to also identify streamers that may not match a brand. Because we transcribe streams to figure out 'Is this appropriate for this brand?' and then also to determine if something is happening that may not match a brand. So that's something that we've had to invest a lot in to make sure that for a brand, their content is appearing next to creators and content that fits what they're trying to do."
This is almost certainly separate from the much-reviled Amazon AI training, existing instead alongside Twitch-native features like AI-generated captions, stream summaries, and auto clips, as well as AI-powered platform safety initiatives—some of which streamers are able to opt out of and others of which they cannot. YouTube has relied on automated systems like this for years to demonetize videos its arcane labyrinth of algorithms deems inappropriate for whatever reason.
Twitch, on the other hand, has only somewhat recently begun dipping its toes into these waters, most notoriously by creating a category for hot tub streamers that advertisers could opt out of in 2021 and then, more recently, by requiring streamers to slap a “politics and social issues” label on corresponding broadcasts in 2024 so advertisers could avoid Hasan Piker and streamers of his ilk. (It is no coincidence that this happened following a bad-faith campaign around Piker and supposed antisemitism.)
Streamers, wary of both AI and YouTube’s infuriatingly un-transparent hall of mirrors, are not enthused about this new development.
“Ohhh so that’s why ad revenue dropped,” longtime streamer JP "itmeJP" McDaniel said on Twitter.
“I’m genuinely worried that the sheer fact that I’m a lesbian will cause the AI to remove ads,” said Vtuber LuminusRed. “In the past I’ve tested it and confirmed that I get less ad money if I had queer-related tags on my stream like ‘lesbian,’ so I am worried this dumb AI monitoring will mean I no longer get money from ads on Twitch.”
Content creators make money both by receiving a tiny percentage of the money a platform makes from each ad played during their broadcasts and from brand deals they strike themselves. For the latter, there’s etiquette in place—a series of unspoken (or contracted) rules. Some creators are worried that, on Twitch, that sanitized mode of conduct will soon become the default.
“For sponsorship streams I: do not curse, stay PG, avoid politics, [and stick to] Sesame Street-level humor,” said veteran writer, speaker, and content creator Steven Spohn. “When not sponsored I: swear like a drunken sailor, stay one [terms of service] notch below Eyes Wide Shut, slam the Orange Man Baby & cronies, worship Cthulhu, the dark overlord, the rightful heir to the universe. Almost all the creators I know do this. You sanitize and go brand-friendly when you need to; be a goofball when you don't. Agents literally tell you how to act in the briefs we get. We don't need AI ratting us out to Daddy when we tell too many dick jokes on our own time.”
There’s also the matter of whys and hows, or lack thereof.
“Because AI is going to be so accurate at understanding people, understanding context, accents, etc,” drag queen and streamer Annie Krevice said sarcastically.
“I wonder if a creator will ever have access to know what they got flagged for,” said Vtuber AlistJpeg. “My assumption is they probably won’t.”
Aftermath reached out to Twitch for more details about whether streamers will be notified when they’ve lost ads, as well as whetherAI or Twitch staff will make the final call. The company did not reply as of this publishing.
But that is the question: When a company begins replacing its own connective tissue with AI, where does accountability lie? Recently, companies like OpenAI have taken to offloading blame for much more serious matters like government hacking onto their agents, rather than accepting culpability themselves. Media and governments have helped prop up this framing. But AI is incapable of actual thought, meaning that it cannot make decisions or be held responsible for them. People are still in charge. They’ve just found a new, exceedingly convenient scapegoat.
Recommended
Nathan Grayson
