Ofcom finds “sharp jump” in correctly labelled Twitch gambling streams

Ofcom has ruled that Twitch’s content classification labelling (CCL) shift this year resulted in a “sharp jump” in the correct classification of gambling streams on the platform.

In May, the Amazon-owned streaming platform rolled out CCLs to help users block streamers hosting potentially offensive and mature content under a host of categories, including mature-rated games, gambling, violence, drugs, sexual themes and significant profanity or vulgarity. 

The shift also automatically blocked gambling and sexual content for logged out users or those without a Twitch account.

Gambling content is applied to content creators “participating in online or in-person gambling, poker, or fantasy sports that involve the exchange of real money”.

When enforcing the change, Twitch granted creators a 30-day grace period “to become accustomed to the new labelling system” but were issued warnings for the mislabelling of content. That grace period ended on 20 July 2023.

Ofcom elected to investigate the change as part of its discussion paper series, with the UK’s communications regulator releasing a 66-page document on 2 September.

In its main findings, Ofcom noted that while creators began to label their streams more accurately, the CCL change did not result in creators altering the type of content they created.

In addition, it said: “We found that the CCL change did not alter viewer behaviour materially: the number of views of mature streams did not substantially differ before and after the CCL change.”

Prior to the CCL change, the regulator’s analysis of Twitch streamers labelling their content found 65% of gambling streams were accurately labelled as mature, second behind mature-rated games streams at fewer than 50%.

In comparison, sexual themes-linked streams were labelled as mature 84.7% of the time.

However, there was a “sharp jump” in the labelling rate of gambling streams following the 30-day grace period, which Ofcom said had highlighted the “intervention’s role in further increasing labelling of such mature content”.

The probability of accurately labelling gambling streams increased by 22 percentage points after the 30-day grace period, compared to 49 percentage points for mature-rated games and 16 percentage points for games with sexual themes.

Further research found streams containing gambling, drugs or intoxication, and violent or graphic content appear less frequently in mature streams – no more than 8% of total CCLs.

The researchers admitted that there was a lack of a clear control group due to the labelling process being forced upon all Twitch users.

Ofcom did develop an in-house classifier model based on machine learning to compare the pre- and post-CCL Twitch landscape. 

The regulator said its model was “extremely likely to be correct when it predicts that a stream has a CCL”.

When presenting its findings, Ofcom said: “We found that more specific content labels, coupled with penalties for inaccurate labelling, led content creators to positively change their behaviour in terms of how accurately they labelled content. 

“This change resulted in users being provided with much clearer alerts for mature content.”

It added: “However, we did not see any significant impact on the amount of mature content users viewed following the change, nor did we see a significant change in the amount of mature content creators produced.”

EGR has approached Twitch for comment.