Meta exempted advertisers who bring in the most money from standard moderation

Meta✴ has exempted some of the leading advertisers on its platforms from standard content moderation practices. It took the measure to protect its multibillion-dollar business, fearing that its automated moderation systems could inappropriately sanction leading brands, writes the Financial Times (FT).

Image Source: Christin Hume / unsplash.com

According to internal Meta✴ documents seen by the FT, Facebook✴ and Instagram✴ have introduced “protection mechanisms” to benefit customers who spend significant amounts on the platforms. These mechanisms take precedence over automated moderation controls on platforms—posts from top advertisers are reviewed only by humans. For example, there is a term called “P95 customers,” which refers to customers who spend more than $1,500 per day. Such clients are exempt from the automated moderation system, and their posts are sent for manual human review. The information comes before Meta✴ CEO Mark Zuckerberg said this week that the company would overhaul its fact-checking program and reduce the severity of its automated moderation tools.

Meta✴ discovered that its automated moderation systems were incorrectly flagging accounts with high ad spend as violating social media rules, according to internal company documents from 2023. The number of false positives from automated moderation systems was disproportionately high for high-spending accounts, Meta✴ told the FT; The company, however, did not say whether any of the “protective mechanisms” were temporary or whether they continue to operate on an ongoing basis. A Meta✴ spokesperson called the FT’s submissions “simply inaccurate” and “based on a selective reading of documents that make clear what we have stated publicly: preventing errors in the application of the rules.” Advertising revenue makes up a significant portion of Meta✴’s revenue, which amounted to almost $135 billion at the end of 2023.

Image source: Corinne Kutz / unsplash.com

In an effort to prevent fraudulent or harmful content from appearing on the platform, the company reviews advertisements using artificial intelligence-based moderation tools and employing humans to complete the task. One of Meta✴’s internal documents states that there are seven mechanisms in place aimed at protecting business accounts that generate more than $1,200 in revenue in 56 days, and individual accounts that spend more than $960 on advertising in the same period. It also states that such mechanisms help the company make decisions about actions when receiving signals from AI moderation systems – they were designed to “suppress” signals, taking into account criteria such as the level of advertising spending.

Some accounts with high advertising costs are disconnected from automated moderation and transferred to specialized specialists for maintenance, Meta✴ admitted, but emphasized that no advertiser is exempt from the obligation to comply with the rules applicable to the platforms. The barrier categories for high-spending advertisers are categorized as “low”, “medium” and “high” in terms of how “justifiable” they are. Mechanisms based on the level of expenses have a “low” priority, and the advertiser’s business reliability criterion corresponds to “high”. Taking into account the threshold values ​​specified in the documents, several thousand advertisers could be exempt from standard moderation measures. According to Sensor Tower analysts, the top ten largest advertisers on Facebook✴ and Instagram✴ include Amazon, Procter & Gamble, Temu, Shein, Walmart, NBCUniversal and Google.

In one of the documents, Meta✴ employees propose to more actively offer protection from AI moderation to clients classified as “platinum and gold buyers” of advertising – the excessive strictness of automated systems, in their opinion, “costs Meta✴ revenues and harms our reputation.” They were proposed to be removed from the operation of a number of systems, with the exception of “very rare cases,” but they did not completely exempt the category of “platinum and gold” clients from moderation – in 73% of cases, the actions of automated systems were justified.