Meta exempted some of its top advertisers from its usual content moderation procedure, shielding its multibillion-dollar business amid internal concerns that the corporation’s systems mistakenly penalised top brands.
According to internal documents from 2023 seen by the monetary Times, the Facebook and Instagram owner introduced a series of “guardrails” that “protect high spenders”.
The previously unreported memos said that Meta would “suppress detections” based on how much an advertiser spent on the platform, and that some top advertisers would instead be reviewed by humans.
One document suggested that a throng called “P95 spenders” — those spending more than $1,500 per day — were “exempt from advertising restrictions” but would still “eventually be sent to manual human review”.
The memos predate this week’s announcement by chief executive Mark Zuckerberg that Meta was ending its third-event truth-checking programme and dialling down its automated content moderation, as it prepares for Donald Trump’s profit as president.
The 2023 documents display Meta had found that its automated systems had incorrectly flagged some top-spending accounts for breaches of the corporation’s rules.
The corporation told the FT that higher-spending accounts were disproportionately subject to erroneous notifications of feasible breaches. It did not respond to questions asking whether any of the measures in the documents were temporary or ongoing.
Ryan Daniels, a Meta spokesperson, said the FT’s reporting is “simply inaccurate” and “based on a cherry-picked reading of documents that clearly state this attempt was intended to address something we’ve been very community about: preventing mistakes in enforcement”.
Advertising makes up the majority of Meta’s annual revenues, which were nearly $135bn in 2023.
The tech giant typically screens adverts using a combination of artificial intelligence and human moderators to stop violations of its standards, in an attempt to remove material such as scams or harmful content.
In a document titled “high spender mistake prevention”, Meta said it had seven guardrails protecting business accounts that bring in more than $1,200 in returns over a 56-day period, as well as person users who spend more than $960 on advertising over the same period.
It wrote that the guardrails assist the corporation “decide if a detection should proceed to an enforcement” and were designed to “suppress detections . . . based on characteristics, such as level of advertising spend”.
It gave as an example a business that “is in the top 5 per cent of returns”.
Meta told the FT it uses “higher spend” as a guardrail because this often means the corporation’s adverts will have greater reach, and so the consequences could be graver if a corporation or its adverts are mistakenly removed.
The corporation also acknowledged that it had prevented some high-spending accounts from being disabled by its automated systems, instead sending them for a human review, when the corporation was concerned about the accuracy of their systems.
However, it said that all businesses were still subject to the same advertising standards and no advertiser was exempt from its rules.
In the “high spender mistake prevention” memo, the corporation rated different categories of guardrails as “low”, “medium” or “high” in terms of whether they were “defensible”.
Meta staff designated the habit of having spend-related guardrails as having “low” defensibility.
Other guardrails, such as using knowledge of the trustworthiness of the business to assist it decide whether a detection of a policy violation should be automatically acted on, were labelled as “high” defensibility.
Meta said that the term “defensible” referred to the hardship of explaining the concept of guardrails to stakeholders, should they be misinterpreted.
The 2023 documents do not name the high spenders that fell within the corporation’s guardrails, but the spending thresholds recommend thousands of advertisers may have been considered exempt from the typical moderation procedure.
Estimates from economy intelligence firm Sensor Tower recommend that the top 10 US spenders on Facebook and Instagram include Amazon, Procter & Gamble, Temu, Shein, Walmart, NBCUniversal and Google.
Meta has achieved record revenues over recent quarters and its ownership is buying and selling at an all-period high, following the corporation’s recovery from a post-pandemic slump in the global advertising economy.
But Zuckerberg has warned of threats to its business, from the rise of AI to ByteDance-owned rival TikTok, which has grown in popularity among younger users.
A person familiar with the documents argued the corporation was “prioritising returns and profits over user integrity and health”, adding that concerns had been raised internally about circumventing the standard moderation procedure.
Zuckerberg said on Tuesday that the complexity of Meta’s content moderation structure had introduced “too many mistakes and too much censorship”.
His comments came after Trump accused Meta last year of censoring conservative talk and suggesting that if the corporation interfered in the 2024 election, Zuckerberg would “spend the rest of his life in prison”.
The internal documents also display that Meta considered pursuing other exemptions for sure top-spending advertisers.
In one memo, Meta’s staffers proposed “more aggressively offering protections” from over-moderation to what it dubs as “platinum and gold spenders”, which together bring in more than half of advertising returns.
“untrue positive integrity enforcement against High worth Advertisers costs Meta returns [and] erodes our credibility,” read the memo.
It suggested an alternative of a blanket exemption for these advertisers from sure enforcements, except in “very rare cases”.
The memo shows that staff concluded that platinum and gold advertisers were “not a suitable segment” for a broad exemption, because an estimated 73 per cent of its enforcements were justified, according to the corporation’s tests.
The internal documents also display that Meta had uncovered multiple AI-generated accounts within large spenders’ categories.
Meta has previously fallen under scrutiny for carving out exemptions for significant users. In 2021, Facebook whistleblower Frances Haugen leaked documents showing that the corporation had an internal structure called “cross-check”, designed to review content from politicians, celebrities and journalists to ensure posts were not mistakenly removed.
According to the Haugen documents, this was sometimes used to shield some users from enforcement, even if they broke Facebook’s rules, a habit known as “whitelisting”.
Meta’s oversight board — an independent “Supreme Court”-style body funded by the corporation to oversee its most challenging moderation decisions — found that the cross-check structure had left risky content online. It demanded an overhaul of the structure, which Meta has since undertaken.
