Meta has rejected a Polish report about scam advertising, arguing that its methodology overstates the scale of the problem while outlining new AI-powered enforcement and reporting tools. The dispute matters because both the criticism and Meta’s defence turn on numbers that ordinary users cannot easily inspect.
Meta AI Watch analysis
What happened
In a statement published on 18 September, Meta said the Instrat Foundation’s report incorrectly treated every advert removed by Meta as fraudulent. The company also challenged the report’s use of European reach data for Poland, assumptions about repeat views and advertising prices, and conclusions drawn from 108 adverts observed on one deliberately configured iOS profile.
Meta further argued that the report extrapolated from three days of data whose results differed by as much as 80 per cent. Those are Meta’s criticisms of the report, not an independent finding that the methodology is invalid. The company called the report misleading and said it was prepared in the context of a legal dispute involving a person who sued Meta.
The company says it has deployed an upgraded AI system in Poland to detect adverts impersonating public figures. Meta claims the system has detected and removed 50 per cent more such adverts than its predecessor. It also says it will require verification for all financial-services advertisers targeting Poland.
Meta said it removed 137,000 scam adverts between July 2025 and June 2026, with more than 88 per cent removed before anyone reported them. It also claimed that the rate at which users reported scam adverts fell by 83 per cent in Poland between July 2024 and June 2026. These figures come from Meta’s own statement and are not independently established in the supplied evidence.
Why it matters
The argument is not merely about whether one report got its arithmetic right. It is about who gets to define a scam advert, which data can represent a country, and whether a platform’s own enforcement statistics are enough to demonstrate that enforcement works. Spoiler: a company grading its own homework will always invite a raised eyebrow.
Meta says its Scam Protection Centre on Facebook and Instagram includes a Meta AI assistant that can examine recent interactions to help people find and report suspicious adverts, accounts or content. That gives readers a practical detail: reporting may be assisted by reviewing recent activity rather than relying on a half-remembered advert and a mounting sense of regret.
Our read
Meta has supplied specific methodological objections and concrete account-level figures, so this is more substantial than a routine corporate denial. But the statement is also advocacy from the platform being assessed. The useful conclusion is not that Meta has won the dispute, or that Instrat’s estimates are correct. It is that the underlying measurements need clearer definitions, representative sampling and independent scrutiny.
For users, the immediate takeaway is to treat Meta’s enforcement percentages as company-reported performance indicators, not a complete picture of scam exposure. The new AI reporting assistant may make complaints easier, but easier reporting is not the same thing as proven prevention.
What to watch
- Whether Instrat publishes a detailed response to Meta’s methodological objections.
- Whether independent researchers can reproduce or challenge the reported Polish figures.
- How Meta defines adverts removed for fraud compared with adverts removed for other policy reasons.
- Whether the AI reporting assistant produces faster, more transparent outcomes for users. The central question is simple enough, even if the spreadsheets are not: can a platform credibly measure its own success against scams without opening its methods to outside scrutiny?
Discussion spark: Should platforms be expected to publish independently auditable scam-ad data, or would exposing detailed detection methods create more opportunities for fraudsters?
Sources and evidence
- Source update (18 September 2026, 14:55 UTC)
not affiliated with, endorsed by, or operated by Meta