Meta calls Polish scam-ad report fiction and counters with its own numbers

Meta has answered a Polish report on deceptive advertising with both barrels, calling the work of the Instrat foundation "fiction" and countering with numbers of its own: 137,000 scam ads removed in a year, 88% of them before any user reported them, and an 83% fall in Poland's scam-ad report rate since July 2024. The dispute matters because every number on the table comes from the platform being audited, and because Meta's defence now leans on freshly deployed AI, from an anti-impersonation system to an assistant that helps people report scams.

Meta AI Watch analysis

What happened

Meta's Polish-language newsroom post, published on Friday, responds to a report the Instrat foundation released earlier in the week which, by Meta's account, estimates the revenue deceptive ads generate in Poland. Meta says the report is fiction prepared by a law firm representing a person who has sued the company, and that its central allegation, that Meta knowingly turns a blind eye to scam advertising, is entirely false.

The rebuttal then dismantles the report's arithmetic as Meta presents it, before pivoting to Meta's own figures and a list of new measures for Poland.

The central claims

• Every removal counted as a scam
  Meta says the report treats all ads Meta removed as fraudulent, when removals also cover restricted goods, third-party intellectual property and ad quality.
• EU reach doing duty for Poland
  Per Meta, the authors used reach figures covering all 27 EU countries as if they described Polish users, inflating the result.
• A year from three days
  Data from October, December and January differed by up to 80%, then was averaged into an annual figure with no margin of error, Meta says.
• A primed sample of 108 ads
  One iOS device, one profile deliberately trained by viewing and clicking to attract certain ads, which Meta calls unrepresentative.
• 137,000 scam ads removed
  Meta's own count for July 2025 to June 2026, with over 88% taken down before anyone reported them.
• Report rate down 83%
  User reports of deceptive ads per impression in Poland fell that far between July 2024 and June 2026, on Meta's numbers.

Why it matters

The new measures are the practical half of the post. Poland is in the first group of European countries to get Meta's upgraded anti-impersonation AI system, which the company says has detected and removed 50% more ads impersonating public figures than its predecessor. Every financial-services advertiser targeting Poland will face verification, part of a company-wide target for 90% of Meta's worldwide ad revenue to come from verified advertisers by the end of 2026. And a Scams Protection Hub on Facebook and Instagram now includes a Meta AI assistant that helps people find and report a scam from their recent interactions.

The strategic half is subtler. Platform accountability fights increasingly turn on machine-generated statistics, and whoever owns the measurement owns the argument. Meta has just demonstrated the playbook: attack the auditor's maths in detail, then refill the vacuum with your own dashboard.

Our read

Meta's methodological criticisms, as it presents them, are the kind that can gut an estimate: reach data spanning 27 countries standing in for one of them, a year extrapolated from three days that differed by up to 80%, and a sample of 108 ads gathered from a profile deliberately primed to attract the very thing being measured. If those characterisations are fair, the report's headline number was in trouble before Meta wrote a word.

But a rebuttal is still a defence exhibit. Every figure answering the report comes from Meta itself, and the report speaks in this post only through Meta's summary of it. Instrat's response is the missing half of this argument, and the sensible move is to admire the footwork without declaring a winner.

What to watch

• Instrat's response
  The foundation's answer to the methodology criticisms is the missing half of the dispute.
• Verification rollout
  Whether 100% verification for financial-services advertisers targeting Poland ships, and on what timetable.
• The 90% target
  Progress towards verified advertisers supplying 90% of Meta's worldwide ad revenue by the end of 2026.
• Independent measurement
  Any outside check of scam-ad prevalence on Meta's platforms, in Poland or across the EU.

Discussion spark: Meta attacks the report's maths but answers it with its own measurements. Can a platform's enforcement statistics ever settle a dispute about its own enforcement, or does this argument need an outside referee?

not affiliated with, endorsed by, or operated by Meta