Problems with Transparency Reports:
Inconsistent URL naming schemes, e.g., Reddit has used six different schemes since 2019
Unstable URL naming schemes, e.g., Pornhub includes an arbitrary number in the name component that changes over time
No consistent index page: Amazon has no index X has archive and most recent report at different URLs No link to index page even if it exists
Problematic presentation
- Most companies do not include machine-readable data
- Reddit includes graphs as bitmaps, uses bright but low-contrast colors (white labels on yelllow bars)
- Pornhub also uses bitmap graphics, makes all sections expandable, defaulting to closed
- Google uses drop-downs to select reporting period for each metric
- HTML pages do not format reliably for printing to PDF, e.g., Pornhub ends up using black background for some of the content, rendering it unreadable
Problematic data
- Meta rounds counts
- TikTok discloses fractions of fractions of... with two digits of precision
- Reports to NCMEC usually are well-defined but see Discord
- Other metrics are imprecise, e.g., Meta's misleading label change
- Reported metrics change, e.g., Reddit, TikTok
- Data is incomplete, e.g., TikTok also breakdown of reasons for Reddit's chat messages or content-based appeals for 2023 and earlier
- Data includes redundant or false report submissions, e.g., Pornhub in 2020 and Amazon in 2025. Those two instances also account for the top two biggest mean differences. Also Pinterest 2025 H1, child safety policy was over-enforced due to "internal error", only acknowledged in footnote of transparency report. The firm does not explain its discrepancy in report counts for 2024.
Problematic reporting: see previous point