Let me speak a bit to the specific numbers there. The 10% was based on some internal research to identify the topline ceiling of potential violations. That included non-violating content that is often caught up in our systems. It included things that are not actually frauds and scams as well. That is a higher number than the reality.
The same document that was leaked and referenced in the Reuters article in 2024 pointed out that the actual number is likely in the 3% to 4% range. Again, that was in 2024. We reduced that by 50% in 2025, and we continue to work on that today.
On the angle of charging more, the ad auction system is incredibly complex. What it does is it reduces the value of a bid in an auction if we suspect but cannot prove that an ad is a scam.
This speaks to where we are on the scam attack chain. What we see is the very early stages of a scam, often with potentially innocent-looking ads, innocent-looking engagements or new friendships being formed. It is challenging—