It's a great question that probably requires a subject matter expert. I'd be happy to connect off-line and do a deeper dive with our subject matter experts on this. In sum, this would be continuous analysis running in the background on the types of behavioural cues. For example, I talked about our onboarding process. When we take on a new advertiser and are doing a trust assessment, we have our advertiser verification process. Then, we have something called limited ad serving.
The point here is that scam networks rely on three things. They rely on speed, volume and deception. We've built our systems to be economically unattractive to scammers, so that we are too slow, too limited in reach and apply too much transparency for it to be worth their time.
In the first place, we have the advertiser verification process. Then, we have limited ad reach. That throttles their ability to reach users. A scammer is casting a wide net. They want to reach millions of people in the hopes that a handful might fall for that deception. Then, we apply the transparency rules of your ad centre, being able to see who the advertiser is and where they are located. This is where, on that limited ad serving.... In the background, Gemini will be assessing behavioural patterns to understand if this actor is trustworthy or if there are....
For example—