Vendors like WhiteOps, Integral Ad Science, Moat, Forensiq, DoubleVerify, etc. that are used to see if visitors to a website viewing the ads are actually people during normal browsing or whether they are bots or other fraudulent behavior.

None of them work well.

What do they do to decide if a user if real or fake? I am not remotely surprised that none of them work well

That's their trade secret but distills down to measuring all the various factors that a person actually reading a page would do: time, scroll speed/depth, click rate/location, previous history, mouse movements, http, ip, network traffic, browser profiles, etc.

Although these days with headless chrome and how easy it is to just record and replay someone's session, it becomes incredibly hard to detect with accuracy, especially while trying to not affect browsing behavior beyond the typical ad load experience.

There is lots of overlap with this and other security-focused companies that work with financial and government companies to protect their own websites and apps, although it's easier when you own and control the site, proxies and backend.

Jesus Christ, that is creepy and perverted.

Welcome to marketing! Get out while you are still sane.

What is exactly? It's an industry with lots of money (100s of billions USD) flowing through with little oversight and cheap startup costs thanks to cloud vendors and commodity tech.

If there are criminals willing to invest intense effort committing physical crimes, it's expected that there would be a just as much put into something like this. Similar to a lot of the bitcoin/crypto dealings where there is plenty of organized crime involved.

Pages viewed/clicked per second. Some sites contain tiny little links that no human should every click, but a bot might. Others get more advanced by comparing visors. If a group of visitors all display the same pattern (think near identical behavior) then they are bots.

There are lots of other ways used as well. Client-side JavaScript that tracks mouse movement, for example. A combination of all of these combined with ML models give surprisingly accurate results.

Probably something like IAS's video viewability tools[1]. There are some big players in the traffic verification space as advertisers like myself don't want to pay for garbage traffic.

[1] https://integralads.com/capabilities/video/