The classification problem
When a detection system identifies part of the traffic as automated, the remainder initially means only: not classified as a bot by those rules. It does not logically follow that every remaining visit has been positively confirmed as human.
Why modern automation is harder to classify
Automated systems can use real browsers, residential proxies, emulated or plausible device signals and increasingly realistic interaction patterns. AI agents may also act without clearly identifying themselves as automation. The boundary between technical plausibility and real human use becomes more demanding.
Ad fraud is broader than the bot question
Ad fraud can also involve manipulated supply paths, spoofing, hidden ads, refresh mechanics, click manipulation or conversion fraud. The question 'was it a bot?' is therefore often too narrow. What matters is whether the paid advertising service technically and economically matches what was agreed and reported.
Positive evidence instead of remainder logic
Media Forensics also asks what evidence supports a plausible human-use context behind an impression or interaction. Multiple signals, timing, technical environment, supply chain and observability limits are evaluated together.
Unknown is a valid category
A useful classification separates plausibly human activity, legitimate automation, confirmed problematic patterns and areas that remain unclear because evidence is insufficient. 'Unknown' is methodologically stronger than forcing every event into a complete-looking classification.
This article describes principles and questions. Specific fraud, bot or quality rates cannot be transferred directly to individual campaigns without understanding methodology, scope and data basis.