LLMs infer gender and age from ad patterns alone
A study conducted by UNSW Sydney and QUT, analyzing over 435,000 Facebook ads from 891 Australians, found that large language models (LLMs) can infer user demographics such as gender and age solely from advertising activity patterns.
Key Takeaways
- The study covered more than 435,000 Facebook ads from 891 Australians.
- UNSW Sydney and QUT found LLMs could infer gender and age from ad activity patterns alone.
- The analysis focused on advertising behavior, not on ad content or user-provided profile data.
- Facebook was the platform used in the dataset examined by the researchers.
Why It Matters
If ad activity patterns are enough for LLMs to infer basic demographics, the privacy boundary around monetization data is thinner than many teams may assume. For streaming and ad-tech operators, that raises the value of scrutinizing what can be derived from impression, targeting, and activity logs even when direct identifiers are absent. The study also shows how model capability can turn routine ad telemetry into a profiling signal. What to watch next: whether researchers extend this method beyond gender and age to other demographic attributes using similarly large ad datasets.
Read full article at scworld.com