Keeping Privacy-Violating and Other Harmful Ads Off Meta's Apps

Meta's apps, including Facebook, Instagram and Messenger, show ads from a very large number of advertisers. As a product manager at Meta, how would you keep bad ads, for example ads that violate people's privacy, from being shown to people on these apps? ```hint Look past the ad itself List the points in an ad's life where a problem could be caught, then ask where a privacy violation would actually be visible, since it is not always in the image or text a reviewer sees first. ``` ### Clarifying Questions - Which kinds of bad ads are in scope: privacy violations only, or also scams, misleading claims, prohibited products and discriminatory targeting? - What counts as a privacy violation here: ads that imply they know a viewer's personal traits, ads or landing pages that collect personal data deceptively, ads for tools that spy on other people, or ads targeted with data the advertiser should not have? - Should success be judged by how many people see bad ads or by the harm those ads cause, and how much review delay and friction for legitimate advertisers is acceptable? - Does the question cover only new ads at submission, or also ads that are already running? ### What a Strong Answer Covers - A working definition of bad ads, with privacy violations broken into concrete patterns that need different detection - Defenses across the ad's lifecycle: who may advertise, review of the creative, landing page, lead form and targeting before launch, monitoring while live, and user reports - How automated systems and human reviewers split the work by risk, confidence and reach, and how repeat offenders are handled - A metric for people's exposure to violating ads, with guardrails for false positives, review delay, appeals and revenue - Adversarial behavior such as cloaked landing pages and returning banned advertisers, and the trade-off between strictness and advertiser friction ### Follow-up Questions - An advertiser shows reviewers a harmless landing page but sends real users to a page that harvests their personal data. How would you catch this? - Stricter review cut violating impressions, but small advertisers now wait much longer for approval and appeals have doubled. What do you do? - How would you measure the bad ads your system never catches?