Case Study: Topic Exclusions as Brand Safety Lite
I have covered platforms like this for more than seven years, and Negative targeting reduces adjacency anxiety for cautious brands.
The useful question is not whether AI shows up in the stack. It already does. The question is who owns taste, safety, and measurement when the platform rewrites the asset.
- Write the hypothesis first
- Capture baselines before launch
- Separate correlation from causality
| Signal | Detail |
|---|---|
| Metric | Before → After / Signal |
| Shopping share | 22.1% → 28.4% |
| DSP Prime Day | 17.9% → 25.5% |
| Title recognition | ~33% knew Ted Lasso source |
Pros
- Clear decision metric
- Time-bound observation window
- Named data source
Cons
- Short windows can mislead
- Platform data is partial
- Creative confounders
None of this means the tools are useless. It means the craft job moved upstream. The people who win this cycle will brief better, version obsessively, and treat platform defaults as suggestions rather than destiny.
If you only remember one habit from this piece, make it this: screenshot the served creative, not just the uploaded master. What ran is the truth; what you approved in the deck is often a polite fiction.
Source: eMarketer. Numbers and product claims above come from that reporting; interpretation is mine.