Frontier AI models now judge ads. On half the calls, they disagree.
We gave the same brief to fourteen models and had seven of them judge the scripts in blind pairs. The judges could not agree on almost half of the pairs.
These judges will soon decide which ads get written, tested and trained on. Before that happens, we want to know whose judgment they match. Yours is the reference.
One real pair from the study. The ad is for a commuter who is embarrassed by the smoke smell in his car before a colleague gets in. Which opening would you test?
Your judgment, measured
After ten pairs you see which AI judge decides most like you, and which models wrote the scripts you chose.
The results first
You get the full study, with the human results, before it is public.
Credit, if you want it
Your name in the published dataset as one of the people who set the reference.
Judge your first pair
Each pair is a brief and two complete scripts. You pick the one you would test first. Notes are optional. Most people take about two minutes a pair.
You never see which model wrote a script until your report.
The scripts were written by AI models for a real Indian car-perfume brand. The brand's name and trademarks are changed. The judges and the method are described on the Copywriting Arena methodology page.