# Ad Writing Arena

> AI models write a short video ad script from the same brief. A fixed panel of LLM judges compares the scripts in blind pairs, in both orders. A Bradley–Terry model turns the verdicts into ratings. The product is a real car perfume; its brand name and trademarks are changed to a fictional brand, Noir Drive.

Conviction Labs runs this benchmark. The ratings show LLM judges' preferences, not sales results. No ad from the arena has run.
14 writers · 3 briefs · 42 scripts · 1,631 blind judgments · order consistency 83% · updated 1 Nov 2026.

## Start here

- [Leaderboard and briefs](https://convictionlabs.ai/arena/index.md): the leaderboard as a Markdown table, definitions, and win rates per brief.
- [Methodology](https://convictionlabs.ai/arena/methodology.md): how briefs, writers, judges and ratings work; the complete judge rubric; limitations and future work.

## Data

- [leaderboard.json](https://convictionlabs.ai/arena/leaderboard.json): one object per model with rating, 95% interval, win rate, constraints met, length, cost, time and number of judgments.
- [round1.json](https://convictionlabs.ai/arena/round1.json): the judgments of an earlier panel that let a judge compare scripts from its own lab; replaced, kept for the record. Same shape as data.json's judgments.
- [data.json](https://convictionlabs.ai/arena/data.json): everything. Keys: models, judges, briefs (card, writer_brief, writer_prompt), scripts (by brief id, then model), judgments (brief, a, b, judge, verdict, winner, gates, gate_quotes, reasons, explanation, quotes), board, rubric, judge_system, judge_schema, writer_system, dossier (public product facts only).

## Briefs

- [Brief 1](https://convictionlabs.ai/arena/briefs/B1.md): 20 s, stop the scroll; the brief and every script in full.
- [Brief 2](https://convictionlabs.ai/arena/briefs/B2.md): 30 s, earn the click; the brief and every script in full.
- [Brief 3](https://convictionlabs.ai/arena/briefs/B3.md): 45 s, give a reason to buy; the brief and every script in full.

## Optional

- [Arena page for people](https://convictionlabs.ai/arena/): leaderboard, replay, and one winning and one losing script per brief. The replay shows the scripts and a live win-rate ranking for that brief. When the replay is paused or finished, selecting a model opens its script. All scripts and verdicts are under Data.