Copy Arena — draft, read, refine, launch

When I set out to improve the copy on my website, I turned to a practical system called Copy Arena. This system eliminates the guesswork from selecting effective headlines by testing them with real readers. The process begins with drafting several headline candidates and ends with launching the most successful one. The idea was inspired by a founder's directive logged on September 12, 2026, in ledger F82: "we run a bunch of copy ideas and refine them until we find the best one."

The Problem It Solves

Before implementing Copy Arena, I faced a recurring issue: multiple headline changes within a few days, none of which were validated by readers. This was evidenced by a failed headline test where none of three blind company readers recognized the live headline (Traviseric.com/design-review/home-v2-2026-09-09/headline-ledger.html). A headline must resonate with readers within five seconds and prompt them to take action.

Output of Copy Arena

The system produces a ranked list of headline candidates, each evaluated by three personas over a five-second read. Each candidate receives scores on specific criteria such as for_me, promise_understood, and action_named, ranging from 0 to 5. It also captures the exact words that either disengaged or captured the reader's attention. The results, including a recommended winner, are saved to design-review/<topic>/copy-arena-<date>.html, ensuring I have a documented history of headline performance.

Implementation Steps

1. Define the Reader

Before drafting any headlines, it's crucial to establish the reader personas. Each site has one primary reader, and I create three personas based on this reader using files like JUDGE-PROMPTS.md and READER-PERSONAS.md. These personas are detailed with specifics such as their business type and what they value most. For my site, traviseric.com, the reader is defined by ledger F80 as a company with its own systems.

2. Draft the Candidates

I draft between six to twelve headline candidates, each comprising a full first screen: eyebrow, headline, lede, primary action, and a secondary action. The drafts vary in approach, focusing on the reader's situation, the mechanism offered, fears alleviated, and proof points. Importantly, each candidate adheres to the claim lock and honesty doctrine, avoiding exaggerated claims. These drafts are stored in candidates.json.

3. Render for Reader Experience

Using the command:

node client-factory/scripts/copy-arena/shoot-candidates.mjs --url <preview-or-local-url> --candidates candidates.json --out <dir>

I render each candidate as it would appear to a reader. This involves swapping the hero text on the real page and saving images of the page's appearance in both phone and desktop formats. This step ensures the five-second test is based on visual impact rather than mere text.

4. Conduct Blind Reads

For each candidate and persona combination, I employ an agent with the Pass A prompt from JUDGE-PROMPTS.md to evaluate the image. Judges work independently, ensuring unbiased feedback. The output is collected as a JSON block for analysis.

By following this methodical approach, I've transformed my headline selection process from guesswork to data-driven decision-making. Copy Arena provides a structured way to test and refine marketing copy, ensuring that the final choice resonates with real readers.