How to Subject Line Format Test
Format is the silent variable in most subject line experiments. Whether you lead with an emoji, a number, a question, or a personalized name, the structure of the line shapes opens before the words even register. This guide shows how to subject line format test methodically with marginal, the hosted email MCP server your AI agent can drive directly.
What a format test actually measures
A subject line format test isolates structure rather than copy. Instead of comparing two unrelated lines, you hold the core message constant and vary the format — question vs. statement, short vs. long, brackets vs. plain text, number-led vs. word-led. That isolation is what makes the result repeatable across future campaigns.
- Question format: "Ready to renew your plan?"
- Statement format: "Your plan renews this week"
- Number-led: "3 things to do before renewal"
- Personalized: "{first_name}, your plan renews soon"
- Bracketed tag: "[Reminder] Plan renewal"
Running the test through an email MCP server
marginal exposes four MCP tools that map cleanly onto a format test workflow. You describe the formats you want to compare, and your AI agent calls the tools end to end — no manual copy-paste between an ESP dashboard and your editor.
Connect any MCP client — Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, Zed, or OpenAI Codex — to the endpoint at https://marginal.sh/mcp using a Bearer API key. From there the agent orchestrates the experiment.
- generate_variants — produce subject lines in each target format
- launch_test — start the subject line A/B test on a managed send
- get_results — pull open/click tracking and within-test lift
- recommend_next — suggest the next format to try based on results
Reading lift and picking a winner
Once the test is live, marginal tracks opens and clicks per variant and reports within-test lift so you can see which format outperformed the control. For a format test, focus on open rate first — the structure of the line is primarily an open-rate lever — then confirm the winner did not hurt downstream clicks.
Don't declare a winner on tiny samples. Let enough volume accumulate that the lift is stable, then let recommend_next propose the follow-up: a sharper version of the winning format, or a new structural angle to challenge it.
Make it part of your agent workflow
The value of running format tests through AI agent email marketing is compounding. Every test teaches your agent which structures resonate with your list, and those learnings feed the next round of generate_variants calls. The free tier covers 100 experiments per month, which is enough to test a new format weekly with room to spare.
See the tool reference and example payloads in the docs at https://marginal.sh/docs/ to wire format tests into your existing campaign pipeline.
Get started with marginal
marginal is a hosted email marketing MCP server at marginal.sh. Sign up free, create an API key, and connect https://marginal.sh/mcp from Cursor, Codex, or Claude Desktop.
- 100 experiments/month on the free tier
- Four MCP tools: generate_variants, launch_test, get_results, recommend_next
- Listed in the MCP Registry as sh.marginal/mcp