Preview Text vs Subject Line Test: What to Measure First
When inbox opens stall, growth teams often debate the preview text vs subject line test question. Both shape what a recipient sees before clicking, but they pull different levers — and confusing the two muddies your results. Here's how to think about each, and how to run clean tests with marginal.
What each element actually controls
The subject line is the headline of the inbox. It's the boldest text and the biggest single driver of whether someone opens at all. Preview text (the gray snippet rendered after the subject in most clients) is the supporting line — it adds context, urgency, or curiosity that the subject couldn't fit.
Because they sit side by side, changing both at once makes it impossible to attribute a lift. A proper preview text vs subject line test isolates one variable so the open-rate delta you measure is real.
- Subject line: primary open driver, highest visual weight, shortest character budget.
- Preview text: secondary context, varies by client truncation, easy to ignore if left blank.
- Changing both = confounded results — you won't know which caused the lift.
How to sequence the tests
Most teams get more leverage by nailing the subject line first, then optimizing preview text on the winner. The subject controls the larger share of open variance, so lock it before fine-tuning the snippet that supports it.
Run each as a single-variable subject line A/B test. Hold preview text constant while you test subjects, then hold the winning subject constant while you test preview text. Two clean experiments beat one ambiguous one.
- Phase 1: fixed preview text, two or more subject variants.
- Phase 2: fixed winning subject, two or more preview-text variants.
- Track within-test lift on opens for each phase separately.
Running both tests with marginal
marginal is a hosted email MCP server that lets an AI agent drive these experiments end to end. Connect a client like Cursor, Claude Desktop, or Windsurf to the endpoint at https://marginal.sh/mcp and ask the agent to generate variants, launch a test, and report results.
The tool surface keeps each phase isolated: generate_variants produces candidate subjects or preview-text lines, launch_test ships the managed send with tracking, get_results returns open and click metrics with within-test lift, and recommend_next suggests the follow-up experiment — for example, moving from subject to preview text once a winner is clear.
- generate_variants — draft subject or preview-text candidates on demand.
- launch_test — managed send with open/click tracking built in.
- get_results — within-test lift so you compare phases fairly.
- recommend_next — guides the handoff from subject to preview text.
Why an MCP workflow helps here
Sequencing two dependent tests by hand is fiddly: you have to remember to freeze the right variable each round. With AI agent email marketing through marginal, the agent carries that context across phases and won't accidentally vary both at once.
The free tier covers 100 experiments per month, which is plenty to run a multi-phase preview text vs subject line test on a real campaign. See the docs at https://marginal.sh/docs/ for tool schemas and setup.
- Auth uses a Bearer API key (marg_live_...) — no marketing-platform keys required.
- Hosted server, so there's nothing to deploy or maintain.
- Works with Cursor, Claude Code, Cline, Continue, Zed, OpenAI Codex, and more.
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