Email Subject Line A/B Test Generator

Why this prompt works

Most 'write me 10 subject lines' prompts produce ten variations of the same subject line wearing different hats. The fix here is the explicit list of techniques the model has to cover (curiosity gap, social proof, urgency, personalisation, question format), and the requirement that each pair tests one specific variable. That last bit is what turns the output from 'here are 10 alternatives' into 'here are 10 actual experiments'. The 50-character cap matters because mobile inboxes truncate longer subject lines, and the per-pair hypothesis forces the model to articulate why it thinks Version B might win, which is a sanity check on whether the variable is testable at all.

When to reach for it

• You're filling out an email calendar and need a starting set of A/B tests, not just 'good subject lines'.
• You've run a few campaigns and want to systematically test the variables you haven't covered yet (e.g. you've tested length but never personalisation).
• You're handing the test plan to someone else and want each pair to come with the rationale baked in, so they don't accidentally test two variables at once.
• You're new to lifecycle email and want a structured way to learn what your audience responds to.

How to customise it

The audience segment is the load-bearing variable. 'Existing customers who haven't purchased in 30 days' produces tighter output than 'past customers'. Specify the offer too: 'a 20% off code on running shoes' beats 'a discount'. If your list is small (under 5,000), drop the request to 10 pairs and ask for 5; running 10 simultaneous tests on a small list gives you no statistical signal. For brands that already have a known winning style (e.g. consistently low-emoji, plain-text), tell the model 'avoid emoji' explicitly, otherwise it'll insert them by default.

What good output looks like

You'll get 10 numbered pairs, each with the two subject lines, the variable being tested in plain English ('emoji vs no emoji', 'first-name personalisation vs generic'), and a one-line hypothesis. Lengths usually come in just under the 50-character limit. The hypotheses are the most useful part for sceptics on your team: they make it harder to dismiss a test as random tweaking. The numbered list output pastes cleanly into a planning doc or spreadsheet without reformatting.