Ten AI subject lines, one send: how to choose without a split test
A five-point scoring method for picking between generated options when your list is too small for a split test to mean anything.
Ten subject lines on the screen, all decent, none obviously best. You read them until the words stop meaning anything, pick number four because it was first to feel right, and send. Two weeks later you do it again with no more information than you had the first time. The problem is not that you lack a testing tool. It is that on a list of a few hundred people a split test cannot tell you anything reliable, so you need a way to choose that does not pretend to be statistics.
Why small lists cannot split test honestly
Say 520 people get your campaign and you split them in half. Each version goes to 260. Version A gets 61 opens, version B gets 68. That looks like a win of about three percentage points, and it is very likely to be nothing at all: a slightly different mix of people, one version landing while more of them were at lunch, an office network being slow. Run it again next week and the ordering may well flip.
Open rates have also become a soft measure. Privacy features on phones and mail apps pre-fetch images and mark messages as opened when no human has looked at them. So you are reading a noisy signal from a small sample. Two sources of error, one conclusion, and it is often the wrong one.
This is genuinely awkward, and it is worth being blunt about it: there is no clever method that turns 520 subscribers into trustworthy data from a single send. What you can do is stop choosing at random. A repeatable scoring pass will not tell you which line wins, but it reliably removes the four or five that were going to underperform.
Choosing on instinct, or choosing on criteria
The point of the right-hand column is not that scoring is scientific. It is that scoring is consistent, and consistency across a year is what eventually teaches you something about your own audience.
| Picking by feel | Scoring, in the store.fan composer |
|---|---|
| Ten options, one gut call, no reasoning you can repeat | Five criteria, a score out of ten, and a decision you can defend to yourself next month |
| No memory of what you used last time | Your sent campaigns are listed, so you can see the phrasing you already used |
| Length judged by eye in a desktop editor | The subject field takes up to 150 characters, so you count deliberately and write for the first 35 |
| One line written for everybody at once | Audience picked first, so the line is written for buyers, leads or members specifically |
| Success measured on opens alone | Per-product views, buyers and revenue show whether the send moved anything real |
| Every campaign a fresh guess | A running note of line, audience and outcome that becomes your own evidence over six months |
The five-point score
Give each option 0, 1 or 2 on each criterion. Maximum ten. Do all ten lines quickly, in one pass, without going back to improve your favourite.
- 1Specificity. Is there a concrete noun, number or name in it? “The 42-page glaze guide” scores 2. “Something new for you” scores 0. Vagueness is the most common failure in generated lines and the easiest to spot.
- 2Phone truncation. Does the meaning survive the first 35 characters? Cover the rest with your thumb and read what is left. If the important word is at the end, it does not exist.
- 3Honesty. Does the line describe what is actually inside? A curiosity gap you do not close teaches people to stop opening, and that cost lands on the next email, where you will not attribute it correctly.
- 4Human sound. Would you say it to somebody? Read it aloud. Anything with a colon in the middle and a benefit on both sides is a headline, not a subject line.
- 5Difference. Is it distinct from your last three subject lines? Repetition in the inbox reads as noise, however good the individual line is.
Scoring ten options on a list of 520
A photographer briefs a draft for a $19 lightroom preset pack and gets ten lines back. Three go immediately on specificity: “A little something for your weekend”, “You are going to love this”, “New drop, don't miss out”. Two more fail the thumb test, since both put “presets” after 50 characters. One reads as an advert out loud: “Better edits, faster workflow: the pack is here”. That leaves four.
Of the four, “12 presets, and the one I use on every wedding” scores 9: specific, front-loaded, true to the email, sounds like her, and unlike anything she sent last month. The runner-up, “The preset pack is live ($19)”, scores 8 and would have been perfectly fine. That is the useful outcome here. Two good options and a defensible reason, in about four minutes, rather than twenty minutes of staring and a coin flip. She notes the line and the date in the same file where she keeps the briefs, and moves on.
Set up your storefront and your first product free, then decide whether the AI drafting is worth adding.
Create your storeThe mistake most people make
They optimise the subject line and ignore the sender. Most people decide whether to open based on who it is from, not on the cleverness of six words. If your name is unfamiliar, because it has been eleven weeks since your last email, no subject line rescues that. Send regularly enough to be recognised, keep your list clean, and treat the subject as the second most important thing on the line. If you are still building the list itself, free lead magnets and the customer list are where that starts, and both work before you ever write a campaign. The full picture of what sits on which plan is on the pricing page.
Broadly, once each half of a send is in the thousands, and only for differences large enough to see. Below that, consistency and record-keeping teach you more than any single test.
You can send to different segments, but buyers and leads are not comparable audiences, so the result tells you about the audiences rather than the lines. It also spends twice from your monthly allowance.
The field allows up to 150 characters, but write for the first 35 or so. Longer lines are not penalised, they are simply not read past the truncation point on most phones.
One, occasionally, if it is how you normally write. As a habit it fades quickly, and a line that needs an emoji to be interesting usually needs a rewrite instead.
Ten options is a luxury, not a problem, as long as you have a method. Four minutes of scoring beats twenty minutes of deliberation, and the note you keep afterwards is the thing that will eventually make the choice obvious.
Start collecting the list today: a free account gives you the page, the products and full CSV export whenever you want it.
Open a free accountTurn your knowledge into income
Launch your Store.Fan in minutes — sell digital products, courses, and calls straight from your bio. Free to start.
