An AI subject line generator produces multiple subject line options from a description of your email. The useful ones are trained on or fed your own campaign history, so the suggestions reflect what your audience has actually opened rather than what performs well in general.
The value is not that the machine writes a better line than you. It usually does not. The value is that writing twelve subject lines by hand is tedious enough that most people ship their second idea, and the eighth idea is often the good one.
Give it enough to work with
A generator handed "newsletter about our new feature" returns twelve variations of "Introducing our new feature". Garbage in, generic out. What changes the output:
The actual news. Not "a new feature" — "campaigns can now be scheduled per subscriber time zone". Specifics are the raw material; without them the model has nothing to be specific about.
Who is receiving it. "Customers who have sent at least one campaign" produces different lines than "people who signed up and never sent anything", because the second group needs a reason to care that the first group already has.
Your winners. Paste the five best-performing subject lines from your last six months. This is the single highest-leverage input, because it communicates register, length and formality more precisely than any adjective. A generator built into your platform can do this automatically — ours scores suggestions against your own campaign history rather than a generic corpus.
What you will not say. If your brand does not use exclamation marks, emoji or false urgency, say so. Models default to all three.
The patterns to reject on sight
Generated subject lines cluster around a handful of shapes. Some are fine; several are tells.
- Colon constructions. "Email deliverability: what you need to know." Grammatically neat, completely flat, and instantly recognisable as machine output.
- Curiosity with no content. "You won't believe what we found." This gets opened once and resented afterwards, and the resentment shows up as spam complaints.
- Manufactured urgency. "Last chance!" when it is not. Recipients calibrate fast, and a deadline that is not real trains people to ignore your real ones.
- The rhetorical question. "Struggling with your open rates?" Everyone has read ten thousand of these.
- Everything capitalised in Title Case. Sentence case reads as a person; title case reads as a press release.
What survives editing is usually the shortest option, or the one with a concrete number or noun in it.
Length, and the truth about it
Mobile clients show roughly 30–40 characters of subject line, desktop shows 60 or more, and the useful rule is that the first four or five words carry everything. Whether the total is 35 or 65 characters matters far less than whether the front of the line is loaded.
Longer subject lines are not inherently worse; front-loaded ones are better. Ask your generator for lines where the most important word appears first, which is an instruction models follow well.
And write the preview text as a continuation, not a repeat. Many clients display the subject and preview together, so a preview that restates the subject wastes the most valuable line of text in the inbox. See email preview text.
Test them properly or do not test them
Generating twelve options is only useful if you can tell which one worked, and most subject line "tests" cannot.
The problems: open rate is now a distorted metric because of Apple Mail Privacy Protection proxy loading; a difference of two percentage points on a 2,000-person split is usually noise; and running a test on one campaign tells you about that campaign rather than about your audience.
The workable approach is to pick two genuinely different options — not two phrasings of the same idea — split at least a few thousand recipients per arm, and judge on clicks rather than opens where you can. Then log the result and look for patterns over ten campaigns rather than reacting to one. Email A/B testing and email test sample size cover the arithmetic.
You can also sanity-check a line before sending with our subject line tester, which flags length problems and spam trigger words without needing a campaign.
A workflow that takes four minutes
- Write one subject line yourself, before opening the generator. It anchors you to what the email is actually about.
- Generate twelve with full context supplied.
- Delete anything matching the patterns above — usually eight of them.
- Pick the two most different survivors.
- Test, log, send.
The generator is doing the tedious part. The judgement stays with you, which is the correct division of labour for every AI feature in email. There is more on that split in AI email writer.
Frequently asked questions
Do AI-generated subject lines get higher open rates?
Sometimes, but not because the model writes better lines. The lift comes from having twelve options to choose from instead of two, which raises the quality of the one you send. Studies claiming large automatic improvements are usually comparing against unoptimised baselines.
How long should an email subject line be?
Front-load the first four or five words and stop worrying about the total. Mobile clients truncate around 30–40 characters, so what matters is that the meaning survives truncation, not that the line is short.
Should I use emoji in subject lines?
They can lift opens for consumer audiences and are a liability in B2B, where corporate clients render them inconsistently. Test with your own list rather than following a general rule, and never use one that repeats what the words already said.
Can AI test subject lines for me?
It can predict, using your history, which of several options is likely to perform. Prediction is useful for narrowing twelve options to two; it does not replace an actual split test, because the model is guessing from patterns rather than measuring your audience today.
