ChatGPT Prompts for Email Subject Lines That Open

Ask ChatGPT for “10 catchy subject lines” and you’ll get ten lines that read fine on your desktop preview and die on a phone - truncated mid-thought, interchangeable with every other AI-written subject line in the inbox, and impossible to learn anything from because you shipped one and guessed.

My fix is building three things into the prompt itself: the device constraints the inbox will impose, the variant structure a properly sized A/B test needs, and the context that makes a line yours instead of anyone’s. Here’s the constraint set, then eight prompts by campaign type that carry it.

The constraint set (why these numbers)

Every prompt I wrote below embeds the same rules, based on what inboxes actually display:

• Mobile truncation is the binding limit. An iPhone in portrait shows about 41 characters of a subject line, while Android portrait can cut off at 27, per Omnisend’s testing across devices. If the value of your line lives after character 40, most readers never see it.
• 5–7 words is the working target (Omnisend’s recommendation) - short enough to survive Android, long enough to say something.
• Front-load the specific. Whatever earns the open - the number, the deadline, the name - goes in the first 27 characters, where every device shows it.
• Know your baseline before testing. The average true open rate across industries is 20.73% once Apple’s Mail Privacy Protection inflation is stripped out, per Brevo’s 2026 benchmark of 175,000+ senders - dashboard numbers showing 33–43% include Apple’s automated pre-fetch opens, not humans. Judge your tests against deflated reality, not the vanity number. Personalization is worth testing early: Campaign Monitor reports (citing Oberlo’s data) that personalized subject lines generate on average 50% higher open rates - a claim worth validating on your own list rather than taking as law.

The base prompt

Every campaign-type prompt below extends this base. Set it up once per brand:

You are a lifecycle marketer writing email subject lines for
[brand - what it sells, to whom, in one sentence].
Audience: [segment + how warm they are].
Voice: [2–3 adjectives; one thing we never sound like].

Rules for every line you write:
- 5–7 words, under 41 characters
- the specific value lands in the first 27 characters
- no clickbait gap you can't pay off in the email body
- no ALL CAPS, no more than one punctuation mark, no emoji
  unless I say otherwise
- write lines as A/B PAIRS: same offer, two different angles,
  labeled with the angle used (curiosity / value / urgency /
  social proof / plain)

Wait for the campaign brief.

Why it works: the rules turn taste into constraints the model can actually follow, and demanding labeled pairs means every output is a ready-made test with a hypothesis - you learn which angle wins, not just which line. The character budget does silently what no “make it punchy” instruction ever does.

The campaign prompts

Paste one of these after the base prompt.

1. Newsletter

Campaign: regular newsletter, edition topic: [topic + the single
most interesting thing in it].
Write 5 A/B pairs. At least 2 pairs must lead with the specific
detail, not the newsletter's name. Plain-angle lines should read
like a note from a colleague, not a broadcast.

Why: newsletters die of sameness - “[Brand] Weekly #47” trains people to archive. Forcing the most interesting thing into the line makes each edition compete on content, and the plain angle is often the sleeper winner for warm lists. (Finding that most interesting thing is what my newsletter prompts are built around - the edition planner there feeds this prompt directly.)

2. Promotion / offer

Campaign: [offer, exact terms, real deadline].
Write 5 A/B pairs. Pair angles must include: value-led (the offer,
stated plainly), urgency-led (the deadline, no fake scarcity), and
benefit-led (what the product does, offer second).
Banned words: unmissable, exclusive, hurry, last chance.

Why: promos tempt models into carnival-barker mode. The ban list and “no fake scarcity” rule keep the urgency honest - which matters beyond ethics, because a list burned by fake deadlines stops believing real ones.

3. Welcome email

Campaign: first email after signup for [what they signed up for].
The email delivers: [the thing + one next step].
Write 4 A/B pairs. Tone: confirmation first, personality second.
One pair should set expectations ("what you'll get from us").

Why: welcome emails are opened on recognition, not persuasion - the subscriber is looking for proof the signup worked. Clever lines here lower opens. Clarity raises them. This prompt deliberately reins the model in.

4. Abandoned cart

Campaign: cart abandoned [timeframe] ago containing [product type].
Incentive available: [yes/what, or none].
Write 4 A/B pairs. One pair references the product plainly, one
asks a low-pressure question. If no incentive exists, do not imply
one. Nothing guilt-based.

Why: the “do not imply an incentive” rule exists because models invent discounts under pressure to be persuasive - and a subject line promising savings the email doesn’t contain is a spam complaint generator.

5. Re-engagement

Campaign: subscribers inactive for [period]. We will remove
non-responders after this sequence (true).
Write 4 A/B pairs: one honest-goodbye angle ("we're about to stop
emailing you"), one what-you-missed angle with a specific:
[detail], one plain check-in. No "we miss you" clichés.

Why: honest-goodbye lines routinely outperform sentimental ones because the stakes are real and stated. The specificity requirement stops the what-you-missed angle from collapsing into vague nostalgia.

6. Cold B2B outreach

Campaign: first-touch email to [role] at [company type] about
[the specific problem you solve for them].
Write 5 A/B pairs. Lowercase-first-letter variants allowed. Angles
must include: their-problem-stated-plainly, a genuine question
about their process, and a reference to [specific trigger:
their company event / role change / public statement].
Banned: "quick question", "touching base", any flattery.

Why: cold email lives or dies on looking individually written. The trigger-reference angle is the only one that can’t be mass-produced - which is exactly why it works - and the ban list removes the three most pattern-matched openers in B2B inboxes.

7. Event / webinar

Campaign: [event, date, the one thing attendees will be able to
do afterward].
Write 4 A/B pairs across the sequence: announcement, reminder
(date in first 27 chars), last-call (honest deadline).
The capability promise, not the event name, carries the line.

Why: “Join our webinar” sells the format, which nobody wants. The outcome is the product. Splitting the sequence into stages stops you reusing one announcement line three times with “REMINDER:” bolted on.

8. The rewrite prompt

Here are our last 10 subject lines with open rates and list
context: [paste, with baseline].
Diagnose the pattern: which angles won, which lost, what we
never try. Then write 5 A/B pairs that exploit the winning
pattern and test one angle we've never used.

Why: this is the compounding prompt - it turns your send history into strategy. Models are strong pattern-readers when you give them real performance data, and “one angle we’ve never used” keeps the testing program from converging on a local maximum.

What a testing log looks like

The compounding value of angle-labeled pairs shows up once you keep score. The minimal log I recommend - illustrative numbers, real structure:

Campaign	Angle A	Angle B	Winner	Margin
Sept newsletter	curiosity	plain	plain	clear
Fall promo	value	urgency	value	narrow
Oct newsletter	plain	social proof	plain	clear
Re-engagement	honest-goodbye	what-you-missed	honest-goodbye	clear

Four tests in, a hypothetical log like this is already saying something: this list opens for clarity, not cleverness. That sentence - your list’s fingerprint - is worth more than any individual winning line, because it travels: it informs the next campaign’s pair selection, gets pasted into prompt 8 as context, and eventually earns a place in the base prompt itself (“this audience historically prefers plain, specific lines - weight the mix accordingly”). Keep the log embarrassingly simple. A spreadsheet nobody updates is worth less than a four-column table someone does.

One discipline keeps the log honest: record the margin as clear or narrow rather than raw percentages, unless your list is large enough that the percentages mean something. Small-list tests produce vivid-looking differences that are pure noise, and a log full of noise teaches confident nonsense.

Testing what comes out

Two rules I’d hold to make the pairs worth the effort. Test one variable: the pairs are built around angle, so keep sender name and send time fixed. And mind your sample - on small lists, a few dozen opens’ difference is noise, so let tests run to confidence and prefer fewer, bigger tests. (Where subject-line testing fits into a wider AI workflow is covered in the complete guide to AI marketing.)

The line the subject line can’t cross

One constraint sits above all the prompt rules: the subject line is a promise the email body has to keep. Open rate is the easiest metric to inflate dishonestly. A curiosity-gap line the body doesn’t pay off buys you one open today and a reflexive archive next week, and enough of those teach both your subscribers and their spam filters to stop trusting the sender name. That’s why the base prompt bans clickbait gaps and why the honest-deadline and no-implied-incentive rules appear in the campaign prompts: they protect the list’s learned expectation that your emails are worth opening, which is the only asset here that compounds. When a test wins opens but smells like a bait-and-switch, the log entry that matters is “don’t do this again,” whatever the percentage says.

FAQ

Should subject lines use the recipient’s name? Test it - that’s what the data supports, not a blanket rule. Name-insertion is the weakest form of personalization anyway. Segment-specific relevance (prompt 6’s trigger angle) is the strong form.

Do emoji help or hurt? Category-dependent and audience-dependent. They’re banned in the base prompt so their use is a deliberate test, not a model habit. If you test them, treat the emoji as the single variable.

What about preview text? It’s the second line of your subject line - roughly 35–90 characters depending on client, and it should extend the subject’s promise, not repeat it. Add also write matching preview text, max 80 characters, that adds new information to any prompt above.

Also in this library: ChatGPT prompts for blog posts, ad copy prompts for Google, Meta and LinkedIn and social media caption prompts, all of which sit on top of a brand voice system prompt. The body these lines open for has its own newsletter prompts, and for where subject line testing fits the whole email program, see AI for email marketing managers.