TL;DR
Use AI to generate options, not to decide. Feed it your segment, your one offer, your single CTA verb, and your last few winning subject lines. Ask for 5 subjects (each under 40 characters of front-loaded text), matching preview lines, and a 120-200 word mobile body with one button. Then judge against your own click history, because open rate is no longer a trustworthy signal. For one-off campaign drafts, Claude Sonnet 4.6 or ChatGPT (GPT-5.5) on the free tier is enough; for launches and win-backs where brand voice matters, draft inside Claude with a Project that holds your voice and past winners.
The task
You are sending a campaign email and want a higher click rate than last week’s. You need 5 subject lines worth A/B testing, a body short enough to read on a phone in line at a coffee shop, and a CTA the reader can act on without thinking. The real constraint is the inbox preview pane: on an iPhone 16 in portrait, only about 41 characters of the subject show before truncation (the iPhone 16 Pro Max shows around 46), and Gmail’s mobile app cuts even tighter, near 37 characters. So the first 35-40 characters of your subject plus the preview text have to earn the click on their own.
Why click rate now matters more than open rate
Since Apple Mail Privacy Protection (MPP) pre-loads tracking pixels, opens are inflated and unreliable. Litmus measured Apple clients at roughly 49% of all email opens, and industry case studies show campaigns jumping from a ~28% to a ~52% “open rate” with no change in clicks or conversions. MPP does not touch clicks, so as of June 2026 the metrics that actually mean something are click rate, click-to-open, and downstream conversions.
Here is what “good” looks like in the 2026 benchmark data so you can calibrate:
| Metric (2026 averages) | Typical range | Source benchmark |
|---|---|---|
| Open rate (all industries) | ~20.7% (33.9% incl. Apple MPP) | Brevo 2026 |
| Click rate (campaigns) | ~2.1% | Brevo 2026 |
| Click-to-open rate | ~6.8% | Brevo 2026 |
| Automated flow click rate | ~5.6% (vs ~1.7% for one-off campaigns) | Klaviyo 2026 |
The headline takeaway: automated flows (welcome, browse-abandon, post-purchase) out-click one-off blasts by roughly 3x, and drive far more orders. If you can move a message from a manual campaign into a triggered flow, do that before you optimize the copy.
When AI helps, and when it does not
AI is very good at subject-line ideation and tone tuning. It is poor at knowing what your specific list responds to, which only your last 10 campaigns’ click data can tell you. Use AI to generate; use your own click history to choose. AI should never invent an offer, deadline, or scarcity claim (“only 3 left”) that you cannot back up, and it should never write your unsubscribe footer, which is regulated legal text under CAN-SPAM and similar laws.
Pick the right tool for the job
You do not need a specialist email tool to write good email. The general assistants are stronger at brand voice and iteration. As of June 2026:
| Tool | Best for | Cost (June 2026) |
|---|---|---|
| Claude (Sonnet 4.6) | Drafts that go out as-is; least editing; holds brand voice | Free tier, or Pro $20/mo |
| Claude (Opus 4.7) | High-stakes launches, investor updates, win-backs | Claude Pro $20/mo or Max $100/mo |
| ChatGPT (GPT-5.5) | Fast bulk ideation: 10 subjects, 3 body variants, 4 segments at once | Free $0 (US Free shows ads), Plus $20/mo |
| Gemini 3.1 Pro | Drafting inside Google Workspace / Gmail | Google AI Pro $19.99/mo |
A practical split many teams use: ChatGPT for fast ideation and variant counts, Claude for the draft that actually ships. If you write recurring campaigns, set up a Claude Project (or a ChatGPT custom GPT) seeded with your brand-voice sentence, your three best past emails, and your segment notes. Every draft then inherits those patterns instead of you re-pasting them.
What to feed the AI
- Audience and last interaction (signed up last week / dormant 90 days / repeat buyer)
- The one offer or news that is actually new for this reader
- Primary CTA as a single specific verb: Shop, Read, Reserve, Reply
- Brand voice in one sentence (“warm, direct, no exclamation marks”)
- Any compliance constraints (financial, health, or regulated-industry language to avoid)
- Your last 3 subject lines that earned clicks, and 3 that did not, so the model learns your audience’s pattern
Copy-ready prompt
Write a marketing email.
Audience and last interaction: [segment, recency]
Offer / news: [one line]
Primary CTA verb: [Shop / Read / Reserve / Reply]
Brand voice: [one sentence]
Compliance constraints: [language to avoid]
Subject lines that earned clicks: [last 3]
Subject lines that did not: [last 3]
Return:
1. Five subject line candidates. Put the key message in the first 35 characters
so it survives iPhone/Gmail truncation. Mix: curiosity, news, benefit,
question, low-key. Mark which is closest to my past winners.
2. Preview text (35-90 chars) for each subject. It must EXTEND the subject,
never repeat it.
3. A 120-200 word body. Mobile-first: short paragraphs, one idea each, scannable.
4. Button text (2-3 words) plus a one-line fallback text link.
5. A "low click rate" alternate subject + preview pair to resend to non-clickers.
Do not invent scarcity, deadlines, or testimonials. If you would write
"limited time," write [VERIFY: real deadline?] instead.For a transactional-adjacent email (shipping update with a soft upsell), add: “The email must read as a shipping update first; the promo is secondary and clearly labelled.”
Recommended output structure
Five subject-plus-preview pairs, then a body with one idea per paragraph, then a single button. No stacked CTAs. Treat preview text as line two of the subject, not as a tagline, and never leave “View this email in your browser” sitting in that slot, which throws away 40-90 characters of conversion surface.
How to check the output is usable
- Read the subject and preview together as one line. Does it earn the click without the body?
- Confirm the key promise lands inside the first ~35 characters, before mobile truncation.
- The body has one CTA. If there are two, cut the weaker one.
- It reads like one human writing to one human, not a brand addressing a list.
- No invented scarcity, fake testimonials, or unverifiable claims.
- Preview it on a real phone (or Litmus for cross-client rendering). Anything cut off in the iOS preview is a problem.
Common mistakes
- Clickbait subject with a thin body: opens once, unsubscribes after two.
- Stacking 3 CTAs because “people want choice.” They want one obvious action.
- Letting AI write fake urgency (“ends tonight!”) that contradicts real merchandising.
- Skipping preview text and burning 40-90 characters of conversion space.
- A/B testing too small a list. For subject tests, aim for roughly 250+ clicks (not opens) per arm at a 95% confidence threshold before you trust the winner; with small lists, subject tests are still your fastest signal because they move the largest number.
- Judging the test on open rate. After Apple MPP, decide subject-line tests on clicks, not opens.
FAQ
Subject A/B test or content A/B test? Subject first: bigger swings, faster signal. Test the body only after the subject is settled. Decide the winner on clicks, since opens are inflated by Apple MPP.
How long should the body be? As long as it earns. Newsletters can run 400+ words; promo emails convert best at roughly 120-200, because most reads happen on a phone.
Which AI should I actually use? For everyday campaign drafts, Claude Sonnet 4.6 or ChatGPT (GPT-5.5) on free tiers is plenty. For launches and win-backs, draft in Claude (Opus 4.7) inside a Project that holds your brand voice and best past emails so the tone stays consistent.
Should AI write the unsubscribe footer? No. It is regulated legal text in most jurisdictions (CAN-SPAM, CASL, GDPR). Paste your approved version verbatim.
My open rate jumped but sales did not. What happened? Almost certainly Apple MPP inflating opens. Ignore the open spike and watch clicks, click-to-open, and conversions instead.
Related
- Email marketing prompts — additional phrasings and segment-specific variants
- Email subject line prompts — dedicated subject-line generator
- Abandoned cart email prompts — when the campaign is a recovery sequence
- AI newsletter tutorial — full newsletter workflow
- AI email triage tutorial — managing incoming, not outgoing, email