Best ChatGPT Prompts for Email Marketing (2026)

Subject lines are where most AI-assisted email work collapses. The model defaults to curiosity-gap clickbait because that's what the training data is full of. The fix is specificity: give the model your audience, the single thing you want them to feel, and constraints that prevent lazy patterns.

**Prompt — Batch subject line generation:** ``` You are a senior email copywriter. Generate 10 subject lines for an email to [AUDIENCE — e.g., "SaaS founders on a free trial"] about [TOPIC — e.g., "upgrading before their trial ends"]. Constraints: - Each subject line under 50 characters - No question marks - No "Here's how" or "Don't miss" openers - Mix of angles: urgency (2), social proof (2), curiosity (2), benefit-first (2), conversational (2) - Brand tone: [TONE — e.g., "direct, no hype, founder-to-founder"] For each subject line, add a one-sentence note on why it works and which segment it targets best. ```

**Why it works:** The angle constraint forces the model to produce genuinely different options rather than variations on a single approach. The character limit prevents the lazy long-form subject line that performs well in prompts but poorly in inboxes. The one-sentence annotation makes the batch useful as a briefing document, not just a pick-one list.

**Prompt — Single subject line with pre-header:** ``` Write a subject line + preview text pair for this email: Email purpose: [PURPOSE — e.g., "announce a 48-hour sale on annual plans"] Audience: [AUDIENCE — e.g., "monthly subscribers who have been active for 90+ days"] One feeling to create: [FEELING — e.g., "relief — they've been thinking about upgrading and this makes the decision easy"] Brand voice: [VOICE — e.g., "warm, honest, never salesy"] Format: Subject: [subject line, under 45 chars] Preview: [preview text, 85-100 chars, continues the thought from the subject — do not just summarize the subject] ```

GPT-5 and Claude Opus 4.8 both perform well here. Claude tends to avoid superlatives and hype language by default — useful for B2B. GPT-5 leans slightly more conversational, which works better for DTC and consumer newsletters.

A welcome sequence is the highest-leverage email series any business owns — it runs at 100% deliverability to the most engaged segment you'll ever have, the people who just subscribed. The common AI mistake here is asking for all five emails at once and getting inconsistent tone, repetitive copy, and no coherent narrative arc.

**Prompt — Welcome sequence planner (run first):** ``` Plan a 5-email welcome sequence for [BRAND/PRODUCT — e.g., "a B2B SaaS project management tool for agencies"]. Context: - New subscriber source: [SOURCE — e.g., "opted in via a free project template download"] - Subscriber goal: [GOAL — e.g., "save time on client project reporting"] - Our USP: [USP — e.g., "auto-generates client reports from Slack activity"] - Brand voice: [VOICE — e.g., "pragmatic, agency-fluent, zero fluff"] - Send cadence: Day 0, Day 1, Day 3, Day 7, Day 14 For each email, output: - Subject line (under 50 chars) - Preview text (85-100 chars) - Primary purpose (one sentence) - Core message (2-3 bullet points) - Primary CTA and URL - Emotional job this email does Do NOT write the full email copy yet — just the plan. ```

**Prompt — Individual welcome email (run after planner):** ``` Write Email [NUMBER] of a 5-part welcome sequence. Plan reference: [PASTE THE PLAN OUTPUT FOR THIS EMAIL] Brand voice: [VOICE] Email format: Plain text, conversational, no HTML flourishes Length: 150-200 words max (not counting subject/preview) CTA placement: One CTA only, in the final paragraph Do not: Use "I hope this email finds you well", "Just wanted to", "Feel free to", or "Don't hesitate to" ```

Running the planner first and the email writer second gives you a consistent arc across the sequence. Claude Opus 4.8 is particularly strong for welcome sequences — it maintains voice consistency across long multi-turn outputs better than any model currently available. See our Claude vs ChatGPT for newsletter writing breakdown for more detail on which model holds tone longer.

Abandoned cart emails have a 40-45% average open rate — far higher than standard promotional email — because the recipient already indicated intent. The copy job is not to sell; it's to remove the specific friction that caused the abandonment. Generic cart reminders leave money on the table.

**Prompt — Three-email abandoned cart sequence:** ``` Write a 3-email abandoned cart sequence for [PRODUCT — e.g., "a $197 online course on freelance pricing"]. Product details: - What it is: [DESCRIPTION] - Price: [PRICE] - Key promise: [PROMISE — e.g., "raise your freelance rates 30-50% in 90 days"] - Common objections for this price point: [OBJECTIONS — e.g., "not sure it will work for my niche, nervous about the investment"] Sequence structure: - Email 1 (1 hour after abandon): Gentle reminder, no urgency, lead with curiosity about what held them back - Email 2 (24 hours after abandon): Address the most common objection directly, include social proof (I'll use [SOCIAL PROOF — e.g., '3 testimonials from service businesses']) - Email 3 (48 hours after abandon): Final nudge with a specific low-risk offer or guarantee restatement For each email: Subject line, preview text, full body copy (under 200 words), CTA button text Tone: [TONE — e.g., "empathetic, no pressure, honest about what the course does and doesn't do"] ```

The structure above outperforms single-email reminders because it maps to the actual psychology of cart abandonment: the first email catches forgetters, the second email addresses doubters, and the third email converts fence-sitters. GPT-5 handles objection-handling copy very well, especially when you give it specific objection language rather than generic pain points.

Inactive subscribers hurt your deliverability and skew your open rate metrics. A re-engagement campaign either wins them back or gets them to self-unsubscribe — both outcomes improve your list health. The AI challenge here is writing copy that feels personal without being creepy, and urgent without being desperate.

**Prompt — Re-engagement sequence (3 emails):** ``` Write a 3-email re-engagement sequence for subscribers who have not opened in [TIMEFRAME — e.g., "90 days"]. Brand: [BRAND] What has changed since they last engaged: [CHANGES — e.g., "we launched a new free tier, doubled the template library, and cut the price of annual plans 20%"] Sunset plan: If they don't engage after Email 3, they will be removed from the list (mention this only in Email 3) Email 1 — Subject angle: acknowledge the silence without guilt-tripping. Body: short, warm, one reason to come back. Email 2 — Subject angle: show what they've missed (concrete, not vague). Body: 3 bullet points of real updates. CTA to explore. Email 3 — Subject angle: "Last chance to stay on the list" — direct, no manipulation. Body: honest note that you're removing inactive subscribers, give them a clear reason to stay, and a clear path to unsubscribe if they want to. Format each email: Subject, Preview, Body (100-150 words), CTA text Tone: [TONE — e.g., "honest, no pressure, treats subscriber as an adult"] ```

**Prompt — Single high-stakes re-engagement email:** ``` Write one re-engagement email for a subscriber who was previously a paying customer but cancelled [TIMEFRAME — e.g., "6 months ago"] and is now on a free list. Context: - Why they likely cancelled: [REASON — e.g., "price was the barrier; they mentioned it in the cancellation survey"] - What's new: [UPDATE — e.g., "we now have a Starter plan at $29/month that didn't exist when they left"] - Tone: direct acknowledgment that you know why they left, no pretending you don't - CTA: link to the new pricing page Length: under 180 words. No hollow compliments. Start with the most useful sentence, not a greeting. ```

Newsletter copy is the hardest email marketing task to delegate to AI because voice consistency matters more here than anywhere else. Readers subscribe to newsletters because they like a specific person's perspective — AI output that reads like a content mill destroys that trust fast.

The solution is a voice document prompt that you run once, then reference in every future newsletter prompt. This is the same technique professional ghostwriters use.

**Prompt — Build your voice document:** ``` I'm going to paste three newsletter issues I wrote. Your job is to extract a voice and style guide I can use to brief an AI writing assistant. Analyze these emails and output: 1. Sentence length pattern (short/medium/long, typical range) 2. Vocabulary level (grade level, technical density) 3. Structural signatures (how I open, transition, and close) 4. Things I never say (phrases or patterns absent from my writing) 5. Point of view and authority stance (how I position myself relative to the reader) 6. Three example sentences that are distinctly "my voice" — keep them verbatim [PASTE EMAIL 1] --- [PASTE EMAIL 2] --- [PASTE EMAIL 3] ```

**Prompt — Newsletter issue (use after you have your voice document):** ``` Write a newsletter issue about [TOPIC]. My voice guide: [PASTE VOICE DOCUMENT OUTPUT] Issue structure: - Opening hook (2-3 sentences, not a question, not a stat — an observation or a specific scene) - Main section: [MAIN CONTENT — e.g., "3 lessons from my experiment running GPT-5 against Claude Opus 4.8 for email A/B testing"] - Each lesson: one-sentence takeaway, 2-3 sentences of evidence or story, one actionable implication - Closing: one sentence that connects the issue back to the reader's situation, then a soft CTA to [CTA — e.g., "reply with their own test results"] Length: 400-550 words total Format: Plain prose, no bullet points in the main body, minimal headers ```

Claude Opus 4.8 outperforms GPT-5 on voice-consistent long-form email. The Anthropic model card credits improved instruction-following in Opus 4.8, which shows in multi-constraint tasks like this. See also our AI for newsletter writing deep-dive.

The most common A/B testing mistake is testing variations that are too similar — changing one word in a subject line and calling it a test. To get statistically meaningful results in reasonable time, your A and B variants need genuinely different approaches, not cosmetic differences.

**Prompt — Subject line A/B variants with strategic divergence:** ``` Generate A/B test variants for a subject line. These should be genuinely different strategic angles, not slight rewrites of the same idea. Email context: [CONTEXT — e.g., "announcing a free audit tool for email deliverability, sent to 10k B2B marketers"] Hypothesis to test: [HYPOTHESIS — e.g., "does leading with the free offer outperform leading with the pain of bad deliverability?"] Variant A: [ANGLE A — e.g., "lead with the offer"] Variant B: [ANGLE B — e.g., "lead with the problem"] For each variant produce: - 3 subject line options (pick the strongest to test) - Matching preview text - One sentence on why this angle should win based on this audience Also flag: what would make this test inconclusive (e.g., if audience size is too small, if the send time difference could confound results). ```

**Prompt — Full email A/B body copy:** ``` Write two versions of the same email body for an A/B test. Email goal: [GOAL — e.g., "get recipients to start a free trial"] Audience: [AUDIENCE] Version A approach: [APPROACH A — e.g., "lead with social proof — open with a customer quote, then explain the product"] Version B approach: [APPROACH B — e.g., "lead with the problem — open with a specific scenario that describes the reader's pain, then present the product as the fix"] Constraints for both versions: - Same word count: 160-180 words - Same CTA text and link - Same PS line (I'll write this myself) - No overlap in the first three sentences between versions Output both versions with their approach labeled, then a brief note on what the test will actually reveal. ```

For more on the cold email angle specifically, our 10 ChatGPT prompts for cold email reply rates post covers the testing frameworks that have the highest ROI for outbound.

Segmented emails outperform broadcast emails by a wide margin, but writing separate copy for each segment is time-consuming. AI is exceptionally good at this task: you write the core email once, then prompt the model to adapt it for each segment without losing the central message.

**Prompt — Segment adaptation (core-to-segment):** ``` I have a core email I want to adapt for three audience segments. Keep the central message and CTA identical across all versions. Only change: the opening hook (first 2 sentences), the primary example or use case referenced, and the tone calibration. Core email: [PASTE CORE EMAIL] Segment 1: [DESCRIPTION — e.g., "freelancers, solo operators, price-sensitive"] Segment 2: [DESCRIPTION — e.g., "small agency owners, 5-20 employees, time is the constraint"] Segment 3: [DESCRIPTION — e.g., "enterprise marketing managers, budget not the issue, worried about internal buy-in"] For each segment, rewrite only the opening hook and primary example. Show me a before/after for each segment so I can see what changed and why. ```

**Prompt — New subscriber vs. long-term subscriber variants:** ``` Adapt this promotional email for two subscriber segments. Original email: [PASTE EMAIL] Version for new subscribers (joined in last 30 days): - Reference that they're new and haven't fully explored the product yet - Softer sell — focus on value discovery, not just the offer - Acknowledge that this is early in their relationship with us Version for long-term subscribers (12+ months): - Reference their loyalty without being cheesy - Direct offer presentation — they already know the product - Make them feel like this offer is specifically for people who've stuck around Keep both versions under 175 words. Same CTA text. ```

Gemini 2.5 Pro handles high-volume segmentation tasks particularly well when you're adapting across many segments — its context window and batch processing capabilities mean you can run all segment variants in a single prompt rather than chaining calls.

Cold email is the highest-volume, lowest-quality AI use case in email marketing. The internet is drowning in AI-generated cold outreach that reads identically because everyone used the same generic prompt. The prompts below force specificity that makes your outreach stand out.

**Prompt — Research-first cold email:** ``` Write a cold outreach email using this research about the recipient. Recipient research: - Name: [NAME] - Company: [COMPANY] - Role: [ROLE] - Specific thing I noticed: [OBSERVATION — e.g., "they published a LinkedIn post last week about struggling to track which email campaigns drive actual pipeline, not just opens"] - Their likely goal: [GOAL — e.g., "attribute email revenue more accurately"] My offer: [OFFER — e.g., "DDH's email attribution dashboard — free trial, no credit card"] Connection to their specific observation: [CONNECTION — e.g., "our pipeline attribution feature solves exactly what they described — it tracks email clicks through to closed deals in CRM"] Email constraints: - Under 100 words - First sentence references their specific observation (not generic flattery) - One clear CTA: a 20-minute call or a trial link, not both - No: "I came across your profile", "I hope this finds you well", "I wanted to reach out" - Sign off with first name only ```

**Prompt — Cold email sequence (3-touch):** ``` Write a 3-touch cold email sequence. The goal is to get a reply, not make a sale. Prospect context: [CONTEXT — e.g., "Head of Email at a 50-person DTC brand, managing Klaviyo, responsible for $2M+ email revenue"] Offer: [OFFER] Primary value prop: [VALUE PROP] Email 1 (Day 0): Lead with a relevant observation or data point about their industry/role. End with a low-friction ask (watch a 2-min demo, not a full call). Email 2 (Day 4): Brief follow-up. Add one new piece of value (a case study, a specific result, a tool they'd find useful). Do not just repeat Email 1. Email 3 (Day 9): Permission-based close. Give them a graceful out. Acknowledge this is the last touch if they don't respond. Each email under 90 words. Subject lines that would make sense as replies in a thread. ```

For broader marketer-focused prompts, our best ChatGPT prompts for marketers 2026 covers the full marketing workflow beyond email.

Promotional emails are the workhorses of ecommerce and SaaS marketing. They have a clear job: communicate an offer clearly and get clicks. The AI failure mode here is burying the offer or adding so much context that the reader never reaches the CTA. The prompts below enforce an offer-first structure.

**Prompt — Product launch email:** ``` Write a product launch announcement email. Product: [PRODUCT NAME] What it does: [ONE SENTENCE DESCRIPTION] Who it's for: [AUDIENCE] The specific problem it solves: [PROBLEM] Price and availability: [PRICE + LAUNCH DATE] Launch offer (if any): [OFFER — e.g., "50% off for the first 48 hours for existing subscribers"] Email structure: - Subject line and preview text - Opening: state the problem in one sentence without mentioning the product - Paragraph 2: introduce the product as the solution - Paragraph 3: 3 bullet points of specific capabilities (not benefits — what it actually does) - Paragraph 4: the offer, deadline, and CTA - PS: a social proof signal (testimonial fragment, waitlist number, or early user result) Tone: [TONE — e.g., "confident but not hypey, speaks to the product's actual value"] Length: 200-250 words excluding subject/preview ```

**Prompt — Flash sale email:** ``` Write a flash sale email. Offer: [OFFER — e.g., "40% off all annual plans"] Deadline: [DEADLINE — e.g., "ends Sunday midnight PT"] Audience: [AUDIENCE — e.g., "monthly subscribers who have been active for 3+ months"] Email constraints: - Lead with the offer in the subject line — no teasing - First sentence of body restates the offer with the deadline - No more than 3 paragraphs before the first CTA button - After the CTA: one paragraph addressing the most likely objection to upgrading ([OBJECTION — e.g., "worried about the annual commitment"]), then a second CTA - No countdown timer copy ("hurry, only X hours left") — just the specific deadline - Length: 150-180 words ```

**Prompt — Post-purchase upsell email:** ``` Write a post-purchase email that introduces an upsell offer. Context: The recipient just purchased [PRODUCT — e.g., "the Email Audit Starter Pack"] Upsell offer: [OFFER — e.g., "Email Mastery Pro at $97, 40% off for the next 72 hours"] Connection between purchase and upsell: [CONNECTION — e.g., "they bought the audit tool — the Pro tier gives them the fixes, not just the diagnosis"] Email timing: 1 hour after purchase confirmation Structure: Acknowledge the purchase warmly (1 sentence), connect their new purchase to the natural next step, present the upsell as the logical continuation (not a new sale), CTA Length: 120-150 words Tone: helpful, not pushy — the framing is completing the loop they already started ```

Not every email task deserves the same model. In 2026, the cost difference between GPT-5.5 and GPT-5 nano is roughly 150x per token — using the wrong model is as wasteful as hiring a senior copywriter to write envelope copy.

**GPT-5 (OpenAI)** — best for: subject line generation, A/B variants, cold outreach, promotional copy. GPT-5's training on massive commercial email data means it has strong intuitions about conversion copy and urgency language. It defaults to slightly punchy, direct copy. Use it when you need high-velocity output and direct response mechanics. OpenAI model documentation.

**GPT-5.5 (OpenAI)** — best for: longer sequences, re-engagement campaigns, any email where tone nuance matters more than CTA optimization. GPT-5.5's improved reasoning shows in complex multi-email sequences where context from earlier emails needs to carry forward.

**Claude Opus 4.8 (Anthropic)** — best for: newsletter copy, welcome sequences, voice-consistent long-form email, brand-sensitive communication. Claude's instruction-following at the Opus tier is the strongest available — it respects the "never say X" and "always structure as Y" constraints better than any other model. This matters most in high-constraint copywriting tasks. Anthropic model overview.

**Gemini 2.5 Pro (Google)** — best for: segmentation-heavy campaigns where you need to adapt one email across many personas simultaneously, data-grounded email copy where you're incorporating analytics or research, and international campaigns where the model's multilingual strength adds value. Gemini model documentation.

For a direct head-to-head evaluation of ChatGPT and Claude specifically for sales and email copy, see ChatGPT vs Claude for sales emails 2026.

The prompt patterns above work because they apply a few principles consistently. Understanding the principles lets you modify the prompts for your context rather than treating them as fixed templates.

**Specificity beats brevity.** A 200-word prompt that includes your audience, tone, constraints, and structural requirements will outperform a 10-word prompt every time. The model's job is to fill in gaps — every gap you leave is a place where it defaults to the statistical average of all email copy in its training data.

**Constraints prevent defaults.** The "do not use" lists in the prompts above ("no 'I hope this finds you well'", "no question marks in subject lines") are not arbitrary. They block the model's most common fallback patterns. The model knows these patterns work on average — the constraints force it off the average and into territory specific to your brand.

**Structure the output, not just the input.** Every prompt above specifies the output format explicitly: word counts, section order, what to include in each paragraph. This matters because AI models optimize for completing the output format you gave them — if you ask for "an email," you get what looks like an email. If you ask for "subject line (under 45 chars) / preview text (85-100 chars) / body (150-175 words) / CTA text," you get a production-ready component for your ESP.

**Run model comparisons on your own copy.** The model rankings in this guide are based on DDH internal testing — your industry, voice, and audience may produce different results. The fastest way to find your best model is to run the same prompt through GPT-5, Claude Opus 4.8, and Gemini 2.5 Pro and read the outputs side-by-side. For the prompt engineering principles that make comparisons meaningful, see how to write better prompts: 15 rules.

And if you want to track what each model output costs before you commit to a workflow, use our AI Prompt Cost Calculator — paste your estimated monthly email generation volume and get the line-item cost across every model.

Here is how these prompts fit into an end-to-end workflow for a mid-size email operation generating 30-50 emails per month.

**Week 1 setup (one time):** Run the voice document prompt on your three strongest past emails. Save the output as your brand voice brief. This is the single most valuable prompt you'll run — every subsequent email prompt references it.

**Per campaign:** (1) Use the subject line batch prompt to generate 10 options; pick two for A/B testing. (2) Use the full email body prompt with your voice document attached. (3) Use the segmentation adaptation prompt to create audience-specific variants from the core email. (4) For automated flows (welcome, abandoned cart, re-engagement), run the sequence planner first, then the individual email prompt for each email — this prevents the tone drift that happens when you write sequences email-by-email without a plan.

**QA checklist before sending:** Read the output aloud — AI copy often scans fine but sounds odd spoken. Check for the specific phrases you put in your "never say" list (models occasionally ignore constraints on longer outputs). Verify the CTA link is correct (AI cannot know your URLs — it will hallucinate plausible-looking paths if you don't specify). Run the subject line through your ESP's spam score checker — AI subjects occasionally trigger spam filters with urgency language.

The best ChatGPT prompts for copywriters post extends these patterns into landing page, ad, and social copy if you're looking to use the same prompt workflow across your full content operation.