AI Prompts for Ecommerce Email Campaigns: Copy-and-Adapt Templates

Most AI prompts for ecommerce email fail for one reason: they ask the model to write about a product it has never seen. "Write a sale email" produces sale-shaped sentences with nothing real inside them. The prompts below are built the other way round — you supply the facts, the model supplies the structure, and anything it does not know comes back as a placeholder instead of an invention.

What these prompts are actually for

A generic prompt gives the model no product data, no offer terms and no constraints, so it fills the gap with plausible-sounding filler: "our best-selling serum", "ships within 24 hours", "loved by thousands". None of that is your copy. It is the model's guess about what ecommerce copy usually says.

The templates in this article are structured to prevent that. Each one has an input block you fill with real facts, an instruction block that limits what the model may write, and an output format so you get something you can paste into an editor rather than a wall of prose.

You can use them in three places: a chat assistant like ChatGPT or Claude, a built-in AI writer inside your email platform, or a saved snippet your team reuses. If you would rather skip the prompt entirely, Mailerflow's AI email builder works the other direction — you describe the email and get a complete responsive design plus subject lines and preview text. Both approaches are fine; the prompts are for when you want control over the wording.

The rule that keeps AI copy factual

Put two hard constraints in every prompt you write, whatever the campaign:

1. Use only the facts supplied. If something needed for the copy is missing, output [PLACEHOLDER_NAME] rather than guessing.
2. Make no claims about results, speed, quality, popularity or savings unless the exact wording was supplied in the input.

The failure mode is easy to picture. Give a model a product name and a price and it will happily add "our best-selling", "award-winning" or "ships in 24 hours" because those phrases are common in the training data. If your warehouse takes three days, you have just published a promise you cannot keep.

Price handling deserves its own rule. State the price and currency explicitly ("£48.00 GBP"), tell the model not to calculate discounts unless you provide the arithmetic, and tell it never to round. A model asked to apply "20% off" to an unstated price will produce a number, and that number will be wrong.

Prompt 1: Product launch announcement

You are writing a product launch email for an ecommerce brand.

FACTS (use only these):
- Product name: [NAME]
- What it is, in one plain sentence: [DESCRIPTION]
- Who it is for: [AUDIENCE]
- Attributes: [3-5 FACTUAL ATTRIBUTES, e.g. material, size, weight, colour options]
- Price: [PRICE + CURRENCY]
- Available from: [DATE]
- One honest limitation or "what it is not": [LIMITATION]
- Single action: [CTA, e.g. "View the product page"]

RULES:
- Use only the facts above. If something needed is missing, output [PLACEHOLDER_NAME] instead of inventing it.
- No superlatives. Do not use "revolutionary", "game-changing", "best-selling", "award-winning".
- Do not invent specifications, shipping times, stock levels or reviews.
- Do not state or imply any discount.

OUTPUT:
1. Email body, 120-180 words, plain language, one clear CTA.
2. Three subject lines, each under 45 characters.
3. Two preview texts, each under 90 characters.

When to adapt: for a restock, replace the launch framing with availability ("back in stock from [DATE]") and drop the limitation line if it no longer applies. For a bundle, list the components as separate bullet points in the input so the model treats them as distinct items rather than merging them into one vague product.

Prompt 2: Sale and promotion email

You are writing a promotional email for an ecommerce brand.

OFFER FACTS (use only these):
- Offer: [e.g. 20% off, or £10 off orders over £60]
- Applies to: [exact scope]
- Exclusions: [list, or "none"]
- Starts: [DATE + TIMEZONE]  Ends: [DATE + TIMEZONE]
- Code or automatic: [CODE] / [AUTOMATIC AT CHECKOUT]
- Minimum spend: [AMOUNT + CURRENCY, or "none"]
- Deadline wording I am allowed to use: [e.g. "ends Sunday 23:59 GMT"]

RULES:
- Do not state or imply a savings amount. Do not calculate anything.
- Do not say "everything" if exclusions exist.
- Do not invent urgency, countdowns or stock warnings.
- Use only the facts above; output [PLACEHOLDER] for anything missing.

OUTPUT:
1. Email body that leads with the offer terms.
2. A short "what is included" list.
3. A plain closing line with the deadline.
4. Three subject lines under 45 characters and two preview texts under 90 characters.

When to adapt: for a tiered sale, give the model the tiers as a small table in the input and ask it to render them as a list, not as prose — prose is where tier conditions get muddled. For free shipping, state the threshold and currency explicitly and add "do not describe shipping as fast, free of delays, or guaranteed".

Prompt 3: Abandoned cart email

You are writing an abandoned cart email for an ecommerce brand.

CART FACTS (use only these):
- Items: [LIST EACH ITEM WITH PRICE AND CURRENCY]
- Discount offered: [YES + TERMS] / [NO]
- Shipping terms: [FACT]
- Returns policy: [FACT]
- Support contact: [EMAIL OR LINK]

RULES:
- Do not guess why the person left the cart.
- No guilt, no fake scarcity, no "we noticed you forgot".
- Include the item list verbatim.
- Under 120 words.
- Use only the facts above; output [PLACEHOLDER] for anything missing.

OUTPUT:
1. Version A: no discount.
2. Version B: with the discount, only if one was supplied.
3. Two subject lines under 45 characters for each version.

When to adapt: this is the prompt most worth turning into a recurring automation rather than a one-off send. Mailerflow's AI automation builder lets you describe the workflow in a sentence and it drafts the trigger, waits, conditions and emails for review before activation — useful when you want the cart email to fire on a delay rather than manually.

Prompt 4: Product spotlight for a single SKU

You are writing a single-product email for an ecommerce brand.

SKU FACTS (use only these):
- Product: [NAME]
- Category: [CATEGORY]
- Materials or ingredients: [LIST]
- Dimensions or sizing: [DETAILS]
- Care or usage instructions: [DETAILS]
- Price: [PRICE + CURRENCY]
- Top support questions about this item: [2-3 QUESTIONS]

RULES:
- Do not compare this product to any competitor product or to other brands.
- Do not make health, safety or efficacy claims.
- Use only the facts above; output [PLACEHOLDER] for anything missing.

OUTPUT:
1. Short intro (2-3 sentences).
2. "What you get" list.
3. "Good to know" list drawn only from the supplied facts.
4. One CTA.
5. Three subject lines under 45 characters.

When to adapt: for regulated categories — supplements, cosmetics, medical devices — add a line telling the model to avoid any health or efficacy claim and to flag anything it cannot verify from the input. Then read the output yourself before it goes anywhere near a send button.

Prompt 5: Win-back and re-engagement

You are writing a re-engagement email for an ecommerce brand.

FACTS (use only these):
- Time since last purchase or open: [FACT]
- What has changed since then: [NEW PRODUCTS, NEW SHIPPING TERMS, ETC.]
- Offer, if any: [TERMS] / [NONE]
- Single action: [CTA]

RULES:
- Acknowledge the gap without over-apologising.
- No "we miss you" clichés.
- Do not claim to know the reader's preferences unless supplied above.
- Use only the facts above; output [PLACEHOLDER] for anything missing.

OUTPUT:
1. Short version, under 80 words.
2. Longer version, 120-150 words.
3. Three subject lines under 45 characters.

When to adapt: send this to a segment, not the whole list. Mailerflow builds dynamic segments on fields, tags, engagement, events and website behaviour, and can generate a segment from a plain-language description — handy when "people who bought once and went quiet for 90 days" is easier to describe than to build.

How to personalize the output without breaking it

Models handle merge tags inconsistently. One run produces Hi [First Name], the next Hi {{first_name}}, the next Hi there. Standardise on your platform's exact syntax and add a line to the prompt: "use the merge tag {{first_name}} verbatim, and do not invent a fallback value."

There are three levels of personalization worth aiming for:

• Merge fields for name and last purchase. Simple, low risk.
• Segment-level variation — one body for first-time buyers, one for repeat buyers. Ask the model for both in a single run so the tone stays consistent.
• Behaviour-level — the email references what the contact actually viewed or read. Mailerflow supports AI personalization steps and website-behaviour data through its JS SDK, so a spotlight email can reflect a page a contact visited.

The warning that matters: do not over-personalize from thin data. If all you know is a first name, do not write copy that implies you know anything else. It reads as surveillance, not service.

Subject lines and preview text to pair with each prompt

The pairing rule: preview text should extend the subject line, not repeat it. Both need to survive being read on a lock screen, so front-load the concrete detail — product, price, date — and leave the adjectives out.

Mailerflow's AI email builder returns subject lines and preview text alongside the design, so you can generate and compare options in one step. If you would rather test than guess, A/B testing of campaigns is available.

Mistakes to avoid

• Pasting real customer data into a chat tool. Use placeholders and merge tags; keep personal data in your platform.
• Letting the model do arithmetic. Supply the final figures and forbid calculation.
• Accepting invented specs or shipping times. Read every number against your own product data.
• Sending without a human read. Tone and legal claims are your responsibility, not the model's.
• Reusing one prompt for every campaign type. Vary the structural instruction, not just the product name, or every email will sound the same.
• Forgetting to check merge tags render. Send a test to yourself before every new template.
• Treating generated copy as final. It is a first draft that saves you the blank page, nothing more.

Keep your vetted prompts in a shared doc or your repo so the team reuses them instead of rewriting from scratch each time. A prompt that has been checked once is worth more than five written in a hurry.

FAQ

Can I use these prompts with any AI tool? Yes — they are plain text. Built-in writers inside email platforms may need the instruction block trimmed, since they often already handle formatting and structure.

Will the AI get my prices wrong? Only if you let it calculate. Supply exact figures with currency, forbid arithmetic, and tell it never to round.

Should I use AI for transactional emails like order confirmations? Keep those template-driven with variables. AI is better suited to marketing copy where structure and phrasing vary; order confirmations need to be identical every time.

How do I stop every email sounding the same? Change the input facts and the structural instruction, not just the product name. A launch prompt and a win-back prompt produce different rhythms because the constraints differ.

Describe your next email or automation in one sentence and let Mailerflow's AI build it — free plan, no credit card.