Email marketing prompts. Twelve ready-made requests, to fill in with your own details.
A vague prompt gives you a generic email, full of adjectives and promises nobody made. Here are twelve prompts for the emails you write every week, and the rules to check before you hit send.
A good prompt says who is writing, to whom, what the email must achieve and what it must not make up. The official guides from Google and Anthropic stress the same points: clear and specific instructions, context, limits on length and tone, examples.
A marketing email is still a marketing email, even when AI writes it. Article 130 of the Italian Privacy Code requires the recipient’s consent, with an exception for existing customers, and the Italian Data Protection Authority (Garante) explained it in its guidelines of 4 July 2013.
Below are the twelve prompts in Word and PDF, in Italian and English, with a completed example, to download without leaving an email address. Each prompt also says what to check in the draft before sending it.
Written by: Mattia Esposito, AI and marketing consultant, founder of Itria AI. From Puglia, based in Turin since 2023, he works remotely in Italy and abroad. LinkedIn profile.
The twelve prompts, to download
A two-page file: the context to put at the top of every request and twelve prompts for everyday emails, from the first reply to an enquiry to asking for a review. The parts to replace are in square brackets, and under each prompt is the line to check before sending.

| File | Contents | Link |
|---|---|---|
| ItalianWord, to edit | The basic context and the twelve prompts, each with its check line. | |
| EnglishWord, to edit | The same prompts in English, for writing to customers who don’t speak Italian. | |
| ItalianPDF | The copy to keep open next to the assistant. | |
| EnglishPDF | The same copy, in English. | |
| ExamplePDF, filled in, in Italian | The follow-up prompt filled in for a fictitious window company, with a draft and the four checks made before sending. |
The basic context tells the assistant not to make up prices, dates, names or figures, and to write “TO COMPLETE” where information is missing. That line is what stops the email offering a discount nobody agreed.
What makes a prompt work
A prompt works when someone new could read it and do the job. Anthropic turns this into a test in its prompting guide: “Show your prompt to a colleague with minimal context on the task and ask them to follow it”. If the colleague would have doubts, so will the model.
Google’s guide for Gemini breaks a prompt into parts: clear and specific instructions, the text to work on, the context, constraints on what to do and what not to do, the response format, and examples showing the right result. For long texts it advises putting the material first and the question at the end.
| Part of the prompt | What to write | In the file |
|---|---|---|
| Who is writing | The company, what it sells, to whom. | The basic context, at the top of every prompt. |
| To whom | The type of recipient and what they already know. | A customer who asked, a lapsed customer, the newsletter list. |
| The task | What the email must achieve, in one sentence. | A reply, a confirmation, a yes or a no. |
| The real material | The email received, the quote, this month’s facts. | The square brackets to fill in. |
| The constraints | Maximum length, tone, what not to promise. | From 50 to 250 words depending on the email. |
| What not to make up | Prices, dates, names, figures. | “TO COMPLETE” in place of missing data. |
| The format | Subject, body, signature; or a list. | Ten subject lines, each with what it promises. |
| An example | One of your own emails that worked, to copy the tone. | Pasted after the context, when needed. |
The twelve prompts, one per row
Twelve prompts, one for each email a small business writes over and over. Each has a word limit, because a sales email read on a phone stops at the first few lines, and a single closing request, because two calls to action split the reply.
| Prompt | When to use it | What to paste |
|---|---|---|
| 1. Replying to an enquiry | An email arrives with questions on product or price. | The email, without signature or contact details. |
| 2. The quote email | The quote is ready and needs a covering email. | Product, quantity, price, lead time, validity. |
| 3. Follow-up after a week | The quote has had no reply. | The email sent and one new piece of information. |
| 4. Last message | The deal has stalled for weeks. | How long, and the direct contact. |
| 5. Lapsed customer | A customer hasn’t bought for months. | The last purchase and a genuine new item. |
| 6. Newsletter | The monthly send. | Three facts and the opt-out line. |
| 7. New product | A service or product is launching. | Who it’s for, price, date. |
| 8. Ten subject lines | The email is written, the subject line is missing. | The finished email. |
| 9. Shortening | The draft is too long. | The draft and what must stay. |
| 10. Changing the tone | The email sounds cold or too familiar. | The draft and the tone you want. |
| 11. Complaint | A customer writes in angry. | The complaint without personal data, and the verified facts. |
| 12. Review | A customer is happy. | The product and the link. |
Before you hit send: consent and Article 130
Sending marketing emails requires the recipient’s consent: Article 130(1) and (2) of the Italian Privacy Code, Legislative Decree 196 of 2003, says so. The Italian Data Protection Authority calls it the prior consent, or opt-in, requirement in its guidelines on promotional activity and spam of 4 July 2013.
There is an exception for customers, which the Authority calls “soft spam”. Paragraph 4 allows you to use, without asking for consent, the address a customer gave during a sale, for similar products or services, provided the customer was informed and does not object. Every send must say they can opt out, free of charge and easily.
Paragraph 5 forbids hiding the sender and sending marketing emails without a contact point for exercising data rights. Article 21 of the GDPR adds that anyone who objects to direct marketing must no longer receive it. A prompt changes none of these rules: it just makes them easier to forget, if the email goes out without being reread.
The bill, when the rules are broken, is high. Article 166 of the Privacy Code places breaches of Article 130 under the fines in Article 83(5) of the GDPR: up to €20 million or 4% of worldwide annual turnover, whichever is higher.
What not to paste into a prompt
Customers’ personal data the email doesn’t need: names, addresses, phone numbers, purchase history. A draft only needs the facts of the offer and the tone, and the name is added afterwards, in the software the email is sent from.
The reason is where the text ends up. ChatGPT’s plans page states that on the Free, Go, Plus and Pro plans content is used to train the models, with an option to opt out; on the Business plan business data is not used for training by default. How data rules and AI rules fit together is on the page about the AI Act and the GDPR.
What to check in the draft
Three things, every time: the facts, the promises and the recipient. A model can confidently write a price or a date that doesn’t exist, a behaviour known as AI hallucination: which is why the basic context tells it to leave a placeholder instead of inventing.
The completed example shows the check on a real follow-up email: the quote date and the free survey are true, but the sentence “the price holds” only applies until the offer expires and needs clarifying. The way of working in which the machine prepares and a person approves is called human in the loop.
When there are too many emails for a prompt
Marketing and sales are the first area where Italian businesses use AI: 33.1% of those that use it, according to Istat, Enterprises and ICT 2025. Among these businesses, 59.1% use generative AI, and overall 16.4% of businesses with at least ten employees use at least one AI technology, twice the previous year’s figure.
A prompt copied by hand works while the emails are few. When enquiries arrive every day and the data sits in your business software or CRM, a system can read the enquiry, fetch the right data, draft the reply and leave it for a person to approve. The difference between an assistant you ask and a system that works through a sequence of steps is on the page about AI agents and chatbots.
Questions and answers
How do you write a prompt for a marketing email?
With four things: who is writing and to whom, the exact task, the real material to use, and the constraints, meaning length, tone and what not to make up. These are the points in the official guides from Google and Anthropic: clear and specific instructions, context, constraints and examples.
The prompt gives you a draft. The email that goes out is reread by a person, who checks prices, dates and promises.
Can you use ChatGPT to write newsletters?
Yes, for the draft. The text must be reread before sending, and customer data doesn’t go into a personal plan: on ChatGPT’s Free, Go, Plus and Pro plans, content is used to train the models unless you opt out.
The rules on newsletters stay the same as ever: consent from the people who receive it and a way to opt out in every send.
Do you need consent to send marketing emails?
Yes. Article 130 of the Italian Privacy Code requires consent for marketing communications by email. The exception in paragraph 4 covers customers: an address given during a sale can be used for similar products or services, provided the customer was informed and does not object.
Every send must include a way to opt out, free of charge and easily.
What should you not paste into a prompt?
Customers’ personal data that the email doesn’t need: names, addresses, phone numbers, purchase history, especially in a service that uses content for training.
A draft only needs the facts of the offer and the tone: the customer’s name is added afterwards, in the software the email is sent from.
Which AI should you use for email marketing?
Any general-purpose assistant will do for drafting, on a plan that doesn’t use your data for training. How the prompt is written matters more.
When there are many emails and the data sits in your business software or CRM, you need a system connected to that data, which drafts the emails and leaves them for a person to approve.
Notes on sources
- Italian Legislative Decree no. 196 of 30 June 2003, the Personal Data Protection Code, Article 130, paragraphs 1, 2, 4 and 5, and Article 166, paragraph 2, on Normattiva: texts in force as of 8 November 2026.
- Italian Data Protection Authority, Guidelines on promotional activity and combating spam, 4 July 2013: sections 2.5 (prior consent), 2.6 (requirements for consent) and 2.7 (soft spam). The term in quotation marks is the Authority’s.
- Regulation (EU) 2016/679, the GDPR, Article 21(2) and (3), and Article 83(5), on EUR-Lex.
- Google, Prompt design strategies, Gemini API documentation; Anthropic, Prompting best practices, the source of the quoted sentence. Read on 8 November 2026.
- Istat, Imprese e ICT, 2025: businesses with at least ten employees; areas of AI use and technologies used.
- OpenAI, ChatGPT plans page, read on 8 November 2026. Providers’ terms change: reread them on the day you choose.
- The prompts and the example are written by Itria; the example uses a fictitious company and customer, and the draft shows how an output can look, always to be reread before sending.
A prompt writes one email. A system drafts them all, and a person approves them.
When enquiries arrive every day, an AI agent can read the email, recognise what the customer is asking, fetch the right data from your business software and draft the reply; a person rereads it and sends it. The piece that sorts incoming enquiries classified 96.0% of a hundred enquiries correctly, across four channels and five languages, with a median of 7.3 seconds from enquiry to ready draft: measured on a test bench on 4 September 2026, on a test system.