ChatGPT for writing B2B emails: method, prompts and pitfalls

In brief: ChatGPT writes fast, but rarely gets it right first time. Everything hinges on the prompt and on the proofreading. Here is the method for turning AI into a reliable B2B writing assistant, the prompts that get replies, and the deliverability and GDPR reflexes never to neglect before sending.

Why most emails that come out of ChatGPT fall flat

The tool is everywhere. According to Salesforce’s State of Marketing 2024, 75% of marketers have already adopted AI, and in France the Infopro Digital / BtoB Leaders barometer published in early 2026 indicates that 85% of B2B decision-makers use it or plan to. Yet open a dozen prospecting emails received this week. Many smell of text generated in two clicks. Smooth phrasing, interchangeable hooks, that famous “I hope this message finds you well”.

The problem is not ChatGPT. It is the way it is used.

A B2B email that gets a reply rests on a precise context: the recipient’s situation, a recent signal, a credible reason to contact them now. When you ask the AI “write me a prospecting email to sell my software”, it produces exactly what it was asked for, a generic message for a generic recipient. And generic, in prospecting, ends up in the bin.

HubSpot measures it in its State of Marketing 2026: 52% of marketers believe that AI has made content so easy to produce that it has become less effective overall. A significant share of consumers also say they spot and ignore visibly generated content. Ease has a downside. When everyone has the same tool, it is the quality of the brief and the human proofreading that make the difference.

The prompt is where everything is decided

OpenAI repeats it in its official prompt engineering documentation: the more precise and explicit the instructions, the better the result. A good B2B email prompt comes down to five components.

Compare the two requests. “Write an email to sell my expense report management solution” gives a flat text. “Write a 90-word email to the CFO of an industrial SME that has just recruited a management controller. Objective: secure a 20-minute conversation on expense report automation. Direct tone, no superlatives, a single question to close” gives a usable message. Same tool, incomparable result.

It is the logic of the detailed brief that applies here. You are not delegating the thinking, you are passing it on. ChatGPT executes, it does not guess your strategy. To place this mechanism within a coherent B2B email marketing strategy, AI remains a fast executor in the service of an intention that stays with you, and you alone.

Operational prompts by use case

Here are four common situations in B2B prospecting and the prompts that cover them. Adapt the variables in square brackets to your reality.

The subject line that decides the open

The subject line conditions everything else. A perfect email with a dull subject line will never be read. ChatGPT excels at producing variants, provided you frame the angle.

Generate 8 B2B email subject lines for a [HR director of a mid-sized company] whom I am contacting about [digitising onboarding]. Four angles: curiosity, quantified benefit, direct question, reference to a recent fact in the sector. Maximum 50 characters. No exclamation marks, no anti-spam trigger words.

Then ask the tool to rank its proposals by probable open rate and explain its reasoning. You keep the two best for a test. Building a high-performing subject line deserves attention, and our advice on writing a punchy email subject line usefully complements what the AI proposes.

The first prospecting email

Write a first cold prospecting email of 100 words maximum. Recipient: [logistics manager of an e-commerce company with 50 to 200 employees]. Signal: [the company has just opened a second warehouse, announced on its LinkedIn]. My offer: [a shipment management tool]. Objective: a reply, not a click. Human tone, a hook tied to the signal, a single question to finish.

The instruction “a reply, not a click” changes the nature of the text. The AI stops pushing a link and builds a conversation.

The follow-up that doesn’t harass

Write a short follow-up, 60 words, following a first email that went unanswered, sent [5 days] ago to [a sales director]. Bring something new, [a customer case from their sector], rather than repeating “I’m just getting back to you”. Light tone, no guilt-tripping.

Rewriting an existing email

The most profitable case on a daily basis. You have a message that works, you want to adapt it without starting from scratch.

Here is an email that converts well: [paste your text]. Rewrite it for [the head of an accounting firm], keeping the structure and the central argument, but adapting the examples to their profession. Keep the length.

It is precisely on this type of task that the time saving becomes tangible. BCG, in its “AI at Work” study of June 2024, notes that 58% of regular users of generative AI say they save five hours or more per week. Litmus, for its part, observes a clear shortening of email production times since these tools became widespread.

Adapting the tone to your contact

A frequent mistake is to serve the same register to everyone. An HR director in industry, a CIO in a fintech and a craftsman running a very small business do not read the same way. Specify the register in your prompt and the result is transformed.

This finesse of personalisation ties into a wider debate on AI for B2B email marketing, where the challenge is no longer to produce fast but to produce right.

What ChatGPT ignores and what ruins your deliverability

This is the blind spot of almost every guide on the subject. An AI-written email can be excellent in substance and still trigger the anti-spam filters.

Why? Because ChatGPT knows nothing about your sending infrastructure. It is unaware of the state of your sender reputation, your authentication, your history. It sometimes produces phrasing that sets off alarms at Gmail or Outlook: a build-up of trigger words, a structure too regular from one email to the next, overuse of links.

Three reflexes before hitting send.

Vary the phrasing from one campaign to the next. If ChatGPT serves you ten emails built on the same skeleton, the filters spot the pattern. Explicitly ask for different structures.

Watch out for trigger words. “Exclusive offer”, “free”, “click here”, “urgent” remain classic markers. A simple instruction in the prompt is enough to ban them.

Never neglect human proofreading. AI has no awareness of what happens after sending. Deliverability depends on your technical configuration and your reputation, not on the literary quality of the text. No writing tool replaces a healthy sending infrastructure.

Prospect data in ChatGPT: the GDPR reflex

A point that most articles pass over in silence. When you paste a prospect’s name, role, email and LinkedIn profile into ChatGPT to personalise a message, you are transmitting personal data to a third party. In France, this act falls under the GDPR.

The CNIL, the French data protection authority, in its recommendations on AI systems published in April 2024, calls for minimising the data supplied and for transparency on the use of artificial intelligence. In plain terms, avoid pouring your raw prospecting database into a prompt. Work with anonymised profiles or already public data, and keep fine personalisation for the proofreading stage, in your emailing tool, not in the consumer AI tool.

This is not an abstract legal detail. It is a question of responsibility towards the people you contact.

AI writes, you decide

The real question has never been “will AI replace the copywriter”. Selim Niederhoffer, in the second edition of his Guide du copywriting, puts it well: AI does not replace the copywriter, it redistributes the work, and strategic design remains irreplaceably human.

The figures confirm this caution. A MarketingProfs and Storyblok survey from August 2025 shows that among B2B marketers who have adopted AI, a small minority, around 6%, see a significant improvement in content performance. AI reduces production time, it does not guarantee engagement. The latter still depends on your knowledge of the recipient and your judgement.

Hence the value of a tool designed for this two-handed work. The AI assistant built into the Ediware editor generates subject lines and rephrasings directly in the context of the campaign, respecting your brand guidelines and without breaking the layout, where a generic ChatGPT forces you into successive copy-pastes. The tool proposes, you decide. It is the only model that holds up over time.

Frequently asked questions

Can ChatGPT write a complete follow-up sequence?

Yes, provided you frame each step. Ask for a sequence of three or four emails with a distinct objective per message and a delay between each. Specify what changes from one email to the next, a customer case, a statistic, a question. Without these instructions, the AI repeats the same idea in three forms.

How do I stop my AI-written emails from ending up in spam?

Deliverability depends first on your infrastructure, not on the text. On the writing side, vary the phrasing from one campaign to the next, ban the classic trigger words and limit links. On the technical side, correct authentication and a good sender reputation weigh far more than the literary quality of the message.

Can you personalise emails at scale with ChatGPT without breaching the GDPR?

Not by pouring your raw prospect database into a prompt. The CNIL recommends minimising the data transmitted. Work with anonymised profiles or public information, and reserve fine personalisation for your emailing platform, which handles this data within a compliant framework.

What is the best prompt for a B2B email subject line?

A prompt that imposes an angle, a length and constraints. Ask for several variants along distinct angles, curiosity, quantified benefit, direct question, setting a character limit and excluding exclamation marks and trigger words. Then have the proposals ranked and test the two best.

Should you tell your recipients that an email was written with AI?

No legal obligation requires it for a classic marketing email. Transparency is therefore a matter of positioning. In practice, an email reworked by a human, with an assumed brand tone, does not need to be flagged: it is indistinguishable from a message written entirely by hand.