"We have a list of a few hundred prospects. Writing each email by hand takes forever, but when we let AI write them, they all sound the same: 'I was impressed by your journey...' Nobody replies." AI tools such as ChatGPT, Gemini or Claude can help, but only if you use them for the right part of the job and a person stays in charge of what goes out.

Why AI outreach sounds robotic

  • The prompt says "write a personalised email" with nothing real to personalise from, so the AI fills the gap with flattery.
  • Every message has the same structure and the same phrases.
  • The AI guesses facts about the company and sometimes gets them wrong.
  • The messages are long, formal and full of words nobody uses in real conversation.

Recipients notice all of this within the first sentence.

What real personalisation means

Personalisation is not inserting a name and a company. It is a real reason why this message is relevant to this person. It works on three levels:

  • Segment: what you say to logistics companies differs from what you say to garment factories.
  • Company: something true about them, such as a new branch, a new product line or a job opening that suggests a need.
  • Person: their role, and what they are likely responsible for.

Often a strong segment message with one accurate, company-specific line works better than a fully rewritten email.

A workflow that keeps a person in charge

  1. Start with a clean, ethically sourced list. People in roles that fit your offer, gathered from public business sources, not bought or scraped lists.
  2. Add research columns. What the company does, one recent and checkable fact, and the likely need. A person can collect this, or AI can summarise a company's own website, with a person checking the summary.
  3. Write the base message yourself for each segment. Short, plain, with one clear question.
  4. Ask AI for the opening line only. Give it the research notes, your base message and two or three examples of your tone. Tell it to use only facts from the notes, avoid compliments and keep to one or two sentences. If there's no real hook, use the segment message unchanged.
  5. A person reviews every message before sending. Check each fact, cut anything invented or awkward, and make sure it sounds like you.
  6. Send in small batches and read the replies. Adjust the base message based on what people actually say.
  7. Keep the basics: your real name and company, and an easy way to opt out. Respect every "no".

I usually start with the base message and only add AI once that message is getting sensible replies on its own.

Mistakes to avoid

  • Sending hundreds of AI messages without reading them.
  • Personal details that feel intrusive, such as comments on someone's family or private posts.
  • Fake familiarity and invented shared interests.
  • Letting the AI claim results or prices you haven't checked.

Who does what

A small team can run this with a spreadsheet and an AI chat tool, as long as someone has time to review. Help is useful for building the research process, setting up prompts and templates so they stay consistent, connecting the sheet to your sending tool, and protecting your email domain. If LinkedIn is part of your plan, see LinkedIn outreach that gets replies.

Start with twenty

Take twenty prospects from one segment. Write the base message yourself, use AI only for the opening lines, review each one and send them. Read every reply before scaling up. See lead generation and B2B outreach, or contact me.

Related service Lead generation and B2B outreach Targeted prospect lists, verified contact data and outreach that reaches the inbox and turns into sales conversations.