AI personalization drafting

Use AI as a drafting layer after you have checked the message's substance. Give it the observation, context, and original wording that make the prospect relevant, then limit it to a defined part of the email. Polished copy can still feel as though it came from a process instead of a person.¹ Real observations, experiences, and context provide the substance; AI handles the expression.²

Prepare the input

Before you prompt it, decide what the draft may use. Remove any detail you cannot defend from what you know about the account or contact.

Keep the AI writer anchored to the listed contact and account fields.³ Personalization at scale requires clean, structured data, and most teams spend as much time correcting the output as they would writing from scratch.⁴

Make a short working note before opening the prompt. Include the verified observation, the wording or context behind it, and what you want the opening to make clear. Ask yourself:

• What did the person or account actually say or do?
• Which detail can I defend if the recipient asks about it?
• What should the recipient understand after reading the opening?

This keeps the model close to facts you can inspect and gives you something concrete to remove when the output starts adding guesses.

Set the drafting boundary

Choose the part AI will write before asking for prose. A narrow assignment gives you one line to check without auditing an entire message.

Start the prompt with the high-level message you want to communicate.⁵ Then ask it to write one specific personalized part, such as the recipient-specific hook.⁶

Use a prompt with these fields:

• Verified detail: the fact you can defend.
• Original context: the words or situation behind the fact.
• Message: what you want the opening to communicate.
• Write only: the recipient-specific opening line.
• Keep: the factual detail and the intended tone.
• Leave out: anything not present in the details above.

Give the model one clear job. After the first draft is sound, you can ask for a shorter version, a more direct version, or a version that keeps the same observation in a different voice.

Run the draft

Use the first pass to tighten the expression while the underlying observation stays fixed. Keep your judgment in the input and the approval step.

Add personally known context and keywords, then ask AI to tighten the draft.⁷ Make the result checkable against the person's original words.⁸ Point from each personalized phrase back to a detail in your working note.

When the same structure repeats across messages, let AI handle the repetitive drafting work while you keep responsibility for verification and personalization.⁹ Keep the fixed part of the message separate from the part that changes for each recipient. This makes errors easier to spot and lets you compare variations without losing the original point.

Move on when the line is clear, tied to a defensible detail, and limited to the job you assigned it. If the model invents a reason for the recipient to care, return to the input and remove the invention.

Review the line

A personalized opening earns its place by showing that you understood something specific. Read it for accuracy first, then decide whether the observation has enough relevance to deserve space in the email.

A specific observation from real experience and a nuanced perspective from someone who lived with the problem can make a prospect feel understood.¹⁰ AI-generated copy tends toward generic, feature-focused phrasing when it lacks genuine customer quotes and data.¹¹

Ask these questions before approving the line:

• Does the observation belong to this recipient, or could it sit in any message?
• Does the line reveal an implication that the raw fact leaves unstated?
• Would I say this sentence aloud in my own voice?
• Can the recipient recognize why I chose this detail?

Cut a line that only praises the company or repeats a broad description. Cut any claim you cannot support. Keep the useful fact and rewrite the expression around it.

Iterate on substance

After one draft passes review, use AI to explore meaningful alternatives while preserving the verified observation. Variation helps when it changes how the same fact is understood.

Ask for alternatives that preserve the fact and change one meaningful part of the line at a time, such as the angle, the degree of directness, or the question the line leaves with the recipient. Keep the versions close enough that you know which change produced the difference.

When AI identifies changes worth carrying forward, use those changes as parameters for the next email you craft.¹² Keep a record of the input detail and the wording choice so the next draft improves a repeatable part of the process.

Move on when one version sounds specific, defensible, and natural to you. More versions do not help once they repeat the same idea in different wording.

What not to do

Watch for polished copy that substitutes for substance.

• Do not ask AI to write the whole email and treat a personalized salutation as personalization; the result is generic content with a personalized salutation.¹³
• Do not request ten versions with different subject lines as a substitute for message variation; the output can remain the same generic content.¹⁴
• Do not assume polished AI copy will outperform manual writing; AI-generated cold emails performed worse than manually written cold emails.¹⁵