Give AI approved writing examples, explicit voice rules and a separate brief for the current email. Save the voice guidance for reuse; update the audience, purpose and verified product or offer facts for each send. Review brand fit and factual accuracy separately. A draft that sounds right still needs correction if it invents a promise.
“Make it friendly and professional” leaves an editor with a question: what would count as getting it wrong? Start by writing guidance that lets a reviewer point to a sentence and explain why it belongs in your emails—or why it does not.
The workflow below is a suggested way to brief and review AI writing. It is not a guarantee of better campaign results. Our guide to AI copy and retention handles that separate question.
Keep the voice stable; adapt the tone to the moment
Mailchimp’s style guide makes a useful distinction: its underlying voice stays relatively stable, while its tone changes with the situation and the reader’s emotional state.
For your own retention emails, write down what should remain recognisable across sends, then specify how the current message should feel. You might want plain, practical language throughout, with more enthusiasm for a product announcement and a quieter approach when addressing frustration.
Nielsen Norman Group’s tone framework offers concrete choices: formality, humour, respectfulness and enthusiasm. It also recommends naming unwanted tones and varying tone to fit the reader’s emotional state and the topic.
Turn a vague adjective into a boundary. Instead of stopping at “playful,” try “light humour about the product’s everyday uses; no jokes about a customer’s problem.” That is an editorial choice you can check in the draft, not a rule every brand should adopt.
Save a voice brief with examples and boundaries
Use this proposed structure as a starting point. The homewares brand and writing references here are invented.
| Brief field | Illustrative guidance |
|---|---|
| Relationship to the reader | A helpful shopkeeper explaining the item plainly |
| Voice | Warm, practical and understated |
| Writing rules | Name the item; use everyday words; make the next action clear |
| Boundaries | No inflated praise, invented scarcity or unsupported discounts; no jokes when addressing frustration |
| Approved writing references | “Our Cedar bowl is made for everyday lunches.” “A replacement lid is available for your jar.” |
| What to imitate | Direct openings, concrete nouns and a useful next step |
For a real brief, replace the invented references with approved writing from your own brand. Annotate what you want copied: perhaps the opening, vocabulary or way of explaining care. Include examples from different situations so the brief does not instruct every email to sound like a sale announcement.
Anthropic’s prompting guidance recommends explicit instructions and relevant, varied examples for its models. It also recommends separating instructions, examples and variable inputs. Treat the structure here as an application to email writing that needs checking with your chosen tool.
Label the examples style references only. Their old prices, offers and product details must not become facts for the next campaign.
Add the current email’s job and facts separately
Keep the reusable voice brief intact and attach a fresh task. Specify who will receive the email, why you are writing now, what the reader should do and which facts are approved for this send.
Here is a prompt template to adapt:
Write an email draft using the supplied sections.
VOICE AND STYLE
[Paste the saved voice brief and approved examples.]
Use examples for writing style only, not product or offer facts.
THIS EMAIL
Audience and lifecycle moment: [who, and what has happened]
Purpose: [what this email needs to communicate]
Approved facts: [current product details, terms and source material]
Desired action: [what the reader should do next]
Tone for this situation: [how to apply the saved voice here]
OUTPUT
Return a subject line, body and CTA.
Use only the approved facts for factual claims.
List missing information separately instead of inventing it.For a welcome email, fill in the actual signup context. For a win-back email, specify the customer situation and any approved offer. Do not silently import a discount from an example just because the example sounds convincing.
This separation is also useful when the voice is right but the message is wrong: correct the current task or facts before rewriting the saved personality.
Rewrite the sentence without inventing a better offer
Consider an invented send to returning customers announcing a new moss colour for the Ridge mug. The approved details are stoneware and hand-wash-only care. There is no discount, stock claim or deadline. The desired action is to view the moss mug.
Both versions below are author-written illustrations, not observed AI output.
| Version | Copy | Editorial decision |
|---|---|---|
| Reject | “Elevate your morning ritual with the perfect dishwasher-safe mug. Grab yours before it disappears!” | Inflated language conflicts with the brief; dishwasher safety contradicts the supplied care instruction; urgency has no support |
| Accept for this invented brief | “The Ridge mug now comes in moss. Made from stoneware; hand-wash only.” CTA: “View the moss mug.” | Names the actual change, preserves the care instruction and follows the understated voice |
The acceptance decision has two parts. “Elevate your morning ritual” is a voice edit for this particular brand. “Dishwasher-safe” is a factual error. A more conversational rewrite would still fail if it kept that error.
Check what the tool actually saves
If your platform has saved voice settings, inspect its current documentation before treating those settings as the brief for every AI feature.
For example, Klaviyo’s current brand-voice instructions give the path Content → Media & brand → Brand → Voice. Use Edit voice to change descriptors and writing rules, Update preview to inspect sample content, then Save. Previewing alone does not save the changes.
Those settings affect the whole account. Uploading a style guide replaces existing descriptors and writing rules, including manual edits, so retain your approved guidance before replacing it.
The same help page says the former Email AI campaign-drafting flow and its voice quick-edit control are no longer available. Saved brand settings are not evidence that this older drafting workflow still exists. Verify the feature you intend to use and inspect its resulting draft.
Review the draft and keep the corrections worth reusing
Anthropic’s hallucination guidance recommends allowing uncertainty and verifying generated claims against supplied sources. Apply that here as an editing step; an instruction to use approved facts does not replace checking the output.
Use these proposed acceptance questions:
| Check | Decision to record |
|---|---|
| Voice | Which sentence conflicts with an approved rule or example? |
| Situation | Does the tone fit the recipient’s circumstances? |
| Facts | Can each product detail, offer and promise be traced to the current brief? |
| Action | Does the CTA ask for the intended next step? |
| Reuse | Is the correction a lasting voice rule or a fact for this send only? |
Try the saved guidance on a welcome, announcement and win-back draft, supplying the context each time. Retain the brief version, current facts, draft, accepted wording and reasons for edits. This is a proposed review exercise, not a validated test set.
When a style correction recurs, consider adding it to the saved brief. Keep product availability and offer terms in the current campaign facts. If the decision is whether this process saves work, measure that separately using the AI production-time guide.
Should I give AI old emails to learn the brand voice?
Use approved emails as style references, with notes explaining what to imitate. Keep their product details and offers separate from the current send’s approved facts. The proposed workflow uses examples to guide writing, not to establish whether an old promotion is still valid.
Should every lifecycle email use the same tone?
Keep the underlying brand personality consistent while adapting tone to the recipient’s situation. Nielsen Norman Group recommends considering the user’s emotional state and topic. In the proposed brief, specify that context for each send rather than applying a promotional tone automatically.
Is on-brand AI copy ready to send?
Only after checking its facts and intended action as well as its voice. In the proposed workflow, reject unsupported product details, discounts, deadlines or promises even when the sentence sounds right. Ask for missing information and verify the final wording against the current approved brief.