You let AI draft an email to a prospect. You read it back and cringe. It sounds like a robot wrote it. Full of buzzwords. Too much enthusiasm. No personality. That does not sound like your business at all.
“I do not want to sound like everyone else” is a valid concern. Generic output is what happens when the workflow lacks approved examples, explicit voice rules, and a person who can reject a draft.
The foundation is a one-page voice source, real examples, a two-pass review, and a record of the edits the tool keeps needing.
Why AI Sounds Generic by Default
A general model is not grounded in your customers, policies, offer, or preferred phrasing unless you provide that context. Without it, the draft tends toward broadly familiar business language.
The fix is to teach AI who you are. Not with pages of instructions. With one good example and clear rules.
Build Your One-Page Brand Voice Doc
You do not need a 50-page brand guide. Open a Google Doc and fill in four sections:
1. Who is your audience? (One sentence) Example: "Small business owners who feel behind on technology and want practical help, not complicated solutions."
2. Your tone in three to five words: Example: "Clear, direct, honest, helpful, no jargon." Not "professional and modern." Those do not help AI. Specific words do.
3. One to two real examples of your writing: A past email you sent to a prospect. A LinkedIn post. A response to a customer problem. Copy and paste your actual words. AI learns from examples better than from descriptions.
4. Dos and don'ts: Keep it to five of each. Examples of "do": "Use short sentences," "Address real problems," "Be specific." Examples of "don't": "No buzzwords like leverage or synergy," "Never fake enthusiasm," "Avoid corporate speak."
That is a starter voice document, not a guarantee of fidelity. Test it on representative drafts, record corrections, and add approved examples for objection handling, audience shifts, and channel-specific tone. Keep a person responsible for rejecting output that is accurate but still sounds wrong.
| Business type | Voice cue to capture | Useful example to paste into AI |
|---|---|---|
| Local service business | Helpful, calm, practical, no scare tactics | A past estimate follow-up or appointment reminder |
| Restaurant or retail shop | Warm, specific, seasonal, community-aware | A menu announcement, review reply, or customer thank-you |
| Professional service firm | Direct, credible, plain-English, specific about outcomes | A proposal email or explanation of your process |
Use Examples in Your Prompts
Instead of telling AI "sound friendly," give it an actual sentence you wrote. Example: "Write this email in the style of this sentence: 'Here is what happened, here is what we fixed, and here is what to expect next.'"
Or: "Draft a LinkedIn post about [topic]. Write it like this example: [paste one of your real posts]."
Real examples give the tool concrete sentence rhythm, length, and vocabulary to follow. Test the result against your own review standard rather than assuming an example guarantees fidelity.
The Two-Pass Review System
Never publish AI output without review. Build a simple two-pass system:
Pass 1: Does it have your voice? Read it as if you sent it. Does it sound like you? If not, mark the parts that sound off and ask AI to rewrite those sections using your voice doc as reference.
Pass 2: Is it accurate? Check facts, links, offers, dates. Make sure nothing contradicts your actual policies or previous statements. AI can sound confident while being wrong. You are the fact-checker.
Record how many edits or rewrites each draft needs. If review takes longer than the old process or the same voice errors recur, improve the source and prompt—or stop using the workflow.
Spot AI Tells Before Your Customers Do
Learn the patterns that scream "AI wrote this":
- Cliches stacked together ("innovative solution," "cutting-edge technology," "powerful insights")
- Over-hedging ("might be," "could potentially," "arguably")
- Fake enthusiasm (exclamation points! Multiple question marks?? Celebration emojis)
- No specifics ("help your business," "improve results," "enhance outcomes")
- Weird word choices (AI sometimes picks uncommon synonyms that sound off)
If you spot these, rewrite the sentence in your own voice or strengthen the instruction and example. Treat repeated “AI tells” as a quality defect the review process should catch before publication.
When NOT to Let AI Write
Customer apologies: If something went wrong, your customer needs a real human acknowledgment. AI writes apologies that feel hollow. Write these yourself. Your customers deserve authenticity when they are upset.
Sensitive or complex issues: A customer complaint about a safety concern. A privacy problem. A lawsuit question. These are not places for AI first drafts. You think these through on your own.
Relationship moments: Welcoming a long-time customer back. Congratulating a partner on something personal. These moments build loyalty. They need your voice, not AI.
Anything that needs legal accuracy: Terms and conditions. Warranty language. Medical or compliance claims. Have a lawyer review anything important before it goes out.
Knowing where to draw these lines is part of what I help clients map out. A good AI workflow has clear "AI assists" lanes and clear "human only" lanes, so nothing important slips through.
Voice rules and approved examples give reviewers a shared standard; the correction log shows what still needs work.
Do and Don't List
Finally, keep a running doc of what works and what does not for your voice:
Do: Be specific about problems | Use short sentences | Name the actual customer benefit | Use "you" and "we" | Reference past work or wins | Show results with numbers
Don't: Stack adjectives | Use words like "delve," "dive," or "revolutionize" | Act like you have all the answers | Use technical jargon without explaining it | Make big claims without proof | Copy competitor language
This list should evolve. After an observation window with enough drafts, review the correction log, update the voice document, and test whether consistency improves without hiding factual errors.
Where voice systems drift
Writing the voice document is only the start. The system drifts when any of these controls are missing:
- The doc lives in someone's head, not on paper. Anyone you hire or any tool you connect ends up guessing.
- Different tools produce different voices. The chatbot sounds one way, the email assistant another, the social scheduler a third. Customers notice the inconsistency before you do.
- The system never gets reviewed. Six months in, the voice doc is stale and AI is drifting back to generic.
A structured setup turns voice from an impression into a review standard. We document approved examples and exclusions, apply them to one workflow, log corrections, and schedule a review before carrying the rules into another tool.
Start Today
Draft a brand voice document, share it with the team, and test it in the next AI prompt. Record which corrections remain; the document is useful only when reviewed drafts become more consistent without hiding mistakes. For prompt examples that use this kind of context, see AI Prompts for Small Business Marketing. For customer-facing support replies, pair it with the customer service AI guide.
If you want to test one voice system across email, social, customer replies, and outreach, book a free 15-minute bottleneck review. Bring two examples that sound right and one draft that does not; we will map the source, rules, review pass, and correction log. You can also see how my service engagements are structured or read about safety and privacy considerations before our conversation.
