Good tone of voice after editing AI copy sounds like one real person at your business talking to one customer: shorter sentences, specific details instead of adjectives, your own recurring words, no stock AI phrasing and a clear next step. The quickest test is whether a competitor could paste it unchanged. If they could, keep editing.
The catch is that tone isn't a single setting. The same voice should sound softer in an apology than in a price list, and most editing goes wrong because people swap individual words while leaving the AI's structure in place: the three-item list, the question at the top, the tidy summary sentence at the end. Those shapes give a draft away faster than any single word does.
The examples below follow one illustrative business, a two-person handmade jewellery seller with its own website and a marketplace shop, through a month of editing its AI drafts. After that come shorter before-and-after pairs from five other kinds of business. Every draft and output here is invented to show common patterns, not copied from a real company.
One rule applies to everything that follows: edit against written rules, not against a feeling. "Make it sound more like us" gets you a different result from every person who edits. "Sizes before adjectives, one warm line, no exclamation marks" gets you the same result from all of them.
A jewellery seller whose AI copy sounded like every other shop
The owner makes rings, earrings and pendants in recycled silver and releases about 40 new pieces each season. For the last two drops she used ChatGPT Plus to draft product descriptions, order emails and Instagram captions. The drafts were fluent and fast, and within a month two repeat customers had asked whether the shop had been sold, because the new listings "don't sound like you any more".
Her editing at that point was light: fix anything factually wrong, cut a sentence or two, publish. Each listing still took about 12 minutes, because she kept rereading drafts that felt wrong without being able to say why. That is the usual sign you have no written voice rules yet.
So before touching any drafts, she printed ten listings and ten customer emails she had written herself before using AI, marked the phrases that sounded most like her, and boiled them down to three traits. She wrote each one as a rule someone else could check.
| Trait | What it means on the page | What it rules out | A line that passes |
|---|---|---|---|
| Plain-spoken maker | Say what it's made of, how, and by whom. Write as "I", because one person makes everything. | "Timeless elegance", "exquisite craftsmanship", "our team of artisans" | "Hammered by hand from recycled sterling silver, so no two are quite the same." |
| Warm, not gushing | One friendly line per piece at most. No stacked exclamation marks. | "You'll absolutely adore these!!" | "It's the pair I wear most days." |
| Useful before pretty | Size, weight, metal, care and dispatch time come before mood. | A paragraph of atmosphere with the size missing | "Drop length 28mm, about 2g each, so fine for all-day wear." |
If you have never written rules like these, the tutorial on how to build a brand voice guide AI can follow covers the full process. For editing, three traits are enough to start, because every edit below traces back to one of them.
Product description: the draft, the problems, the rewrite
Here is the kind of first draft ChatGPT produced from a one-line brief ("small hammered silver hoops, handmade, recycled silver"). It is illustrative, but anyone who has asked an assistant for product copy will recognise it.
Before (AI draft): Discover the timeless elegance of our Hammered Hoops, where artistry meets everyday luxury. Meticulously handcrafted with passion, these stunning earrings are the perfect accessory for any occasion, whether you're heading to the office or enjoying a night out. Treat yourself or someone special to a piece that will be cherished for years to come.
What was wrong with it, trait by trait:
- Plain-spoken maker: "timeless elegance" and "where artistry meets everyday luxury" could describe any earring from any shop in the world. "Meticulously handcrafted with passion" claims care without showing any.
- Warm, not gushing: "stunning", "perfect" and "cherished for years to come" in three sentences is gushing by any measure.
- Useful before pretty: no diameter, no weight, no clasp type, no care advice, no dispatch time. A buyer has to message to ask, or leaves.
- Structure: the "whether you're doing X or Y" construction and the closing call to action are two of the most recognisable AI shapes. The button on the page already says "Add to basket", so the last sentence does nothing.
After (edited): Small hammered hoops in recycled sterling silver, 18mm across. I shape each one around a steel rod, then hammer the front so it catches the light without looking shiny-new. They weigh under 2g, so you forget they're in, and they close with a simple hinged catch. Rinse and dry after swimming; a polishing cloth brings back the shine. Made to order and posted within 3 working days.
The rewrite is about the same length, yet nearly every sentence now carries a fact: a size, a process, a weight, a care step, a date. The one warm phrase ("you forget they're in") earns its place because it tells the buyer something useful about comfort. Notice too that nothing had to be invented. Every detail came from the owner, which is why a brief with the numbers in it matters more than any clever prompt.
Order-delay email: one apology, one reason, one date
Customer emails show tone problems more sharply than product pages, because a vague reply to a worried customer reads as evasive. This was the draft the assistant offered when asked to "write a polite email about a delayed order".
Before (AI draft): Dear Valued Customer, We sincerely apologise for any inconvenience this delay may have caused. We understand how frustrating it can be to wait for an order. Please rest assured that our team is working diligently to process your order as quickly as possible. We truly appreciate your patience and understanding. Warm regards, The Team
Four sentences of apology, no facts. It addresses a stranger when the shop knows the customer's name, mentions a team when there are two people, and gives no new date, no reason and no choice. Nothing in it is rude. It simply says nothing, and customers notice that.
After (edited): Hi [first name], your silver leaf necklace is running late, and I'm sorry. The chain supplier sent the wrong length, so I'm waiting on the right one, due Thursday. I'll finish your necklace and post it on Friday by tracked delivery, and email you the tracking number the same day. If you'd rather not wait, just reply "refund" and I'll refund you in full today. Thanks, [owner's first name]
The edit follows a pattern worth copying for any bad-news email: say sorry once, give the real reason in one sentence, give the new date, offer a choice, sign with a person's name. If you draft many replies like this, the tutorial on drafting customer email replies with AI that sound like you goes further into reply prompts.
Instagram caption for a seasonal collection
Captions are where AI enthusiasm runs furthest. The draft below came back from a prompt asking for "an exciting caption for our new autumn collection".
Before (AI draft): Introducing our NEW Autumn Collection! Embrace the season with pieces inspired by nature's beauty. Each design is crafted to capture the magic of falling leaves. Shop now via the link in bio! #handmadejewellery #silverjewellery #autumnstyle #shopsmall #giftideas #leafjewellery #slowfashion
Besides the generic lines, there is a practical error: seven hashtags. Instagram has capped posts at five hashtags since December 2025, so a caption written from an older habit fails before tone even matters. Assistants trained on older posts still suggest long hashtag blocks, which is one reason to keep platform rules in your prompt.
After (edited): Six new pieces for autumn, each cast from a real oak or ash leaf I picked up on the walk to the workshop. The oak studs are the smallest thing I've ever cast: 9mm from tip to stem. They go live on the website at 7pm tonight, link in bio. #handmadejewellery #silverjewellery #leafearrings
Specific origin, one surprising detail, a launch time, three hashtags. The Instagram captions tutorial has more caption prompts if social posts are where your AI copy drifts most.
Week by week: how the edits became a prompt
Editing every draft by hand forever would cancel out the time AI saves. The point of the month was to turn the edits into instructions, so later drafts would arrive closer to finished.
Week 1: collect and mark
She put ten pre-AI pieces next to ten AI drafts and highlighted every phrase she would never say aloud. Words that turned up three or more times went onto a banned list: timeless, elegance, stunning, exquisite, "perfect for any occasion", "treat yourself", "rest assured". The tutorial on writing a banned-phrase list for your brand shows how to keep that list short enough to be obeyed.
Week 2: write the pairs
She rewrote six drafts by hand, two each of product descriptions, customer emails and captions, and saved each as a before-and-after pair with one line explaining the main change. Six pairs took about two hours.
Week 3: put rules and pairs where the assistant can see them
She pasted the three traits, the banned list and the six pairs into the instructions of a ChatGPT Project, so every chat in that project starts with them. A Claude Project or a Gemini Gem (becoming a skill from November 2026) does the same job. One thing she did not do was build a custom GPT: OpenAI is retiring custom GPTs, which stop running on 11 December 2026, so a project is the sensible home for this now. The prompt she used for new listings looked like this:
You write product descriptions for a one-person handmade jewellery brand.
Voice rules:
1. Plain-spoken maker: say what it is made of, how, and by whom. Write as "I".
2. Warm, not gushing: at most one friendly line. No exclamation marks.
3. Useful before pretty: size, weight, metal, care and dispatch time come first.
Never use: timeless, elegance, stunning, exquisite, "perfect for any occasion",
"whether you're ... or ...", "treat yourself", "cherished".
The before/after pairs in this project show the pattern. Learn the pattern;
do not reuse their sentences.
Brief: [piece name] | [metal] | [size in mm] | [weight in g] | [how it is made]
| [care] | [dispatch time]
Write 50-80 words. If the brief is missing a detail, write [CHECK: detail]
instead of guessing it.Here is an illustrative output for a new signet ring, with the brief missing its size range:
Sample output: A chunky signet ring in recycled sterling silver with a flat oval face, 12mm by 9mm. I cast it from a wax I carve by hand, so the edges stay slightly soft rather than machine-crisp. It weighs about 8g, so you'll notice it on your hand. [CHECK: available ring sizes]. Polish with a soft cloth and keep it away from bleach. Made to order, posted within 3 working days.
What she still fixed: the weight in her brief was an estimate, so she weighed the sample ring before publishing, and she replaced the placeholder with the real size range. The placeholder did its job. Without that instruction, assistants tend to fill gaps with plausible numbers, and a wrong ring size is a return waiting to happen.
Week 4: measure the edit time
She timed ten listings. Editing fell from about 12 minutes each to about 5, mostly spent checking facts rather than rewriting sentences. Across a 40-piece drop that is 40 × 7 = 280 minutes saved, a little under five hours per season, and the listings read like the old ones again. Your numbers will differ, but timing ten pieces before and after is the honest way to find out whether the rules are working.
Five more before-and-after pairs from other businesses
The same three moves (facts before feelings, one human line, a clear next step) work outside jewellery. Each pair below is illustrative and uses a different kind of copy.
Print shop: a quote reply
Before: Thank you for reaching out! We would be delighted to assist you with your business card printing needs. Please find our competitive pricing below, and do not hesitate to contact us should you have any further questions.
After: Hi [first name], 500 double-sided business cards on 400gsm silk come to $64 including delivery. Send the artwork by 2pm Wednesday and they'll reach you on Friday. One thing to check first: your logo is a low-resolution PNG, so it may print fuzzy. A PDF or SVG version would fix that.
The edit leads with the price, gives a deadline tied to a delivery day, and adds the one warning a good printer would give in person. That warning does more for the shop's reputation than "we would be delighted".
Software reseller: a renewal reminder
Before: We hope this message finds you well! As a valued partner, we wanted to remind you that your subscription is approaching its renewal date. Don't miss out on continued access to industry-leading productivity features.
After: Your 12 Microsoft 365 Business Standard licences renew on 14 November at $14 per user per month on your annual commitment. Two of them haven't been signed into for 60 days. If you drop those two before renewal, you'll save $336 a year. Reply by 7 November and I'll make the change.
The quick sum (2 licences × $14 × 12 months = $336) turns a nag into a favour. For a reseller, that kind of useful honesty is the brand voice.
Events company: a first reply to a wedding enquiry
Before: Thank you so much for considering us for your special day! We would be absolutely thrilled to be part of your love story and help bring your dream celebration to life.
After: Thanks for getting in touch about Saturday 14 June. We're free that day. For 120 guests in a marquee, I've attached two sample plans at different budgets so you can see where the money goes. Could you tell me roughly where the venue is and whether you need evening lighting? Then I can send a proper quote within two working days.
The edited reply confirms availability in the second sentence, which is the thing the couple most wants to know, and asks the two questions that block a real quote.
Sign maker: a services page intro
Before: At our company, we pride ourselves on delivering high-quality signage solutions tailored to your unique needs, combining creativity with the latest technology.
After: We make shop fascias, projecting signs, window graphics and van livery. Most fascias take three to four weeks from approved proof to fitting, and we handle the fitting ourselves. If you have a photo of your shopfront, send it and we'll mock up two options for free.
"Signage solutions tailored to your unique needs" is replaced by a list of things the business actually makes and a real lead time. Phrases such as "combining creativity with the latest technology" are exactly what belongs on a banned list, because no customer has ever chosen a sign maker for that reason.
Online clothing shop: answering a sizing question
Before: Great question! Sizing can vary from brand to brand, so we always recommend checking our comprehensive size guide to find your perfect fit.
After: That linen shirt runs roomy. If you're usually a medium in our shirts, most people go down to a small in this one. The medium measures 112cm around the chest, if you'd like to compare it with a shirt you already own.
"Check the size guide" is what a customer hears when nobody has read their question. The edit answers it with a direction (size down) and a measurement they can check at home.
A five-question tone check for any edited draft
Print this and keep it next to whoever publishes. It takes under a minute per piece once it becomes habit.
- Could a competitor paste this unchanged? If yes, it needs a detail only you could write: a measurement, a date, a process, a name.
- Would you say it out loud to a customer standing in front of you? Read it aloud. Anything you'd feel silly saying goes.
- Is the structure still the AI's? Look for the question opener, the three-part list, the "whether you're X or Y" line and the summary sentence at the end.
- Is every number and claim confirmed? Tone edits often add specifics, so check that the specifics are true. A confident wrong size is worse than a vague right one.
- Does the reader know what happens next, and when? A date, a deadline or a single action. Not "do not hesitate to contact us".
If your drafts keep failing question 3, the reasons behind that pattern are covered in why AI marketing copy sounds generic, which pairs well with this check.
How edited copy slides back into sounding generated
Good edits decay. These are the ways it happened to the jewellery seller, or could have, and how each one showed up.
- Example echo. Three weeks in, she noticed "you forget they're in" in nine of fifteen new listings, including a chunky ring where it made no sense. The phrase appeared in two of her six pairs, so the assistant treated it as house style. The fix was to replace one pair and add the line "do not reuse sentences from the examples", which the prompt above now includes.
- Over-correction. Told to be warmer, one draft came back as "These hoops are proper gorgeous, honestly obsessed". Warm, not gushing, means one friendly line, not slang. The rule table settled it in seconds.
- Two voices on one shop. Her part-time helper edited customer emails in his own style, which was brisker than hers. Customers who emailed twice got two personalities. They agreed on the five-question check and a shared sign-off, and she spot-reads five of his replies each Friday.
- Tool or model changes. After an update to the assistant, drafts came back noticeably longer. Because the rules and pairs live in a document she owns, she re-ran three standard test briefs, compared the results with her approved versions and tightened the word limit. Keep your test briefs; they are how you notice drift.
- Email tools with their own tone matching. Gmail's Help me write can now match the tone of your past emails on Google Workspace business plans and Google AI plans. That helps only if your past emails were good. Skim a dozen of your sent replies before trusting it to copy them.
A monthly look at ten random published pieces against the five questions catches most of this before customers do. If you would rather set up the rules, pairs and project with someone, that is part of what my AI implementation consultation covers, but everything above works without outside help.
More questions about editing AI copy for tone
How many before-and-after pairs should go into the prompt?
Start with two pairs for each type of copy you produce, so about six in total for most small businesses. If new drafts start repeating whole sentences from your examples, the pairs are too similar or too few in number for that job. Swap in fresh pairs and tell the model to learn the pattern without reusing the wording.
Should part-time staff edit in their own voice or the brand's?
The brand's, but let their personality show in small places such as a sign-off or one line about why they like a product. Give everyone the same short list of voice rules and the five-question check, and have one person spot-read a handful of edited pieces each week so the voice doesn't split into several.
What if my old copy, written before AI, was not very good either?
Use it for vocabulary and attitude, not quality. Old emails often show how you naturally explain things to customers even when the grammar is loose. Take the phrases and habits that feel like you, fix the clumsiness, and write fresh model answers that become your before-and-after pairs.
Does a tone of voice change need a new AI tool?
Rarely. The rules and example pairs do most of the work, and every mainstream assistant can hold them in a project or saved instructions. Specialist brand-voice tools mainly help larger teams that need shared rules, approvals and reporting across many writers.
Further reads
- How to Write a House Style Guide Your Team and AI Both Follow — Turn your voice rules into a full house style guide for people and AI.
- How to Write AI Prompts That Sound Like Your Business — More prompt patterns for getting on-brand drafts first time.
- Brand Voice Tools Compared: Jasper, Writer and Custom Instructions — When a paid brand-voice tool beats saved instructions.
- How Much Does Human Editing of AI Content Cost? — What a human editor costs if you would rather hand this off.
- AI Content Approval Workflow: Draft, Check, Sign Off — Add a sign-off step so edited copy stays edited.
- Should You Pay for an AI Humaniser Tool? What It Does and Risks — Why paying a humaniser tool is no substitute for these edits.
- How to Write Menu Descriptions With AI That Still Sound Like You — A five-stage method for AI-drafted menu descriptions that keep your kitchen's voice, with a voice card template, a drafting prompt and a before-and-after dish.
- How Photographers Can Handle Client Emails With AI — The email map of a photography booking, a voice card for the AI, and prompts for the awkward ones: RAW file requests, late galleries and unhappy clients.
- Shopify Magic vs ChatGPT for Product Descriptions — Shopify Magic is free and sits in the description box; ChatGPT handles bulk rewrites and brand voice. A decision table, a fair test and sample outputs.
- How to Answer Etsy Customer Messages Faster With AI — Quick replies for the repeat questions, AI drafts for the rest, and a tone check for the hard ones, so Etsy messages get answered well inside 24 hours.
- Using AI in a Yoga or Pilates Studio While Keeping Your Voice — Why AI makes every studio sound the same, and the voice sheet, before-and-after edits and teacher-bio method that keep yours recognisably yours.
- Does Writing Etsy Listings With AI Hurt Your Search Ranking? — How Etsy search actually ranks listings, the specific ways AI-drafted copy loses ground, and a simple test to run on your own shop.
- How Agencies Keep Brand Voice Consistent in AI-Written Content — A filled-in voice card, two client voices side by side, a sample-led prompt and a five-line scoring rubric for keeping AI copy on-brand.
- How Funeral Directors Can Use AI for Admin With Care — Where AI can take funeral admin off your desk, where it must never speak for you, and the read-back routine that stops a misspelt name reaching print.
- How Freelance Writers Use AI Without Losing Their Voice — Keep control of the angle and first draft, then use targeted prompts to find weak passages without flattening your writing.
- How Tutoring Centres Use AI to Write Parent Progress Reports — Short session notes in, specific parent reports out: the note fields, prompt and checks that cut a term of reports from 20 hours to about 7.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI help pages (custom GPT retirement and migration FAQ); Claude help pages (personalisation features: instructions, projects and skills); Google Workspace information on Gmail Help me write tone matching; Instagram hashtag limit as recorded in the series fact sheet (checked September 2026).