Posted in Setup · 9 min read
Writing in your own voice: a 10-minute setup
The six things to tell Reply Pilots once — what you do, how you write, and what you will never claim — so every draft sounds like you. Most of the work is in two of the six fields.
Farhad
In short
An AI draft sounds generic because it was given a category to write for rather than a business, so the fix is specificity in two places — a description of what you actually sell and to whom, and a short list of words you never use. Six fields cover it, they take about ten minutes to fill in properly, and the two above account for most of the difference. Tone sliders and style presets help at the margin, but they cannot rescue a draft written from a description that would fit any competitor.
Key takeaways
- Drafts sound generic because the context was generic, not because the tone setting was wrong.
- Two fields do most of the work — what you actually sell, and the words you never use.
- A banned-words list outperforms every positive tone instruction, because prohibitions are checkable and adjectives are not.
- Write the context as though briefing a new employee, not as though writing an About page.
- Keep one persona per distinct voice. A single persona covering three brands produces the average of three brands.
The first draft comes back and it is fine. Correct, polite, well-structured, on topic. And you will not send it, because it sounds like a company rather than like you, and you cannot immediately say which sentence is the problem.
That reaction is worth trusting, and it is worth diagnosing properly, because the usual response is to go looking for a tone setting. Friendlier. More casual. Add some personality. None of it works, and it does not work for a specific reason: the draft was not written in the wrong tone. It was written for the wrong business — a plausible generic version of your industry, assembled because nothing more specific was available.
Specificity is the entire fix. Six fields, about ten minutes, and two of them do most of the work.
This is not the same problem as an AI promising things you do not offer. That one is about claims, and it is the more expensive failure. This one is about sound, and it is the one that stops you using the tool at all, because a draft you have to rewrite is slower than typing.
Why does an AI draft sound generic?
Because generic is the correct answer to a generic brief.
Tell a language model you run a marketing agency and it has to write for the average of every marketing agency it has ever seen. That average is real, it is coherent, and it is exactly what you are reacting to. It is not making a mistake — it is filling a gap you left, and it fills it with the midpoint of the category.
The second reason is that fluency is the default. Absent instruction, models produce tidy, evenly weighted, slightly formal prose with a preference for tricolons and a habit of restating the question before answering it. Most people do not write that way. You have habits — a sentence length you favour, an opener you always use, a word you avoid — and none of them are in the draft because nobody mentioned them.
The third is that people describe their business the way they describe it on their website, which is a document written to sound impressive rather than to be specific. "We help ambitious businesses grow" is not a description, it is a mood, and a draft written from it will be about growth in general.
If your business description would also be true of four competitors, you have written a category, not a business — and the draft will sound like the category.
What are the six fields?
| # | Field | What to write | If you skip it |
|---|---|---|---|
| 1 | What you actually sell | The services, named, with the ones you do not offer stated | Drafts reference services you dropped, or hedge on all of them |
| 2 | Who buys it | The specific customer, with the ones you turn away | Replies pitched at the wrong reader — too technical, or too basic |
| 3 | How you talk | Formality, sentence length, contractions, first person or we | Correct but flat drafts with no rhythm of their own |
| 4 | Words you never use | Ten to fifteen. The ones that make you wince | The obvious tells survive — delve, seamless, unlock, exclamation marks |
| 5 | Three real examples | Actual replies you sent. Not marketing copy | Habits are missing, because no description captures rhythm |
| 6 | What you will not claim | Guardrails — price, timing, guarantees | Drafts promise things you cannot deliver. This is the costly one |
Fields one and four are the ones that matter. If you have three minutes rather than ten, write those two and stop. Field one gives the draft something real to be about; field four removes the tells. Everything else is refinement on top of those.
How do you write the business context so it is actually useful?
Write it as though briefing a competent new employee on their first morning — not as though writing an About page. The two documents have opposite goals. The About page is written to impress a stranger; the briefing is written so someone can do the job without asking you forty questions.
Too vague to help:
We are a friendly, professional plumbing company serving the local area with
high-quality workmanship and great customer service.
Every sentence there is an adjective. A draft written from it will be adjectives.
Specific enough to work:
Domestic plumbing and heating, mainly boiler repair, servicing and replacement,
plus bathroom installs. No commercial work, no new-build. Covering [area] within
about 20 miles.
Most customers are homeowners aged 35+, usually calling because something has
already broken rather than as planned maintenance. Price-aware but not the
cheapest-quote type — they want it done once, properly.
We turn down: emergency callouts outside hours, anything gas-related on
commercial premises, and jobs where the customer has already had two other
people out and wants a third opinion for free.
The last paragraph is the one people leave out and it is the most useful. What you turn down defines the business more sharply than what you accept, and it stops the draft being accommodating about work you do not want.
Then the voice, in the same concrete register:
Short sentences. Contractions. First person singular — I, not we, because it is
mostly me. Direct: say what the problem is before saying what it will take.
No exclamation marks, ever. Never say "reach out", "touch base", or "at your
earliest convenience".
That last line is doing more work than the four before it.
Why does a banned-words list beat a tone setting?
Because a prohibition is checkable and an adjective is not.
"Be conversational" has to be interpreted, and the interpretation drifts — over a long thread, under a complicated question, when the previous message was formal. "Never use the word leverage" does not drift. It either appears or it does not.
It is also a more accurate model of how voice actually works. Most people cannot describe their writing style, but everyone can immediately name five words they would never type. That list is a sharper fingerprint than any positive description, because it is made of real reactions rather than self-assessment.
A starting list, if you want one to react to: delve, leverage, seamless, robust, unlock, elevate, game-changing, revolutionise, synergy, reach out, touch base, at your earliest convenience, absolutely, exciting news. Cut the ones you actually use, add the two from your own industry that make you wince. Ten to fifteen entries is plenty.
Everyone struggles to describe how they write. Everyone can name five words they would never type. Use the second list.
How do you test it in ten minutes?
Do not test on an easy message. Test on the three that are actually hard.
| Test | Why this one | What a passing draft looks like |
|---|---|---|
| The routine enquiry | Baseline — most of your volume | Answers, does not over-explain, sounds like a person |
| The awkward one — someone unhappy | Tone drifts most under pressure | Direct, does not grovel, does not get defensive |
| The one where you must say no | Generic drafts are pathologically accommodating | Declines clearly, offers the alternative, stays warm |
| The technical one | Reveals whether the context is real | Uses your actual vocabulary, at the right level |
The saying-no test is the most diagnostic. A draft written from a thin context will find a way to say yes to almost anything, because agreeableness is the safe default. If your no comes back as a soft maybe, the problem is field one — you have not written down what you turn down.
Run all four, read what comes back, and change one field at a time. Changing three at once means you cannot tell which one fixed it.
How do you handle several brands without them blurring?
One persona per voice, and be strict about it.
The temptation with several clients is a single setup with a "depending on which client" note in the context. It produces drafts that sound like the average of all of them, which is nobody in particular, and it is worse than either brand's generic default because it is inconsistent as well as bland.
Separate personas mean separate business context, separate voice settings and separate guardrails — which matters more than the voice, since guarantees and pricing policies are exactly the things that differ between clients. Switch persona and the whole configuration switches with it.
The cost is real: three brands means writing the six fields three times. It is about half an hour, and it is the difference between drafts you send and drafts you rewrite.
What should you deliberately not configure away?
The last edit. A well-configured draft is a good first draft, and the ten seconds you spend changing one sentence is where the specific detail usually gets added — the thing only you know, which is what makes a reply land in a group thread or a comment section.
The messages that are the relationship. A long-standing client asking how you are is not a drafting problem. Something warm and slightly off is worse than something short and genuinely yours.
And the first message in an important thread. Configuration makes routine replies fast; it does not make the opening move in a conversation that matters. Write that one yourself and let the tool handle the twelve routine replies that follow, which is where the time actually goes.
What does Reply Pilots do here, and what does it not?
The six fields are the dashboard setup: business context and what you sell, voice settings, a banned-words list, example replies, and guardrails. Personas let you keep a separate set per brand or per account and switch between them from the panel. On top of that, per-platform rules adjust length and formality — the same voice, shorter on X than on LinkedIn — so you are not maintaining several versions of yourself that will drift apart. An optional setting mirrors the tone of the top comments in a thread, which is useful when the room is more casual than your default.
What it does not do: it does not learn your voice by watching you. There is no silent model adaptation and nothing you write is used to train anything — the configuration is what you typed, which means it is inspectable and editable, and also means a thin configuration stays thin until you improve it. It also cannot supply the specific detail that makes a reply worth reading. That is still your job, and it is the reason reading the draft is not optional.
Your next step
Open a blank note and write the two fields that matter, in the concrete register above: what you actually sell including what you turn down, and ten words you would never type. Six minutes.
Then run the saying-no test — take the last message you had to decline and generate a reply. If the draft says no clearly and still sounds like you, the setup is working. If it comes back as a soft maybe, go back to field one: you have not written down what you turn down.
Related reading
- How to stop AI from promising things you do not offer — field six, in full
- The anatomy of a LinkedIn comment that adds one real point — where voice meets a platform that punishes filler
- Comment or DM: where your next client actually comes from — keeping one voice across the handoff
See the documentation for the exact, step-by-step setup instructions.
Frequently asked questions
How long does this really take?
About ten minutes to fill in properly, and most of that is the two fields that matter — what you sell and what you never say. You can improve it later without redoing anything, and the usual pattern is a quick first pass followed by two or three small edits over the first week as you notice what the drafts keep getting wrong.
Why do my drafts still sound generic?
Almost always because the business description is generic. If your description would also be true of four competitors, the draft has nothing to work with except the category, and the category is what generic sounds like. Replace adjectives with specifics — what you actually sell, who actually buys it, what you turn down.
Should I paste in examples of my writing?
Yes, if you have three or four short real ones — actual replies you sent, not polished marketing copy. Real examples carry rhythm and habits that no description captures. Polished copy teaches it to write like a brochure, which is usually the opposite of what you want in a comment box.
What is the difference between voice and guardrails?
Voice decides how a draft sounds; guardrails decide what it is allowed to claim. A draft can be perfectly in your voice and still promise a next-day visit you cannot make. They are configured separately because they fail separately, and the second failure is the expensive one.
Do I need a different setup for each platform?
Not a different voice, but different length and formality — the same person writes shorter on X than on LinkedIn. Reply Pilots handles that with per-platform rules on top of one voice, so you are not maintaining several versions of yourself that will drift apart.
How do I handle multiple clients or brands?
One persona per brand. Each carries its own business context, voice settings and guardrails, and you switch between them from the panel. Trying to serve several brands from one persona produces drafts that sound like the average of all of them, which is nobody.
What should go in the banned-words list?
The words you notice and dislike in other people's marketing, plus any industry jargon you personally avoid. Common entries are delve, leverage, seamless, robust, unlock, elevate, game-changing and any exclamation mark. Ten to fifteen entries is plenty — the list works by removing the tells, not by covering the dictionary.
Will this stop the drafts from sounding like AI?
It removes most of the obvious tells, which is the majority of the effect. What remains is structural — a slight over-tidiness, an even sentence rhythm — and the reliable fix for that is editing one sentence before you send. Reading each draft is worth doing anyway, and it is where the specific detail usually gets added.
Related articles
How Comment & DM Specialists Can Get an AI Reply Assistant Running in 15 Minutes
The tension here isn't whether to set this up — it's whether stopping for it feels acceptable. Here's why a genuinely bounded 15 minutes resolves that specific tension.
Read article →
How DM-to-Book Agencies Can Get an AI Reply Assistant Running in 15 Minutes
Fifteen minutes, and none of it requires stepping away from a live conversation mid-flow. Here's the actual sequence that proves it.
Read article →
How Freelance Social Managers Can Get an AI Reply Assistant Running in 15 Minutes
Fifteen minutes sounds like a marketing number until you see the actual steps. Here's the specific sequence for a solo operator's first client.
Read article →