How to Sound Like Yourself When Using AI to Write | Personal Branding Guide
“There is no greater agony than bearing an untold story inside you.” ~ Maya Angelou, I Know Why the Caged Bird Sings
To sound like yourself when you use AI to write, keep the two things AI cannot supply in your own hands: your raw material (stories, decisions, numbers, opinions) and your final edit.
Let AI do the middle work of interviewing you, organizing your thoughts, and drafting against a written guide to your voice. That one division of labour is the difference between a post that sounds like you and a post that sounds like everyone (the average sum).
Most people use AI the other way around. They type a topic, accept the draft, and publish. The result is fluent, polished, and forgettable, because the model filled every gap with the most statistically likely words. Those words belong to no unique mind.
This guide is for people who do real work and want more of the right people to find it. You probably have a demanding job, a full calendar, and a folder of ideas you never published. AI can give you back the hours. It should not take your voice in exchange.
We start from the basics: how language models produce text, what a “voice” is made of, and why it matters for trust. Then we build a complete system you can set up this week, with every template and prompt included. We are holding nothing back.
What's included in this guide:
Why AI writing sounds generic, explained from the ground up
The six layers that make up a writing voice
A 6-step system: voice fingerprint, raw material bank, AI workspace setup, interview-first drafting, de-AI editing, and a 90-minute weekly routine
Copy-and-paste prompts and a full voice guide template
How the system changes for data leaders, AI leaders, founders, and executives
How to check that it is working, and how to think about disclosure
New to personal branding? Start with our foundational guide, How to Build a Personal Brand That Works. This article zooms in on one step of that system, your voice, and shows how to protect it when AI becomes part of your process.
Why Does AI Writing Sound Generic?
“Whenever you find yourself on the side of the majority, it is time to pause and reflect.” ~ Mark Twain
AI writing sounds generic because a language model predicts the most likely next word, and later training teaches it to prefer safe, widely approved phrasing. Without your specifics and your rules, the most likely text is the average of everything it has read.
This is how the tool works, not a glitch you can fix by typing “make it sound human.”
Start with the simplest fact: a model predicts words
A large language model learns from an enormous amount of text. When you give it a prompt, it calculates which word (technically, which token) is most likely to come next, adds it, and repeats.
Fluency comes from that process. So does blandness.
Picture asking a thousand people to finish your sentence and always choosing the most popular answer. You get something correct and readable. You rarely get something surprising, because surprise is, by definition, unlikely.
Add the second layer: training for approval
After the initial training, chatbots are tuned with human feedback so they become helpful, polite, and safe. That tuning has a side effect. A study presented at ICLR 2024 found that reinforcement learning from human feedback significantly reduces output diversity compared with ordinary fine-tuning. In plain terms, the model develops a house style and keeps returning to it.
That house style leaves fingerprints. Researchers who analyzed more than 15 million biomedical abstracts found a sudden jump in style words such as “delves,” “showcasing,” and “underscores” after chatbots arrived. They estimated that at least 13.5% of 2024 abstracts were processed with LLMs. The tools did not just help people write. They made thousands of people write with the same vocabulary.
Then the third layer: everyone uses the same tools
When millions of people draw from a handful of models, their work converges. Three studies show it happening at three different levels:
Ideas converge. In a Science Advances experiment, writers who received story ideas from AI produced stories that evaluators rated as more creative. Yet the AI-assisted stories were more similar to each other than stories written without help.
Styles converge. A Cornell study of 118 writers in India and the US found that AI autocomplete made both groups write more alike, mostly at the expense of Indian writing styles. Some Indian participants ended up describing their own festivals and food from a Western point of view.
Opinions converge. In a CHI 2023 experiment with 1,506 participants, people who co-wrote with an opinionated AI assistant were twice as likely to write paragraphs agreeing with it. Most did not notice they were being influenced.
That last finding matters most for anyone building a personal brand. AI does not only change how you sound. Left unchecked, it can and will change what you say.
What this looks like in your feed
The effect is already visible on LinkedIn. An analysis by the detection company Originality.ai, first reported by WIRED, estimated that more than half of longer English-language posts on the platform were likely AI-generated by late 2024.
Detector numbers are estimates, and some of those posts were only polished by AI. But today, if your feed has started to read as if one person wrote all of it, you are not imagining it.
The takeaway: generic output is the default state. To get writing that sounds like you, you have to supply what the average does not contain. The rest of this guide shows you how.
What Is a Writing Voice?
“Style is a very simple matter; it is all rhythm.” ~ Virginia Woolf, in a letter to Vita Sackville-West
Your writing voice is the set of repeatable choices that make your writing recognisable: what you believe, which stories you tell, the words you use, the rhythm of your sentences, how you build an argument, and what you refuse to say.
Two people with the same facts sound different because they make different choices at each of these levels.
“Voice” sounds mystical. It is not. Psychologists and linguists have measured it for decades. James Pennebaker spent years counting how people use small “function words” such as pronouns, articles, and prepositions, and found that these barely noticed words reveal a great deal about who we are.
Forensic linguists use similar patterns to work out who wrote an anonymous text. Your voice is real and measurable, and it is mostly invisible to you. That explains why so many experts say “I don’t have a voice.” You do. You just can’t hear it, the same way you can’t hear your own accent.
The six layers of voice
Here is a working model. Each layer is something you can describe, collect examples of, and hand to an AI tool. Pay attention to the last column.
Layer | What it means | Example | Can AI imitate it from samples? |
1. Beliefs | The positions you hold and defend | “Most dashboards fail because nobody owns the decision.” | No. It can only repeat beliefs you give it. |
2. Stories | Your experiences, mistakes, and specifics | The migration that ran three months late, and why | No. It will invent plausible fakes if you let it. |
3. Vocabulary | Words you use, and words you never use | You say “messy,” never “suboptimal” | Partly, with examples and a banned list |
4. Rhythm | Sentence length, pacing, punctuation habits | Two blunt lines, then one long explanation | Partly, with good samples |
5. Structure | How you open, argue, and close | You always start with the mistake, then the lesson | Yes, with clear rules |
6. Boundaries | Topics, tones, and claims you avoid | No hype, no predictions without evidence | Yes, if written down |
The pattern is hard to miss. AI can learn your surface (layers 3 to 6) from examples and rules. It cannot generate your substance (layers 1 and 2). Substance is what readers remember, and it is what builds trust.
So any system for writing with AI has two jobs. It must give the model enough of your surface to stop it from sounding like itself. And it must keep your substance flowing in from you, every single time.
Woolf’s line is worth taking literally. Rhythm is the layer people underestimate most. A paragraph where every sentence is the same length reads like a machine wrote it, even if every word is true.
Why Does Sounding Like Yourself Matter for Your Personal Brand?
“Be yourself; everyone else is already taken.” ~ Oscar Wilde (commonly attributed)
Your voice matters because a personal brand runs on recognition and trust, and both depend on readers believing a specific person stands behind the words.
Generic writing can earn impressions. It rarely earns replies, referrals, or invitations. When everything sounds the same, sounding like a real person becomes the advantage.
Recall what a brand actually is: a shortcut people use to decide whether to trust you. The research summarized in our foundational guide breaks personal brand equity into three parts: appeal, differentiation, and recognition.
Generic AI content fails two of them. It cannot differentiate you, because it sounds like everyone. It cannot build recognition, because there is nothing distinctive to recognize.
Decision-makers are reading, and they reward original thinking
The audience for expert writing is real and paying attention. Edelman and LinkedIn surveyed nearly 2,000 business decision-makers for their 2025 B2B Thought Leadership Impact Report.
They found that “hidden buyers” (finance, legal, procurement, and IT people who shape deals without sitting in sales meetings) lean heavily on the expertise and opinions shown by a company’s executives and experts. Ninety-one percent of them said good thought leadership helps them spot challenges they had not recognized.
Read that carefully. What these readers reward is original thinking from identifiable people. That is precisely the part generic AI writing strips out.
Trust is fragile when readers sense automation
Research on AI and trust is still mixed. Some experiments, including work from UVA Darden, found that labeling content as AI-generated lowered trust, while other studies found smaller effects.
But no one has found that readers trust writing more when it sounds like a template. Sounding like yourself is the low-risk position whatever the research concludes next.
Visibility without vanity
Many technical experts avoid personal branding because it feels like self-promotion.
Writing about your own work, in your own words, is closer to documentation. You are showing what you know, how you think, and what you learned the hard way. That is generosity and not bragging.
Generic AI content | Content in your own voice |
Summarizes what everyone already knows | Shares what only you saw |
Sounds confident about everything | Shows judgment, including doubt |
Could be posted by anyone in your field | Could only be posted by you |
Collects likes from strangers | Starts conversations with the right people |
Builds the platform’s content library | Builds your reputation |
If your work is good, it deserves to be found by the people who need it. Your voice is how they will know it is you.
How Do You Make AI Write Like You: The 6-Step System
“Know thyself.” ~ Inscription at the Temple of Apollo at Delphi
To make AI write like you, work through six steps in order:
document your voice fingerprint,
build a bank of raw material,
set up an AI workspace that holds both,
draft by being interviewed instead of by prompting,
edit out AI patterns,
and fold it all into a weekly routine.
Order matters. Each step feeds the next, and skipping the first two is the reason most “write like me” prompts disappoint.
Step | What you do | What AI does | Time |
1. Voice fingerprint | Collect samples, answer voice questions | Analyzes patterns in your samples | 60 min, once |
2. Raw material bank | Capture stories, decisions, opinions | Organizes and tags them | 20 min to set up, then minutes a day |
3. AI workspace | Create a project with your guide and bank | Follows your guide in every chat | 30 min, once |
4. Interview-first drafting | Talk through your answers | Asks questions, then shapes a draft | 20 to 30 min per piece |
5. De-AI edit | Final read-aloud pass | Flags AI patterns against your list | 10 to 15 min per piece |
6. Weekly routine | Protect one block of time | Turns one session into several pieces | About 90 min a week |
Total setup is roughly two hours, done once. After that, the system runs on about 90 minutes a week.
Step 1: How Do You Find Your Writing Voice?
Find your voice by collecting writing you produced when you weren’t trying to “write,” asking AI to describe the patterns, and correcting that description until it matches how you actually sound.
The output is a written portrait of your voice that any AI tool can follow.
Gather the right samples
The best samples are things you wrote quickly, for a real person, about something you cared about. Formal reports and polished bios are usually the least “you” things you have ever written.
Collect five to ten pieces, around 1,500 to 3,000 words in total:
Slack or Teams messages where you explained something to a colleague
Emails where you disagreed with someone, politely
Thoughtful comments you left on other people’s posts
A transcript of you explaining your work out loud
Any post or article you are proud of and wrote without AI
Leave out anything heavily edited by someone else, anything AI wrote, and anything built from a corporate template.
The transcript is the most valuable sample. Record yourself for five minutes explaining a recent decision to a smart friend, then let your phone or meeting tool transcribe it. Speech captures your rhythm and word choices before your inner editor can sand them down.
As Paul Graham argues in Write Like You Talk, spoken language is easier to read, and most people already speak more clearly than they write.
Ask AI to analyze, not imitate
Paste your samples into a chat with this prompt:
PROMPT: VOICE ANALYSIS
I'm going to share samples of my own writing and speech. Do not rewrite them.
Analyze them and describe my voice so precisely that another writer could imitate it. Cover:
1. Sentence length and rhythm (averages and patterns)
2. How I open and how I close
3. Words and phrases I use often
4. Words and phrases I never use (infer from what is absent)
5. How I handle disagreement, uncertainty, and humor
6. How I use examples, numbers, and stories
7. Punctuation and formatting habits
8. Three things that make my writing different from typical professional writing
Quote short examples from my samples for each point.
If something is unclear, say so instead of guessing.
Read the result critically. The model will get some things right and flatter you on others. Correct it in plain language: “I’m rarely funny in writing.” “I say honestly too often, keep it rare.” Your corrections are the most valuable part of this step, because they contain knowledge only you have.
Answer the questions AI can’t answer for you
Write short answers to each of these. Bullets are fine.
What do I believe about my field that many peers would disagree with?
Which topics do I never want to write about?
Which words make me cringe when I see them on LinkedIn?
Which two writers do I admire, and what exactly do I admire about their tone?
How should a reader feel after reading me: smarter, challenged, reassured, amused?
Which claims will I never make without evidence?
The AI analysis plus your answers becomes your voice fingerprint. In Step 3, it turns into a one-page voice guide.
Step 2: How Do You Build a Raw Material Bank AI Can’t Invent?
“Experience is not what happens to a man; it is what a man does with what happens to him.” ~ Aldous Huxley, Texts and Pretexts
Build a raw material bank by capturing small pieces of your real work every day (decisions, mistakes, surprises, numbers, questions people ask you) in one place you can paste into AI.
This bank is the substance layer of your voice, and it is your strongest defense against generic output. If a draft contains nothing from the bank, it will sound like everyone else, however good your voice guide is.
Here is the reason. A model fills gaps with likely words. A real detail (the client who asked the same question three times, the dashboard nobody opened for six weeks) is not a likely word. It can’t be predicted. It has to come from you. Every specific you add is a spot where the model cannot slide toward the average.
What to capture
Type | Question to ask yourself | Example entry |
Decision | What did I choose this week, and what did I weigh? | Delayed the launch two weeks to fix data quality. Sales was unhappy. |
Mistake | What went wrong, and what did it teach me? | Built the model before we agreed on the metric definition. Rebuilt it twice. |
Surprise | What didn’t go the way I expected? | Executives read the one-page summary and skipped the 40-slide deck. |
Repeated question | What did someone ask me more than once? | “How do we know the AI isn’t making things up?” |
Disagreement | What did I push back on? | Told the team we didn’t need real-time data for a weekly decision. |
Number | What figure stuck with me? | Cut report build time from three days to four hours. |
Phrase | What did I say out loud that landed? | “Data quality is a product problem, not a pipeline problem.” |
How to capture it without adding work
Aim for a few minutes a day at most. Choose whichever option has the least friction for you:
Voice notes. Speaking is much faster than typing. A Stanford study found that speech recognition on phones was about three times faster than typing in English, with fewer errors. Record a 60-second note as you walk out of a meeting.
One running document. A single note titled “Raw material” with dated bullets. No folders, no formatting.
Copy what you already wrote. When you send a good Slack reply or email, paste it into the bank. You have already done the thinking.
A Friday prompt. Answer three questions every Friday: What did I decide? What did I get wrong? What did someone ask me?
Once a week, paste the raw entries into your AI workspace with this prompt:
PROMPT: ORGANIZE MY RAW MATERIAL
Below are my unedited notes from this week. Organize them into a table with columns:
Entry | Type (decision, mistake, surprise, question, disagreement, number, phrase) |
Possible post angle | What's missing (details I should add before this becomes a post).
Do not invent any details, names, or numbers.
If an entry is too vague to use, mark it "needs more".
A word on confidentiality
Your bank will hold sensitive details, so two rules apply. First, check which AI tools your employer approves and whether your plan allows your data to be used for training.
Second, generalize before you publish: swap names for roles, exact figures for ranges, and client names for industries. Our foundational guide has a full section on building authority when you can’t share confidential work.
Step 3: How Do You Set Up ChatGPT or Claude to Write in Your Voice?
“Give me six hours to chop down a tree and I will spend the first four sharpening the axe.” ~ Abraham Lincoln (commonly attributed)
Set up one dedicated AI workspace that permanently holds three things: your voice guide as instructions, your best samples as reference files, and your raw material bank.
Then every new chat starts already knowing how you sound. You stop re-explaining yourself, and drafts stop starting from the average.
Which feature to use
Tool | Feature | What it holds | Best for |
ChatGPT | Instructions, files, and chats in one place. Project instructions override your global custom instructions. | An ongoing content workspace | |
Claude | Project instructions plus a knowledge base of uploaded files | Long samples and source material | |
Claude | A custom style built from writing samples you upload, or from a description | Applying your voice in any chat | |
Most assistants | Custom assistants (such as custom GPTs or Gemini Gems) | A saved bundle of instructions and files | Sharing your setup with an editor or ghostwriter |
Features and plan limits change often, so check each tool’s help page for the current details. The principle holds regardless: put your voice somewhere the tool reads every time.
The voice guide template
This is the core document. Keep it to one page. Models follow short, specific rules better than long, vague ones. Copy the template and fill it in from your Step 1 work:
# Voice guide: [Your name]
## Who I am
[Role, field, years of experience, one line on what I'm known for]
## Who I write for
[One specific reader: their role, their problem, what they already know]
## What I believe
- [Position 1, in one sentence]
- [Position 2]
- [Position 3]
## How I sound
- Sentences: [e.g., mostly short; one longer sentence per paragraph to explain]
- Openings: [e.g., start with a specific moment or mistake, never a definition]
- Closings: [e.g., end on the practical implication, never a recap]
- Tone: [e.g., direct, dry humor, comfortable admitting uncertainty]
- Words I use: [list]
- Formatting: [e.g., no emojis, no more than 3 bullets, no bold in short posts]
## Never do this
- Words: [e.g., delve, leverage, unlock, game-changer, landscape, robust, seamless]
- Patterns: [e.g., "It's not X, it's Y"; rhetorical-question openers;
lists of three by default; em dashes; ending with "What do you think?"]
- Claims: [e.g., predictions without evidence; numbers I can't source]
## How to work with me
- Never invent stories, names, numbers, or quotes. If a draft needs a detail, ask me.
- Use only my raw material bank and what I say in this chat.
- When unsure how I'd phrase something, ask instead of guessing.
- Flag anything that sounds like generic professional writing.
## Reference samples
See attached files: [file names]
Three parts of the template do most of the work:
“Never invent stories, names, numbers, or quotes.” This prevents the most damaging failure of all: a fabricated anecdote published under your name.
“Ask instead of guessing.” This turns the model from an author into an interviewer, which is the shift Step 4 depends on.
The banned list. Much of the recognizable “AI voice” comes from a small set of habits, so naming them explicitly works surprisingly well. Wikipedia’s volunteer editors maintain a detailed field guide to signs of AI writing that makes a good starting list.
What to upload to your workspace
Your filled-in voice guide
Three to five of your best samples from Step 1
Your raw material bank, refreshed weekly
Your known-for statement and content pillars, if you have them (the personal brand guide and the free Authority Framework walk you through both)
Optional: a “posts that worked” file with your three pieces that drew the most meaningful replies
Then test it. Ask the workspace for a short post built on one entry from your bank, and read the result aloud. Wherever it sounds wrong, fix the guide as well as the draft, so the same mistake doesn’t return next week. After three or four rounds, the guide settles down.
Step 4: What Is the Best AI Writing Workflow for Sounding Like Yourself?
“I write entirely to find out what I’m thinking, what I’m looking at, what I see and what it means.” ~ Joan Didion, Why I Write
The best workflow is interview-first: AI asks you questions, you answer out loud or in rough notes, and only then does AI shape your answers into a draft. Your thinking comes first and the model’s words come last.
This flips the common “prompt, accept, post” habit, and it is the biggest single lever for sounding like yourself.
Why it works: when you type “write a post about data quality,” the model has to supply the ideas, the examples, and the words, so it supplies the most likely ones. When the model interviews you, you supply the ideas and examples, and the model only arranges them.
The co-writing study from earlier shows why this order matters. When AI proposes the content, it nudges what you think. When you propose the content, there is far less for it to nudge.
The four-stage workflow
Interview (10 minutes). AI asks you five to seven questions about one topic from your bank. You answer by voice or rough typing. Ramble. Don’t edit.
Shape (5 minutes). AI organizes your answers into a draft, using only what you said and following your voice guide.
Challenge (5 minutes). AI critiques the draft against your banned list and points to claims that need evidence.
Own (10 minutes). You edit the final version yourself and read it aloud. Nothing publishes without this step.
Prompt 1: The interviewer
PROMPT: INTERVIEWER
Act as an interviewer helping me write a post. Topic: [topic or raw material entry].
Ask me one question at a time, up to seven questions.
Start broad, then push for specifics: the exact moment, what I decided,
what I got wrong, what I'd tell someone else, and where I disagree with common advice.
If an answer is vague, ask a follow-up that forces a concrete example or number.
Do not suggest answers or opinions. Do not draft anything yet.
Prompt 2: The shaper
PROMPT: SHAPER
Using only my answers above, draft a [LinkedIn post / newsletter section / article]
of about [length] words. Follow my voice guide exactly.
Keep my phrasing wherever it works, and quote me directly where possible.
Do not add any facts, stories, numbers, or opinions I didn't give you.
Where the draft needs something I didn't provide, insert [ASK ME: what's needed]
instead of inventing it. Open with the most specific moment from my answers.
Prompt 3: The challenger
PROMPT: CHALLENGER
Review the draft against my voice guide. List:
1. Every phrase that sounds like generic AI or corporate writing,
with a plainer alternative in my style
2. Any sentence I wouldn't say out loud to a colleague
3. Any claim that needs evidence or a caveat
4. Places where the rhythm is too even (several similar-length sentences in a row)
5. Anything that could have been written by someone else in my field
Do not rewrite the draft. Give me the list so I can decide.
Notice that the challenger returns a list, not a new draft. That keeps the final decisions with you. If you let the model rewrite again, it often brings back the very patterns you just removed.
Turn one interview into several pieces
A good interview usually holds more than one idea. After shaping your first draft, run this:
PROMPT: IDEA SPLITTER
From my interview answers, list every distinct idea that could stand alone
as a short post. For each, give a one-line angle and the exact quote from
my answers it is built on. Do not draft them yet.
This is how one 20-minute conversation becomes two to four weeks of posts without a single invented sentence. It is the same batching principle professional ghostwriters use: extract once, publish many times.
Step 5: How Do You Make AI-Assisted Writing Sound Human?
“The secret of good writing is to strip every sentence to its cleanest components.” ~ William Zinsser, On Writing Well
Make AI-assisted writing sound human by removing the specific patterns that mark text as machine-made (inflated vocabulary, formulaic structures, even rhythm, empty summaries) and then reading the piece aloud.
If you wouldn’t say a sentence to a colleague, rewrite it. Most of what people call “AI voice” comes from a small, learnable set of habits.
One caution first. No single pattern proves anything. Wikipedia’s guide stresses that its list is a set of observations, not rules, because models learned these habits from human writers. The editors also warn against relying on AI detection tools, which have high error rates.
The goal is not to fool a detector. The goal is to sound like you.
The most common AI tells, and how to fix them
AI tell | Example | Fix |
Inflated vocabulary | “delve,” “underscore,” “pivotal,” “landscape,” “tapestry” | Use the word you’d say out loud: “look at,” “shows,” “important,” “field” |
Grand significance | “stands as a testament to,” “plays a vital role” | State the actual effect, with a number or example |
The contrast formula | “It’s not just X, it’s Y.” | Make the claim directly |
Default lists of three | “faster, smarter, and more efficient” | Use the number of items you actually have |
Even rhythm | Five sentences of similar length in a row | Mix very short sentences with longer ones |
Empty openers | “In today’s fast-paced world…” | Open with a specific moment, claim, or number |
Recap endings | “In conclusion…” or restating every point | End on the implication, or simply stop |
Hedging everything | “may potentially help to some extent” | Commit where you’re sure; hedge only where you’re not |
Engagement bait | “What do you think? Comment below!” | Ask a real question only when you want real answers |
Punctuation tics | Em dashes everywhere, a colon in every title | Use the punctuation you use in email |
Stacked transitions | “Moreover,” “Furthermore,” “Additionally” | Cut them; good sentences connect on their own |
Invented specifics | A client story you don’t remember | Delete it and replace it with a real entry from your bank |
The last row matters most. A stylistic tell makes you sound generic. An invented specific makes you sound dishonest, and it can damage trust in everything else you have written.
The 10-minute de-AI editing pass
Run this checklist on every piece before it goes out:
Specific test: It contains at least one detail only I would know.
Read-aloud test: I read it out loud, and no sentence made me wince.
Banned-list test: I searched the draft for my banned words and patterns.
Rhythm test: There are at least two very short sentences, and no run of same-length sentences.
Swap test: A peer could not post this unchanged.
Claim test: I can defend every claim and number if someone replies “source?”
Opening test: The first line says something specific instead of setting a scene.
Ending test: It ends on a point, not a recap.
The read-aloud test is the cheapest voice check that exists. Paul Graham reads every essay aloud before publishing and fixes anything that doesn’t sound like conversation. When you write something you wouldn’t say, you can hear it.
Want a printable version? The free Authority Framework includes a self-audit for spotting template-sounding content, plus frameworks for finding your point of view before you write a word.
Step 6: How Can Busy Professionals Post Consistently With AI? A 90-Minute Weekly System
“Inspiration is for amateurs. The rest of us just show up and get to work.” ~ Chuck Close
Busy professionals can publish consistently by protecting one 60-minute session and two 15-minute check-ins each week, and by producing everything from a single interview.
That is enough for two or three short posts a week, or one long-form piece every two weeks. Consistency comes from a system you can keep during a bad week, not from motivation.
The math is simple. Most professionals can’t find five hours a week for content. Almost anyone can find 90 minutes. The interview-first workflow makes those minutes count, because talking is faster than typing and one conversation contains several ideas.
The weekly schedule
When | Time | What you do | What AI does |
Daily, in passing | 1 to 5 min | Drop voice notes or bullets into your bank | Nothing yet |
Monday | 15 min | Paste last week’s notes and pick one topic | Organises notes, suggests angles from your material |
Wednesday | 60 min | Interview (20), review drafts (20), final edit and read-aloud (20) | Interviews you, shapes two or three drafts, runs the challenger |
Friday | 15 min | Schedule posts, reply to comments, note what drew real replies | Suggests reply points you rewrite, summarizes what resonated |
The total is about 90 minutes, plus a few minutes a day of capture.
Repurpose without repeating yourself
One interview can feed several formats. The trick is that each piece expresses a different idea from your answers, so nothing feels recycled.
Short post: one story and one lesson from a single answer
Carousel or checklist: the steps or criteria you described
Newsletter section: the full story, with context and caveats
Comment on a peer’s post: your disagreement, in two sentences
Long-form article: three or four interviews on one theme, combined once a month
Monthly and quarterly upkeep
Monthly (30 minutes): Add any new banned phrases you caught to your voice guide. Add your best-performing piece to your samples file.
Quarterly (30 minutes): Read your last ten pieces in one sitting and ask whether a stranger would recognise them as one person. Then run the signal review from the foundational guide: replies, DMs, invitations, and referrals, not likes.
Protect the system in a bad week
You will have weeks with no time at all. Plan for them now:
Keep a reserve of two finished pieces you can publish any time.
Define a minimum week: one thoughtful comment on a peer’s post counts.
Cut frequency before quality. One real post every two weeks beats three generic ones.
This is consistency without burnout: a pace you can hold for a year, built on a process that carries most of the load for you.
How Does This Change for Data and AI Leaders, Founders, and Executives?
“The great enemy of communication, we find, is the illusion of it.” ~ William H. Whyte, Fortune (1950), often misattributed to George Bernard Shaw
The six steps stay the same for every role. What changes is the main threat to your voice: jargon for data leaders, hype for AI leaders, pitch-deck language for founders, and confidentiality for executives.
Tune your voice guide and your challenger prompt to target that one threat.
Role | Biggest voice risk | Add to your voice guide | Best raw material |
Data leader | Buzzwords (“data-driven,” “single source of truth”) and jargon that loses business readers | “Explain every technical term in plain language the first time. Tie every insight to a business decision.” | Decisions your analysis changed; dashboards nobody used, and why |
AI leader | Hype and overclaiming in a field crowded with both | “Every claim must point to something I built or tested. Include at least one limitation.” | What you shipped, what broke in production, what you’d warn peers about |
Founder | Startup clichés (“we’re building the future of X”) | “Write as if talking to one customer, not an investor. No category claims.” | Customer conversations, product calls, honest setbacks |
Executive | Confidentiality limits that push content toward vague and generic | “Speak in principles and patterns. Use roles instead of names and ranges instead of exact figures.” | Lessons across many teams, decision criteria, what you’d do differently |
Data leaders: translate depth into decisions
Your expertise is often invisible because it is technical. Add one line to the challenger prompt: “Flag every sentence a CFO would need explained.”
The goal is to keep the depth and lose the jargon, so your name becomes shorthand for judgment, not just output. See more on positioning for this audience at Influence Theory for Data & AI Leaders.
AI leaders: let proof do the talking
You work in the field where generic AI commentary is most common. That makes a clearly human voice, grounded in shipped work, unusually credible.
Add a proof check to every draft: what did I build, test, or measure that supports this? Where you can’t answer, cut the claim or label it as opinion.
Founders: your voice is the company’s voice
Early on, people trust a person before they trust a startup. Use customer conversations as raw material, with permission and without identifying details.
Your specific customers and decisions are the one thing no competitor can copy, and they make better content than any positioning statement.
Executives: visibility that respects discretion
You may have the richest experience and the least freedom to share it. Fill your bank with patterns rather than specifics: “In three of the last four transformations I led, the first thing to fail was…”
Ask your communications or legal team to review the “Never do this” section of your voice guide once, so every draft starts out compliant. Your brand should also outlast any one employer, so keep your samples, bank, and guide in a place you own.
How Do You Know Your AI-Assisted Writing Still Sounds Like You?
“Honesty is the first chapter in the book of wisdom.” ~ Thomas Jefferson
You know it still sounds like you when people who know you say so, when strangers can predict what you would say next, and when your writing draws replies and conversations instead of polite likes. Measure voice with people instead of using AI detectors.
Four fun tests
The colleague test. Send three recent pieces to two people who know you well. Ask: “Does this sound like me? Which line sounds least like me?” Whatever line they pick goes to your ban list.
The blind test. Mix two of your posts with two from peers on similar topics, remove the names, and ask a friend to spot yours. If they can’t, your substance layer is too thin. Add more from your bank.
The reply test. Track replies and DMs from the audience you actually want. Specific, voice-driven writing tends to start conversations. Generic writing tends to collect quiet approval.
The drift test. Each quarter, put a recent piece next to one of your original Step 1 samples. If your writing has drifted toward the model’s house style, refresh your samples and tighten the guide.
Should you disclose that you use AI?
There is no universal rule yet, and research on how disclosure affects trust is still mixed. Here is a practical, honest standard you can adopt today:
You are the author when the ideas, stories, and judgments are yours and you edited the final version. AI is then a tool, much like an editor or a spell checker.
Disclose when your employer, a platform, or a publication requires it, or when AI produced a substantial part of the content itself, such as an image or a summary you did not write.
Never publish a story, quote, statistic, or claim you cannot personally stand behind, whatever tool helped you phrase it.
If you follow the interview-first system, that third rule largely takes care of itself, because every substantive sentence starts with you.
Frequently Asked Questions
“The important thing is not to stop questioning.” ~ Albert Einstein
Can AI really learn to write like me?
AI can learn your surface style (vocabulary, rhythm, structure, formatting) from good samples and a clear voice guide. It cannot learn your experiences or opinions unless you give them to it. The best results come from pairing a voice guide with your own raw material in every draft.
How many writing samples does AI need to match my voice?
Quality matters more than quantity. Five to ten samples, around 1,500 to 3,000 words in total, usually give a model enough to describe your patterns. Include at least one transcript of you speaking, because speech carries your natural rhythm.
What is the best prompt to make ChatGPT or Claude sound like me?
There is no single magic prompt. The most reliable setup is a project or custom style that holds your voice guide and samples, plus a standing instruction to ask instead of invent. Then use the interviewer, shaper, and challenger prompts from this guide, in that order.
Will people be able to tell I used AI?
If you publish raw AI drafts, many readers will notice, because the patterns are now familiar. If AI only interviews you, organizes your answers, and flags problems while you make the final edit, your writing reads as yours, because it is.
Do AI detectors work on AI-assisted writing?
Not reliably. Even Wikipedia’s AI cleanup volunteers tell editors not to rely on detection tools alone, because of high error rates. Aim to sound like yourself, not to beat a detector.
Is it okay to use AI for LinkedIn posts?
Yes, as long as you remain accountable for every idea and claim. Use AI as an interviewer and editor rather than as the author, and don’t publish anything you couldn’t explain or defend in conversation.
How long does it take to set up an AI writing workflow?
About two hours, once: an hour for your voice fingerprint, 30 minutes to set up the workspace, and 30 minutes to test and adjust. After that, plan on roughly 90 minutes a week.
What if I don’t think I have a voice?
You do. Everyone has consistent patterns in how they explain things, and language researchers have measured those patterns for decades. You just can’t hear your own voice easily. Record yourself explaining your work and let the transcript show you.
Should I use the same voice on every platform?
Keep the same voice and adjust the register. You might be slightly more formal in a long article than in a short post, the way you speak differently in a board meeting and over lunch. Your beliefs, stories, and boundaries should stay the same everywhere.
Your Next Step
“If there’s a book that you want to read, but it hasn’t been written yet, then you must write it.” ~ Toni Morrison
Sounding like yourself with AI comes down to one rule: you supply the substance and the final judgment, and AI supplies the speed in between. Every step in this guide exists to make that rule easy to follow, even in your busiest week.
The whole system, in one view:
Collect samples and build your voice fingerprint.
Capture raw material from your real work, a few minutes a day.
Set up one AI workspace with your voice guide, samples, and bank.
Draft by being interviewed, not by prompting.
Edit out the AI tells, then read it aloud.
Protect 90 minutes a week.
Start today with one small action: record a five-minute voice note explaining a recent decision at work, transcribe it, and paste it into the voice analysis prompt from Step 1. You will see your voice described on the page, possibly for the first time.
The people who need your expertise are out there, reading. Your work deserves to reach them in your own words.
Want help going further?
Get the free Authority Framework for self-assessments, point-of-view templates, and a content audit built for experts.
Or explore hands-on support, from positioning sessions to interview-based ghostwriting built on your own voice, at influencetheory.org.
Keep reading:
Learn more about Influence Theory, a personal branding service, built dedicatedly for data and AI leaders.
Comments