You ask a tool to draft an email, and what comes back is fine. Polished, even. It just doesn't sound like you. So you rewrite half of it, and tomorrow you do the same thing again on the next draft. The fix is a small one, and you do it once: teach the tool your voice, save it somewhere it stays, and every draft after that starts much closer to how you write. This is one of the first wins in The 7 Levels of AI Proficiency, the point where you stop steering every sentence and start designing the standing context the tool works from.
What training AI to sound like you means
You are not retraining the model. You are giving it two things it can use on every draft: a short set of rules that describe your voice, and a few real samples of your writing to mimic. Together they tell the tool what "you" sounds like before it writes a word.
The rules are plain instructions. Short sentences. No insider terms unless you define them in parentheses. Warm but direct. Contractions are fine. No exclamation points. The samples are things you wrote, pasted or uploaded so the tool can match the rhythm and word choice it sees there.
The piece beginners tend to miss is the third one: a place to save all of it so it carries over. Most tools have a persistent setting for standing instructions. Once it is set, you don't re-paste your voice into every new chat. You set it once, and future drafts already have it.
When this is worth doing
This pays off any time you write the same kind of thing over and over in a recognizable voice. Good first candidates:
- Emails and replies you send all day
- LinkedIn posts, newsletters, or other regular content
- Proposals, quotes, or client notes that need to sound like your firm
- Anything a teammate or assistant might draft on your behalf
If you only ask a tool for one-off facts or quick research, you may not need this.
The setup is worth about fifteen minutes for most people, and it earns that back the first time the output carries your name and stays consistent without re-explaining yourself.
The method: three parts that work across tools
The exact buttons differ by tool and the buttons change over time. The method underneath stays the same.
-
Describe your voice in specific rules
"Write clearly" gives a tool almost nothing to aim at. Specific rules give it a target: keep sentences short, define any technical term in parentheses, stay warm but direct, use contractions, skip exclamation points. Five concrete rules beat one vague adjective every time.
-
Give it two or three writing samples
Paste or upload a few things you genuinely wrote. Samples teach voice better than describing it, because the tool can see your sentence length, your word choices, and your rhythm instead of guessing from labels. Pick samples that sound the way you want future drafts to sound.
-
Save it where it persists
This is the step that turns a one-time effort into a permanent setup. The path depends on your tool. In ChatGPT, that home is Custom Instructions, under Settings, then Personalization. Set it once and it applies across your chats without re-typing, and it is available on all ChatGPT plans. In Claude, you have two good options: for a general voice, use Instructions for Claude so the guidance applies across your conversations; for a specific workflow, client, or content system, use a Project and set the project instructions, which Claude uses for chats inside that project. Many major tools now have some version of this, whether it is called saved instructions, projects, memory, or skills. The labels differ; the habit is the same. The point is the same everywhere: write your voice down once, in the spot the tool reads every time.
Do it now
Here is the fastest way to build your voice profile. Open a fresh chat, paste the prompt below, and paste two or three of your own writing samples right under it. Ask for the profile first, then save what it gives you.
I want you to learn my writing voice from the samples below, then write me a reusable voice profile I can save and reuse. Here are 2-3 things I wrote: [paste sample 1] [paste sample 2] [paste sample 3] Do three things, in this order: 1. Infer my voice rules from the samples. List them as short, specific rules (sentence length, tone, words or punctuation I avoid, formatting habits). 2. Write those rules back to me as a clean "voice profile" I can paste into a tool's standing-instructions setting. Keep it tight. 3. Then rewrite this test paragraph in my voice, matching the samples as closely as you can: "Thanks for the note. I took a look at the proposal and I have a few thoughts before we move forward." Match the samples as closely as you can. This is guidance, not a script.
You will get a set of rules, a ready-to-save profile, and a test rewrite so you can sanity-check the result. Read the rewrite aloud. If it sounds like you, save the profile in your tool's persistent setting. If something is off, tell the tool which rule to adjust and run it again. A couple of rounds gets it close.
One privacy note before you paste: leave out client secrets, private financial or medical details, and anything you would not want stored in the tool. Use clean samples that show your voice without exposing sensitive facts. If this is for work, check your company's AI policy and data settings first.
A starter file to download
To make this even faster, grab the companion template. It is a fill-in-the-blank voice profile with labeled sections (who you are and what you do, your voice in five rules, words and phrases you never use, your format defaults, and a short writing sample), a one-line prompt under each, and a filled-in worked example so you can see what "good" looks like. Rename it, fill in your own answers, and paste it into your tool's standing-instructions setting.
Voice profile starter (.md)
A labeled fill-in-the-blank template with a worked example. No signup.
Common mistakes you can step around
A few small things trip people up, and all three are easy to avoid.
- Vague rules. Be professional or sound friendly gives the tool too much room. Trade each adjective for a concrete instruction, like keeping sentences short or defining any technical term in parentheses.
- Skipping the samples. Rules alone get you partway. The samples are what make a draft sound like you, because the tool matches what it can see. Even two short ones change the result.
- Re-pasting your voice every chat. If you find yourself typing your rules into each new conversation, you have skipped the save step. Move the profile into the persistent setting once, and the work stops repeating.
One honest note: rules plus samples get you much closer, and you will still want a quick read before anything important goes out. The tool gives you a strong first draft in your voice. You stay the editor.
Where this fits in the bigger picture
Teaching a tool your voice is a small example of a skill that scales a long way. You are designing the standing context the tool inherits before it does any work. That is the same move as setting up a project with its own instructions, or writing a standing brief a whole team of tools reads from. The 7 Levels of AI Proficiency tracks exactly this change: early on, you tell the tool what to do one request at a time, and a little further along, you build a reusable setup that handles the request the same way every time without you steering each word.
Your voice profile is your first reusable setup. Once it is saved, every email, post, and proposal starts from a draft that already sounds like you, and you spend your time on judgment instead of cleanup. From there, the same habit carries into bigger jobs: a project that knows your clients, a brief that knows your brand, a workflow that runs while you do something else.
Want to see where you stand today and what the next step looks like? The 7 Levels of AI Proficiency assessment takes about ten minutes and shows you the climb.
Related reading: Level 4: The Commander.
Sources
- OpenAI Help Center, "ChatGPT Custom Instructions," help.openai.com (Accessed June 23, 2026).
- Claude Help Center, "How can I create and manage projects?", support.claude.com (Accessed June 23, 2026).
- Claude Help Center, "Understanding Claude's personalization features," support.claude.com (Accessed June 23, 2026).
- The 7 Levels of AI Proficiency assessment, assess.launchready.ai (LaunchReady.ai).
Frequently Asked Questions
What does it mean to train AI to sound like you?
It means giving a tool a short set of rules that describe your voice plus two or three real writing samples to mimic, then saving both in a place the tool reads on every draft. You are not retraining the model. You are setting the standing context it writes from.
Do I need to know how to code?
No. You write a few plain-language rules, paste in some of your own writing, and save it in a settings menu. There is no code involved.
Where do I save my voice so I don't retype it every time?
In your tool's persistent setting. In ChatGPT, that is Custom Instructions under Settings, then Personalization, applied across your chats and available on all plans. In Claude, save general voice guidance in Instructions for Claude so it applies across your conversations, or set instructions inside a Project when the voice belongs to a specific workflow. Many major tools now have a similar place, whether it is called saved instructions, projects, memory, or skills.
How many writing samples should I give it?
Two or three is plenty to start. Pick ones that already sound the way you want future drafts to sound. You can always add or swap samples later if the results drift.
Will it match my voice perfectly?
It gets much closer with rules plus samples, and it is still guidance rather than a guarantee. Treat the output as a strong first draft and give it a quick read before anything important goes out.
Does this only work with one tool?
No. The method is the same across tools: specific rules, a few samples, and a persistent place to save them. The menus and labels differ, and the buttons change over time, so the durable skill is the approach, not any one screen.
Find your AI Proficiency level
The free 7 Levels assessment places you across seven stages of AI capability. Under ten minutes. Research-backed scoring.