To get an AI to write in your voice, give it three things: a few real samples of your own writing, a short description of what makes those samples sound like you, and instructions phrased as what to do rather than what to avoid. Anthropic, OpenAI and Google all document ways to steer tone, but none of them promises a perfect clone of a person's voice. This article gives you a repeatable method, the Voice Brief, and a way to test whether it worked.
The question here is narrow: how do you make AI drafts sound like you instead of like generic AI? It covers what each company's own documentation says about style, our original five-part Voice Brief, a worked example, and a drift test you can run in a few minutes. If your prompts fail for reasons beyond tone, start with why ChatGPT prompts fail and the CLEAR framework.
The Short Answer
Paste two to four samples of writing you are proud of, describe your voice in plain positive statements ("short sentences, direct, one concrete example per point"), name the audience and purpose, and ask the AI to check its own draft against your samples before showing it to you. This is our practical reading of vendor guidance, not a guaranteed result. Anthropic's documentation says examples are "one of the most reliable ways to steer Claude's output format, tone, and structure," and recommends "3–5 examples for best results" when you use them inside a prompt. Treat the number of samples as a starting point and test it on your own writing.
What the Major AI Tools Document About Style (Checked Live on October 9, 2026)
| Source | What it documents |
|---|---|
| Anthropic help center (Claude styles) | To create a custom style, choose "Add Writing Example" and upload a file or paste text; the help article says you can upload "pdf, doc, and txt files," and Claude "analyzes your writing to generate a matching style." There is also a "Describe style instead" option, and presets named Normal, Concise, Formal and Explanatory. |
| Anthropic, Prompting best practices | "Tell Claude what to do instead of what not to do." Also: the formatting style of your prompt "may influence Claude’s response style," and setting a role in the system prompt "focuses Claude’s behavior and tone." |
| OpenAI help center (ChatGPT personality) | A personality is "the style and tone ChatGPT uses when responding to you." There are seven options (Default, Professional, Friendly, Candid, Quirky, Efficient, Cynical). A saved memory that conflicts with a personality "may override or reduce the visible traits of that personality." For requested outputs such as emails or social posts, tone follows your instructions and context rather than the selected personality. |
| Google, Gemini API prompting strategies | "Few-shot prompts are often used to regulate the formatting, phrasing, scoping, or general patterning of model responses." The suggested system-instruction template includes a tone constraint ("Tone: [Specify Formal/Casual/Technical]"). |
Two things are worth noticing. First, the documented features differ: Claude's help center describes learning a style from uploaded writing samples, while the ChatGPT article we checked describes selecting a preset personality. We did not find a statement in the ChatGPT article about learning a style from your samples, so with ChatGPT the practical route is to put your samples and style notes into your custom instructions or into the conversation itself. Second, no source we checked claims that any of these features reproduces a specific person's voice exactly. Treat them as steering tools.
The Voice Brief (Original Framework)
The documentation tells you that examples and clear, positive instructions steer tone. It does not give you a checklist for what to hand the AI. The Voice Brief below is our own method for turning those principles into one reusable block of text. It is an interpretation, not a statement from Anthropic, OpenAI or Google.
| Part | What goes in it | Why it is there |
|---|---|---|
| 1. Samples | Two to four pieces of your own writing, varied in topic, each labeled and kept separate from your instructions. | Examples are the documented way to steer tone; variety stops the AI from copying one topic. |
| 2. Traits | Three to five plain statements about how your writing sounds: sentence length, how you open, how formal you are. | Gives the AI words for what it should notice in the samples. |
| 3. Do-rules | Instructions phrased positively ("open with the main point", "use one concrete example per section"). | Anthropic recommends telling the model what to do rather than what not to do. |
| 4. Reader and job | Who will read this, what they should do afterwards, and one line on why. | Tone depends on the audience; a stated purpose narrows the choices. |
| 5. Self-check | Ask the AI to compare its draft with your samples and fix any mismatch before showing you. | Our own addition. It is not documented by the vendors, so test whether it helps you. |
Part 1 is the same idea we covered in our guide to few-shot prompting: samples work best when they are relevant, different from each other and clearly separated from your instructions. The same risk applies too. If all your samples start with the same sentence, expect the draft to start with it as well.
Why "Don't Sound Like AI" Usually Fails
A common instruction is "don't sound robotic" or "avoid corporate language." It names what you dislike but gives the model nothing to aim at. Anthropic's guidance makes the same point for formatting: instead of "Do not use markdown in your response," it suggests "Your response should be composed of smoothly flowing prose paragraphs."
Applied to voice, "don't be wordy" becomes "keep sentences under 20 words." The number 20 is our illustration; adjust it to your own samples. Anthropic also notes that the formatting style of your prompt may influence the style of the response, so writing your brief in the plain voice you want back is a reasonable habit. That is a documented tendency, not a guarantee.
Where to Put the Brief
| Your situation | Where the brief goes | Note |
|---|---|---|
| One voice for almost everything | Saved style or standing instructions in your AI tool | Keeps you from re-pasting. See system prompts vs. regular prompts. |
| Different voice per task (client email vs. newsletter) | A separate brief pasted at the start of each task | Each task gets its own samples. |
| ChatGPT with saved memories | Custom instructions or the conversation, after checking your memories | OpenAI says conflicting memories can override a personality. |
| Using an API | System instruction plus examples formatted identically | Google advises keeping example formatting the same. |
Once a brief works, save it. A brief is just another reusable prompt, and our guide to building a personal prompt library shows a simple way to store it with the task it belongs to.
Worked Example: A Freelance Designer's Client Update
The example below is an illustration we wrote for this article, not a tested benchmark. A freelance designer wants client status emails that sound like her, not like a template.
- Samples: three real emails she sent to different clients, pasted under the labels "Sample 1," "Sample 2" and "Sample 3."
- Traits: "Short paragraphs. Friendly but direct. I give the status in the first sentence."
- Do-rules: "Start with the status. End with the next step and a date. Keep it under 120 words."
- Reader and job: "A busy client who needs to know whether to approve or wait."
- Self-check: "Compare your draft with the three samples. If it sounds different, revise once, then show me the final version."
She then supplies the facts for this week's update separately, so the AI is not tempted to invent details. The brief controls how the email sounds; her notes control what it says.
The Voice Drift Test
A brief that works on the first draft may drift over a long conversation. This five-step test is our own method for checking.
- Write a short piece with the AI using your brief.
- Without looking at the AI's version, write the same piece yourself in a few lines.
- Compare opening lines, sentence length and any phrases you would never use.
- Add the mismatch to your Traits or Do-rules in plain positive wording.
- Repeat after a long conversation to see whether the voice holds.
If you use AI for planning content as well, the same test fits into the workflow described in how to build an AI content calendar that does not sound robotic.
Limits You Should Know
In our opinion, a brief gets you closer, not all the way. Expect to edit, especially for personal or high-stakes writing. Be careful with what you paste: samples may contain client names or private details, and we have not reviewed any provider's data-handling terms, so check the settings and policies of the tool you use before pasting sensitive text. Finally, vendor features change, so confirm current behavior on the linked pages.
Frequently Asked Questions
How many writing samples should I use? Anthropic's prompting guidance recommends 3 to 5 examples when using examples in a prompt. Two to four samples of your own writing is a sensible start; test whether more or fewer improves the result for you.
Can I describe my style instead of pasting samples? Claude's help center lists a "Describe style instead" option. In any tool, a short list of traits and positive rules is a reasonable substitute, though samples usually show more than a description does.
Will the AI still add phrases that are not mine? It can. Run the drift test, add what you find to your Traits, and edit the final draft yourself.
Does this work for every kind of content? Not necessarily. OpenAI notes that for requested outputs such as emails or social posts, tone follows your instructions and context, which is why a brief written for the specific task matters.
Want a Head Start on the Prompts Behind Your Brief?
AI Prompt Library PRO gives you 500 ready-to-use prompts across common business tasks, so you can spend your time adding your own voice instead of writing every prompt from zero.
Get AI Prompt Library PRO — $27Sources: Anthropic Help Center, configuring and using styles (writing examples, file types, describe style, presets); Anthropic, "Prompting best practices" (examples steer tone, 3–5 examples, positive instructions, prompt style influences response style, role focuses tone); OpenAI Help Center, customizing your ChatGPT personality (definition, seven options, memories may override, tone for requested outputs); Google, Gemini API "Prompt design strategies" (few-shot regulates phrasing, tone constraint). All sources checked live on October 9, 2026. The Voice Brief, the placement table, the drift test and the designer example are our own interpretation and illustration, not statements from Anthropic, OpenAI or Google. Vendor guidance changes; check the linked pages for current wording. Last reviewed: October 2026.