An AI-written draft can sound polished and still feel wrong. It may use words you never use, make the same tidy three-point argument as every other post, or end with an unnecessary question. Keeping your voice requires more than asking a model to “sound human.” Give it examples of your actual writing, a clear view of your intended argument, and a few instructions you are willing to enforce when editing.
Choose samples that represent your normal writing
Start with posts you wrote and would publish again. Select examples that show how you explain ideas, introduce a disagreement, and move from evidence to a conclusion. Avoid supplying only your most formal essay if you are trying to draft a short LinkedIn post: the model needs examples appropriate to the destination.
Repurai lets you provide writing samples and maintain a separate voice for different platforms. This is useful when your newsletter has room for a patient explanation but your social posts use shorter sentences. Your samples should reflect that difference rather than asking one style to cover every context.
Write a short list of instructions that matter
Examples show the pattern; instructions make your preferences explicit. Choose rules that have an observable result. “Use British English,” “avoid emoji,” and “explain technical terms before using abbreviations” are easier to apply than “make it engaging.” Add instructions about claims too, such as avoiding unsupported percentages or invented personal stories.
Do not write an enormous rulebook. A handful of meaningful preferences is easier to review than dozens of competing instructions. If a rule repeatedly requires you to undo good writing, revise the rule. The aim is a useful draft, not perfect obedience to a style checklist.
Give the draft your argument, not just a topic
A source title tells the model what happened. Your angle tells it why the reader should care. Before generating a post, write down the point you want to make, the evidence available, and the limit you want to acknowledge. This is often the most important contribution you make to the draft.
For example, instead of “write about a new coding assistant,” choose “explain why our team would test a coding assistant on maintenance work before using it for a new project.” You now have a decision and an audience. The draft can organize your reasoning rather than filling the page with general enthusiasm about the tool.
Edit for meaning before polishing the tone
Read the generated draft once for its argument. Does it say what you actually believe? Does each claim follow from the supplied sources or your own experience? Remove statements that attribute an experience to you that you have not had. A writing sample should teach style, not authorize the model to invent a new personal story.
Then edit for language. Replace phrases you would not say, shorten an opening that takes too long to reach the point, and keep the useful detail that a reader could act on. Read the post aloud if the rhythm feels artificial. A draft that sounds like you but misrepresents the evidence still needs revision.
Use a repeatable final check
Before publishing, ask whether the post has one clear point, a concrete example, a supported claim, and an ending that fits the argument. Compare it with recent posts to avoid repeating the same introduction or angle. Repurai's drafting workflow can help you create and revise platform-specific drafts, but your judgment decides whether a draft deserves to be published.
Start with one platform, a few representative samples, and a small set of rules. Update them when your writing changes. You can try Repurai's research-to-draft workflow and check current feature availability before choosing the setup that fits your work.