Make AI Write Like You with the Right Writing Samples

A hypothetical writer has an old project update: “Quick update. The first pass is ready. Two rough edges left; I’ll fix those tomorrow.” They want a new reply about a guide. The guide still needs one diagram reviewed, and there is no confirmed completion date.

The old message is useful as a voice reference. It is dangerous as a source of facts. A draft that borrows “I’ll fix those tomorrow” may sound familiar while creating a commitment the writer never made.

Choose a reference that resembles the new job

Start with writing you produced and would be comfortable sending again. For a short work reply, choose another short work reply. An opinion essay may reveal your vocabulary but offers little evidence about how you acknowledge a request or end an email.

Look for choices you can identify in the actual text: whether you open with the answer, how much context you add, sentence length, contractions and how you close. “Warm and professional” leaves these decisions open. “Answer first, use short sentences, omit a sign-off in an ongoing chat” gives a reviewer something to check.

One message can also contain a temporary mood or an unusual phrase. If the first reference leaves an important habit unclear, add another from similar work. More material creates more to inspect; it does not establish a universal number of samples that will work for every tool.

Give style and facts different places in the brief

The Make AI Sound Like You prompt has a separate voice-reference field. For this walkthrough, the inputs are:

Voice reference: “Quick update. The first pass is ready. Two rough edges left; I’ll fix those tomorrow.”

New facts: The guide has been drafted. One diagram needs review. There is no confirmed completion date.

Task: Reply to a teammate asking whether the guide is ready to share.

An authored revision that follows the brief is:

Quick update. The guide is drafted. One diagram still needs review, so it isn’t ready to share yet. I don’t have a confirmed date.

The opening and clipped rhythm come from the reference. The guide, diagram and missing date come from the new facts. The draft answers the teammate’s question rather than reproducing the old update.

Compare that with “Quick update. The guide is ready. One rough edge left; I’ll fix it tomorrow.” It borrows more of the original wording, but changes readiness and invents timing. A closer surface resemblance can be a worse reply.

These are editorial comparisons written to make the boundary inspectable, rather than recorded outputs from a model experiment.

Review the new draft in two passes

First check the commitments. Is the guide ready to share? Is any timing confirmed? Did the draft add a reason for the delay? Compare the answers with the new brief, not the voice sample.

Then check the voice. Does the opening resemble your ordinary replies? Would you use those words? Is the amount of explanation proportionate to the relationship? Remove a copied phrase when it does not fit the new situation.

For the next task, use a different brief with the same reference. A reference that produces one acceptable draft has not yet shown that it transfers across situations. Keep the fact check each time, especially when an old message includes names, deadlines, apologies or promises.

The reference also needs to be available where you write. Use the instructions in the current conversation, or save them through a feature your tool supports. Do not assume a request to remember your voice applies to future chats.

Once the voice is useful, use Write Natural Replies for message-specific drafting. The meaning-preservation guide explains why a friendlier sentence can still change what you have agreed to.