Drafting with AI without sounding like a robot

The first time a team member shows me an AI-drafted caption, I can usually tell within two sentences. The grammar is usually perfect. The problem is that it sounds like nobody. The tone is flatly helpful, the examples are generic, and the closing line is something no Indonesian team member would ever say out loud. This is the voice problem, and it is the single biggest reason teams stop using AI for writing after the first week.

The real fix is giving the AI something to copy from. Before you ask AI to draft anything, gather five to ten pieces of writing your team published: old captions, emails, blog posts, anything that already sounds like you. Paste two or three of them into the prompt as examples. Then ask the AI to match that tone, not a generic professional tone. This single habit changes the output more than any prompt template you can buy.

Feed the AI your best work, not your most polished work.

Bandung teams have an advantage here that they sometimes undervalue. The creative community in Bandung has a distinct, warm, slightly playful voice that carries well into short-form content. When I run content sessions in Bandung, I ask the team to bring their three best-performing posts from the last month, not their most polished ones. The ones that worked. Those are the voice samples the AI should learn from. A team that feeds the AI its own best work gets drafts that need editing, not rewriting.

Here is the workflow I recommend for a team starting out. One person drafts with AI, a second person edits before anything goes live. Fixing grammar is rarely the editor's job; the grammar is usually fine already. The editor asks one question: does this sound like us? If the answer is no, the editor rewrites the opening sentence in their own words and sends it back. Over a few weeks the AI learns the pattern from the corrections, and the drafts get closer to the voice on the first pass.

One mistake I see often: teams try to automate all their content at once. They want AI to write every caption, every email, every blog post, starting Monday. That is how you get a feed that looks consistent but feels hollow. Start with one content type: the one that takes the most time and matters the least. Weekly internal summaries, customer reply templates, or social captions that you can batch. Get the voice right on that one thing for a month. Then expand.

The goal is not to publish AI-written content. The goal is to publish your team's content faster, with AI doing the boring first draft and your team doing the last ten percent that makes it sound human. If you are spending zero time editing AI drafts, your audience can tell. If you are spending thirty seconds fixing the opening line and adding one real detail, they cannot. That thirty seconds is the whole job.