I built a content tool that starts from your voice, not a prompt

Every AI content tool I tried had the same problem: the output sounded like AI.

Not in the obvious way — not the “as an AI language model” way. In the subtler way where you read your own blog post and think: this could be anyone. The vocabulary is right. The structure is fine. But the thing that makes your writing yours — the way you land a point, the analogies you reach for, the rhythm of your sentences — that’s gone.

A year of trying to fix this with prompts. Better system instructions. More context. Style guides pasted into every conversation. It helped, maybe 20%. The content went from “obviously AI” to “plausibly human but not specifically me.” Which is worse, in a way, because you can’t tell it’s bad until you’ve already published it.

The insight that changed everything

The problem isn’t the model. The problem is the input.

ChatGPT starts from a text prompt. AI CMO tools start from a URL. Both are trying to reverse-engineer who you are from artifacts — your website, your LinkedIn, your previous posts. They’re building a profile from the outside in.

What if you started from the inside out?

Hold a button. Talk for fifteen minutes about what you actually think about a topic. Not what you’d write — what you’d say to a colleague over coffee. The unfiltered version, with the asides and the “actually, here’s what really matters” moments.

That’s what Authexis does. It interviews you. You talk. It listens. Then it drafts something that sounds like you said it — because you did. The AI’s job isn’t to generate your perspective. It’s to organize what you already said into something publishable.

What I learned building it

Voice capture is a design problem, not an AI problem. The hardest part wasn’t getting Claude to write in someone’s voice. It was designing the interview flow so people actually talk naturally. Early versions asked essay questions and got essay answers — stilted, formal, not how the person actually thinks. The breakthrough was making the questions conversational: “What would you tell someone who disagrees with you on this?” That’s when people drop the performance and just talk.

The pipeline matters more than the model. When users could paste a topic and get a draft, the output was generic. Adding the interview step — idea selection, guided questions, spoken answers, then drafting — made the quality jump. Each step is simple. The sequence is what creates the result.

Content creation is a decision funnel, not a generation task. The platform scans your domain, proposes ideas ranked by relevance and freshness, and you pick which to write. You set the length and style. You answer the interview questions. You review and approve the draft. At every step, the AI proposes and you decide. That’s the product: AI handles execution, you make the calls that matter.

Distribution has to be seamless or people won’t do it consistently. Building the content is the hard part. Once it exists, posting it to LinkedIn, Bluesky, your blog — that’s mechanical. Queue-based scheduling with posting slots means you fill up when inspiration strikes and posts go out on schedule. Consistency without the daily grind.

Why now

Every content tool is converging on the same approach: paste a URL, get a post. The output is converging too — toward a median that sounds like everyone and no one. The market for “content that could be anyone’s” is saturated.

The market for content that sounds like a specific person — that captures how they think, not just what they know — is wide open. That’s where Authexis lives.

Launching on Product Hunt today. If you’ve ever published something and thought “this doesn’t sound like me,” this is for you.