AI Has Not Earned the Right to Write Raising the Barr Weekly Memo: Issue 715

The default AI request I hear constantly is: “Write this for me.”

People hand AI an unfinished idea and ask it to make the idea sound finished. The result can be polished and persuasive long before the thinking is ready. AI can make an unfinished idea sound like a finished conclusion.

That is the wrong sequence. AI has not earned the right to write. First, it should interview you until your idea is strong enough to publish.

The Article I Had Already Written

I learned that while developing this article. I began with what felt like a strong premise: stop using AI primarily as a writer and start using it as an interviewer. I had proof. The same process had helped me develop my strategic AI profile, uncover four frameworks embedded in my work, and rethink how my four core offerings fit together.

Then I ran a skill I had built for exactly this purpose. Its assignment was simple: “Interview me one question at a time. Challenge weak answers. Do not let me off the hook. I will challenge you, and you challenge me.”

It did exactly that. It asked what professionals commonly believe, pressed me on why that belief was wrong, and demanded a real example, why it mattered, the practical consequence, and the action I wanted readers to take. Every answer made the idea feel stronger.

Then it checked the idea against my recent articles. There was a problem: I had already written much of it.

The overlap was not vague. In “You Built a Career. You Did Not Build an Asset.” (Issue 707), I had already included a section called “AI as the Interviewer, Not the Writer.” The article described how the process surfaced four frameworks and included an interview prompt readers could run.

“Your AI Agrees With You. That Is the Problem.” (Issue 709) covered AI challenging my thinking and the integrated structure of my business. “Last Week’s Conclusion Has a Hole in It.” (Issue 710) examined that structure again. “If AI Has Never Told You You’re Wrong, You’re Using a Mirror.” (Issue 712) showed what happens when AI is deliberately instructed to argue rather than agree.

The interview caught what I had missed: I was about to present thinking I had already published as a new idea.

AI Can Finish the Draft Before You Finish the Thinking

That is the deeper risk of asking AI to write too soon. AI can turn an unfinished idea into an outline and an outline into an article that sounds authoritative. Polished language does not prove that the underlying thought is complete, distinctive, or correct. It proves only that AI can complete the assignment.

Consultants, advisors, and thought leaders risk gaining smoother sentences while losing what makes their thinking distinctive. If the idea has not been challenged, polished language only makes the unfinished thinking harder to see.

The Interview Has a Job

The job of an AI interview is to determine whether there is an article worth publishing. Collecting enough material for a draft is only the beginning. Before writing begins, the interview should establish four things:

1. The idea must be complete. What are you actually arguing, and what are you rejecting?
2. The idea must be differentiated. What does it say that your audience has not heard from you already?
3. The strategy must be clear. Why does the argument matter, what is at stake, and what changes if the reader accepts it?
4. The execution must be practical. What should the reader do differently after reading?

If those questions remain unresolved, the interview is not finished. The writing should wait until you can answer them clearly.

Challenge Must Run Both Ways

This process does not work if you treat AI as the authority and stop exercising your own judgment. While developing my strategic AI profile, AI challenged my answers and I challenged its conclusions. Sometimes it held its position; sometimes I showed it where its reasoning failed. At other times, I discovered that I could not defend my own beliefs as strongly as I thought I could.

That exchange surfaced ideas neither of us began with. The material, experience, and judgment were mine. AI created the conditions that forced me to articulate, test, and structure what had remained instinctive.

At times, I widen that exchange by bringing in a second AI platform. If an idea begins with Claude, I give its reasoning and conclusions to ChatGPT and ask it to identify weak assumptions, missing evidence, and flaws in the logic. I take ChatGPT’s critique back to Claude, ask it to respond, and then challenge both platforms myself. I question their assumptions, press them to defend their conclusions, and reject any reasoning that does not hold.

Two AI platforms can still share the same errors. Their disagreement gives me competing reasoning to investigate, not proof that either one is right.

Their disagreement exposes the fault lines in the idea. I remain the final judge.

That is thinking partnership. Not obedience in either direction.

Make AI Earn the Draft

The lazy approach is asking AI to write before you have done the work of developing an original idea. That shortcut skips the difficult thinking that gives your writing something original to say.

A real interviewer slows the rush toward output. It notices when your answer is rehearsed. It asks for proof. It finds the contradiction. It checks the idea against the prior work you make available, so familiar material does not pass as new. Only then should AI help you write.

The practical question is how to build that discipline into your own AI conversations. Begin by giving AI a precise interviewing assignment that defines how it should question you, what it should challenge, and when it has earned the right to write.

I wanted that process available whenever I needed it, without rebuilding the instructions each time. So I turned it into a skill. The skill preserves the questions, standards, and sequence, allowing AI to apply the same process consistently. In my next newsletter, I will take that skill apart and show you why skills may change how you work with AI.

You can begin now without the skill. Open a new AI conversation, bring it one unresolved challenge, and give it this instruction:

“Here is the challenge I am working through: [challenge]. Interview me one question at a time. Challenge answers that are weak, generic, or unsupported. Do not let me off the hook. I will challenge you, and you should challenge me. Do not begin writing until we have a complete and differentiated idea, a clear strategy, and executable tactics. When you believe we have reached that point, summarize the idea, strategy, and tactics for my approval before writing.”

A finished draft cannot rescue an unfinished idea. Make AI earn the right to write.

This Week’s Carousel

AI Has Not Earned the Right to Write