We Let an AI Draft This Blog — Here's the Guardrail That Keeps It Honest

We use AI to draft posts for this journal. You're probably reading one right now. We're not going to be coy about that, and we're not going to dress it up as some bold experiment. It's a workflow choice, the same as using Figma instead of paper or Slack instead of email. The interesting part isn't that we do it. The interesting part is the one rule that makes it not embarrassing.

The Rule Is Simple

No fabricated evidence. The AI cannot invent statistics, cite unnamed studies, or reference clients, projects, or outcomes that we haven't explicitly provided. If a claim requires evidence, that evidence has to come from us — from a real number we can point to, a real project we can name, a real observation we've actually made. The model's job is to shape and sharpen the argument. The facts are still ours to supply.

That's it. One rule. But it does a surprising amount of work.

Why This One Guardrail Matters More Than a Style Guide

AI writing tools are genuinely good at tone. Give them enough examples of your voice and they'll approximate it well enough to be useful. What they're bad at — structurally, not just occasionally — is knowing what's true about your specific business, your specific clients, your specific results.

Left unchecked, they'll fill that gap with plausible-sounding garbage. "Studies show that 73% of consumers…" Nobody ran that study. "Brands that invest in design see 3x returns…" That number came from nowhere. It sounds authoritative. It is invented. And it erodes the one thing a business journal is supposed to build: credibility.

The fabrication problem isn't unique to AI. Humans write vague, evidence-free content all the time. But AI makes it faster and more fluent, which makes it more seductive and more dangerous. A confidently written lie reads better than a hesitant truth.

How the Process Actually Works

The workflow is not complicated. Someone on the team identifies a topic — usually something we've been arguing about internally, something a client asked that made us think, something we've watched the industry get badly wrong. They write a brief. The brief includes the actual position we want to take, any real data or observations we're willing to stand behind, and a note on what we want the reader to do or think differently after reading.

The AI drafts from that brief. It structures, it sharpens, it handles transitions and sentence rhythm. What it doesn't do is add claims we didn't give it. If the brief doesn't include a stat, the draft doesn't include a stat. If we want to reference a project, we name the project in the brief.

Then a human edits. Not to check whether the AI hallucinated — the process is designed so that's not the main risk — but to make sure the argument is actually sharp, that the position is one we'd defend out loud, and that nothing reads like it was written to sound smart rather than to be useful.

The human judgment at the end isn't optional. It's the whole point.

What We're Not Claiming

We're not claiming the posts are indistinguishable from something written entirely by hand. They're not, always. We're not claiming AI-assisted writing is superior. It's faster in some dimensions and worse in others — it tends toward completeness over edge, and you have to push back against its instinct to balance every argument into mush.

We're also not claiming this approach is right for everyone. If your brand voice is genuinely eccentric, or if your value is tied to a single writer's perspective, the math probably doesn't work in your favor. Use the tool where it earns its place.

The Transparency Question

Some people will read this and feel vaguely annoyed that we used AI to write about using AI. That's fine. The annoyance is worth examining, though. The problem with AI-generated content was never that it existed. It was that it was bad, dishonest, or both. Content farms were churning out garbage long before large language models. The medium isn't the corruption. The lack of editorial standards is the corruption.

What we're arguing for isn't AI writing. It's accountable writing. The guardrail we've described exists to make us accountable — to force us to show up with real evidence, real positions, and real editorial judgment, rather than using the tool as a way to produce content without actually thinking.

If you're using AI in your content workflow and you don't have an equivalent rule, you should probably build one. Not because the model is malicious. Because the path of least resistance always leads toward fluent vagueness, and fluent vagueness is exactly what erodes trust over time.