We Got Caught in the AI-Written Content Detection Fight, Here's What Happened

A syndication partner ran a batch of a client's blog posts through an AI-content detector before agreeing to republish them, and flagged six out of twenty as 'likely AI-generated.' Four of those six were written entirely by a human writer who'd been on the team for three years. One of the other two actually was AI-assisted, heavily edited, and it passed clean.

That's not a fluke, that's the tool. AI-content detectors score based on statistical patterns like sentence-length variance and predictability, and a competent human writer who writes clean, structured, low-burstiness prose scores exactly like a language model, because that's what both are optimizing for.

The actual argument we had to make

We couldn't argue the detector was 'wrong' in any way the partner would accept, because they trusted the tool's number more than our word. What worked instead was showing drafting history — version timestamps, comment threads, the writer's research notes — provenance, not prose analysis.

You can't out-argue a detection score with vibes. You need a paper trail.

What changed afterward

• We started keeping draft history and editorial comments as a standard retention practice, not just for this partner but for every piece
• We stopped using detector scores internally as any kind of quality signal — they measure a different thing than quality
• We pushed back on any client contract clause that cited a specific detection tool's score as a pass/fail gate
• We accepted that AI-assisted writing, properly edited, is going to keep getting flagged and the fight is about process trust, not writing style

The uncomfortable truth is that detection tools aren't measuring whether AI wrote something, they're measuring whether the writing is boringly consistent — and good editing makes writing more consistent, not less. Until that changes, the fix isn't writing worse on purpose, it's building a paper trail you can point to when a false flag shows up, because it will.