I'm not a writer. I'm the developer at a content marketing agency. I build internal tooling, manage the CMS, watch what happens when editors try to scale brand-voice content with AI.
Most AI tools struggle with longform writing. Drafts come back fine for short copy, then drift to a default register over a 2,000-word article. Editors spent hours per article rewriting back to brand voice.
The "formal/casual" sliders we'd been given didn't help. Voice doesn't fit on one slider, it shows up in word choice, sentence rhythm, what the brand refuses to say.
What worked was the way our editors briefed each piece by hand. They'd give the writer a list of brand-specific rules. "Don't start with 'In today's…'." "Use contractions." "Cut anything that sounds like a LinkedIn post." Real lists, brand by brand.
So I built a tool around the same format, voice as a spec, not a vibe slider. Banned phrases, cadence rules, a few worked examples from the brand's existing content as an anchor, then a polish pass that re-reads the draft and rewrites anything that still reads AI.
It's not perfect. But the brand-voice consistency holds across articles in a way the slider tools never did.
How are agencies handling this internally? I assume the bigger shops have their own house-built spec systems, but for solo content marketers + small agencies this seems to be the unsolved problem. What are you doing today?