If your team is publishing more than ever this year and pipeline still feels flat, you're not alone, and it's probably not a volume problem. AI made execution cheap. It didn't make your content sound like you.
You can usually tell within a paragraph. Sentences arrive in tight little bursts. Three short, punchy statements in a row. A dash where a comma would have done the job. Nothing on the page is technically wrong, and none of it sounds like a person who actually knows your business wrote it.
This came up directly in a recent internal masterclass on brand voice, when one of our writers pushed back on a live demo: the output matched the brand's tone settings, and it still read as obviously AI-generated. Short, punchy statements stacked back to back. The cadence everyone's started to notice, dashes or not.
The full recording is below if you'd rather see the before-and-after live.
Why more content doesn't fix this
Language models are built to predict the most statistically likely next word. That's useful when you need something fast and readable. It's a problem when "most likely" means "sounds like every other company's marketing copy," which at scale produces a very specific, very recognizable rhythm. Tone settings help with formality and warmth. They don't touch cadence. You can ask a tool to sound warmer and still get the same three punchy sentences in a row, because warmth and rhythm are two different problems, and publishing more of it just means more of the same problem, faster.
That matters for a simple business reason: buyers are getting better at spotting generic AI content, and every piece that reads as generic is a missed chance to be the vendor a buyer remembers before they start comparing anyone on price.
What actually fixes it
Fixing tone is a settings problem. Fixing cadence and word choice is a rules problem, and it takes two things a tone dial alone can't give you.
The first is an explicit list of words to use and avoid, built from what your best writer already notices when something sounds off. If "client" should always be "partner," write it down. If certain words are banned outright, write those down too. This isn't guesswork. It's making one person's instinct into a rule everyone, including the AI, actually follows.
The second is a way to check the work instead of guessing. Paste a draft in and score it against your locked voice: which banned words showed up, how far off the tone landed, and a rewrite that fixes it. In one internal test, a paragraph full of inflated, generic language scored a 38 out of 100. The rewrite, stripped down to plain language about what the team actually built, scored a 92. Same information. Completely different read, and a completely different chance of actually landing with a buyer.
The part a system still can't do for you
None of this replaces the person who first noticed the problem. Someone has to catch that a sentence feels off and decide it's worth fixing. That single moment of judgment is what the whole system runs on. Skip it, and you get a tool that produces confident, grammatically perfect copy that still reads like it came from nowhere in particular, no matter how much of it you publish.
If your content output keeps climbing and pipeline still isn't, more content isn't the fix. Find out where the real gap is in the T2D3 Growth Workshop.