7 min read

AI content that does not sound like AI

The tells are structural, not stylistic. What actually changes AI writing quality: specificity, source material and a human who removes things.

Last updated 15 August 2026

Readers rarely detect AI writing by style. They detect it by emptiness — text that covers a topic without saying anything only that business could say.

The actual tells

  • No specifics. No numbers, names, dates or examples. Generic because it was generated from nothing but the topic.
  • Perfectly balanced. Every point has a counterpoint. Real expertise has opinions and states them.
  • Uniform paragraphs. Human writing has rhythm; a one-line paragraph after a long one is a human decision.
  • Summary endings. "In conclusion, X is an important consideration." Nobody who knows the subject ends like that.

What genuinely improves it

Not better prompts. Better inputs.

  • Feed it real material — your past work, your pricing, your actual objections and answers.
  • Give it a position to argue rather than a topic to cover.
  • Have a human delete rather than add. Most AI drafts improve by 30% purely by removal.
  • Make it name things: the district, the product, the number, the week.
On detection tools

Do not optimise for AI-detector scores. They are unreliable in both directions, and Google's stated position is about whether content is helpful and original, not how it was produced. Write things worth reading and the question stops mattering.

This is why our content departments are trained on your material and why nothing publishes without your approval — see how that works.

Keep reading

18 Aug 20268 min read

Why multi-agent AI systems fail in production

Most multi-agent demos work and most multi-agent deployments do not. The reason is error compounding — and the fixes are structural, not model upgrades.

Read
Want this done rather than explained?

We build the agent team, connect it to your accounts and supervise the output. Twenty-minute call · See pricing