How It Works

A dashboard, not a PDF you'll never reopen.

Worst offenders first. Every flag links to the exact text, the rule that fired, and a one-click fix.

Two engines. Only one of them ever touches an AI model.

The detection engine is a pure ruleset — a vocabulary blacklist for overused AI-tell phrases, plus structural checks like sentence-length variance and generic openers and closers. It's deterministic and runs at zero cost, which is why the scan itself is free and unlimited across your whole site.

Every flag has a severity — low, medium, or high, based on how strong a tell it is. A stray transition word costs a page a couple of points; a phrase like "in today's fast-paced world" costs a lot more. Those add up into a score out of 100 per page, plus an A–F grade so it reads at a glance. The number for your whole site weights pages by how much real content they have, so a few short contact or legal pages can't quietly average out one genuinely bad page, and one long post can't tank an otherwise solid site either. Repeating the same flagged phrase over and over also stops costing full points after the first few times — by then it's an established pattern, not new evidence. Never a black-box percentage: every point traces back to a specific flag on a specific sentence.

The fix-suggestion engine only runs when you click to generate a fix for a specific flag. It's grounded in the rule that fired and your actual sentence — never an open "make this sound human" prompt — and you review and approve every rewrite before anything changes. That's the part that costs money to run, which is why it's the paid layer.