GPTZero sends the text to the real detector — its verdict (paid: counts words against the monthly budget, sentence-level heatmap). Heatmap is a free local approximation (GPT-2 on this machine, no words spent): it shades each word by how predictable it is, showing why text might look AI — use it to iterate freely, then spend GPTZero to confirm.

Result

GPTZero scoring is never automatic — each click sends exactly the words shown and counts against the monthly budget. The sentence heatmap comes with every score. Target: weighted AI% (AI + ½·mixed) below 20 with meaning and voice preserved. If the revised score is worse than the original, keep the original.