Every plan has two limits and they work differently. The monthly allowance is how many words you can process in a billing period. The per-scan cap is how many words you can submit at once.
On the free plan that is 2,000 humanizer words a month with a 1,500-word cap per scan. Writer raises it to 25,000 a month and 5,000 per scan. Studio is 80,000 and 15,000. Team is 250,000 with no per-scan cap at all.
The per-scan cap is the one people meet first, because it decides whether a document goes through in one pass. A 4,000-word article on a plan capped at 1,500 has to be split into three, and each fragment gets rewritten without sight of the others — which is how repeated transitions survive into the final text.
Detection has its own separate allowance, and it is far more generous: 10,000 words a month free, rising to 2,000,000 on Team. Detection runs on our own model at a fixed infrastructure cost, where rewriting calls a paid model. That is the whole reason for the difference.
Using the detector does not consume humanizer words, and vice versa. They are two separate counters.
If a document exceeds your per-scan cap, split it at a section break rather than mid-argument, and process the sections in order. Continuity suffers least when the breaks fall where the reader would pause anyway.