The unit nobody explains
Every AI product prices in tokens and almost none of them tell you what that means for the thing you are about to write. A token is roughly three-quarters of an English word. A 1,500-word article is about 2,000 tokens of output, plus whatever you sent in — a brief, an outline, a transcript — which is often larger than the output itself.
A realistic article, costed
Take a genuine workflow: a 40-minute interview, transcribed, outlined, written to 1,500 words, then two rounds of edits on individual sections. The transcript is around 6,000 words in, the outline pass costs a few hundred tokens, the draft costs about 2,000 out, and selection-scoped edits cost a few hundred each because they only send the paragraph. Total, roughly 12,000 tokens. On the Starter plan's million a month, that is eighty-three articles.
Where the estimates go wrong
The two things that blow a budget are regenerating whole documents instead of editing sections, and RSS feeds without a ceiling. Both are avoidable and both are why we show the cost before you confirm and cap feeds by default.
Why we do not price per seat only
Per-seat pricing punishes the team member who uses the tool most, which is exactly backwards. A pooled balance means the person doing the heavy production this month simply uses more of it.