Two colleagues are on identical paid plans. One sends forty short questions a day and never hits a limit. The other sends six requests a day, each analysing a long report, and is cut off by mid-afternoon. What explains the difference?

The principle — a usage allowance measures work done, not messages sent, so message counts predict nothing on their own.

Why the key is correct — Anthropic lists the factors that affect usage: the length and complexity of your conversations, the features you use, which model you are chatting with, and the effort level selected. A long report analysed in full is long input, complex work, and often a more capable model — three of those four at once. Forty short questions can easily total less.

Why the others are wrong — nothing here suggests a fault; the behaviour is what the documentation predicts. Billing dates do not differ between two people described as being in the same position. And the heavier user is the one sending fewer messages, so frequency is not what is being penalised.

Remember this — if you are hitting limits, look at the size of each request before you look at how many you sent.

Sources — Anthropic's usage and length limits article.