How AI usage is counted
The numeric audit consumes nothing: it is computed in code, deterministically. Only the written recommendations and the chat with your accounts call a language model — and that is what gets counted. This page explains exactly how.
What a token is
A token is a fragment of text, roughly four characters. “conversion” is two or three; a number is one. Everything going into the model and everything coming out is counted in tokens.
A chat message consumes two kinds:
- · Input — your question, the account identity, the latest audit findings, the memory of past exchanges, and the data the model fetches to answer.
- · Output — the answer itself. About ten times smaller than the input, but five times more expensive per token.
Why the first message costs more
Your account context is cached on the first message of a conversation, then re-read at a tenth of the price for the following ones. Keeping your questions inside one conversation therefore costs far less than opening a new one each time. The cache expires after a few minutes of inactivity.
| First message of a conversation | ≈ 0,049 $ |
| Following messages, cache re-read | ≈ 0,009 $ |
Measured on a real account with five campaigns and three hundred keywords, Claude Sonnet 5.
What each plan allows
| Plan | Messages per day | Tokens per day | Analyses |
|---|---|---|---|
| Free | 7 | 250 000 | 1 / account / week |
| Solo | 50 | 1 000 000 | 7 / account / week |
| Pro | 50 | 2 000 000 | 7 / account / week |
| Agency | 150 | 6 000 000 | 7 / account / week |
Token caps are deliberately generous: in normal use the message count stops you first, not the tokens. Your dashboard counter always shows both, and tells you which one is binding.
About the free plan
The free plan gives you AI access without providing any key: it runs on ours, within a budget of 0 € per month and per user, smoothed over the month so some remains every day. That is what sets the 7 daily messages.
The numeric audit stays whole on the free plan: score, wasted spend, prioritised findings. Only the writing and the chat are capped.
Spending less
- · Keep your questions in one conversation rather than opening one per question: the cache does the rest.
- · Ask precise questions. “Which terms should I exclude on the Brand campaign?” costs less than a general analysis request, and answers better.
- · Run the audit before chatting: its findings enter the context and save the model from recomputing everything.
Your actual usage, day by day and exchange by exchange, is detailed in your workspace.
See pricing