Run cadence
Choose how often a tracked prompt runs, what each cadence costs on your own model key, and why changing it mid-measurement distorts a trend.
Updated October 20, 2026
Cadence is how often a prompt runs. It is also the only dial in the product that directly scales your model spend, which makes it the setting people change most and regret fastest.
Picking one
Weekly is the right default. Fifty-two points a year per prompt is enough to see a trend and cheap enough to run a large set.
Daily belongs on prompts you are actively working. If you shipped a page on Tuesday and want to know by Friday whether anything moved, daily is what gives you the resolution to see it.
Monthly is for prompts you are keeping an eye on rather than working, in a market that does not move much.
What it costs
A probe is one prompt on one assistant. Four assistants means four probes per measurement, each a few cents on your key. Weekly for twenty prompts is a couple of dollars a month; daily for the same twenty is seven times that.
The prompt dialog estimates the monthly cost across all platforms as you choose, and Settings shows the projected cost by engine for the whole project. Those are the numbers to read rather than a per-call price.
Changing it mid-measurement
A trend line built from weekly points and then continued with daily ones is two different measurements on one axis. The points are still correct, the slope is not comparable, and a reader who does not know will draw a conclusion from the join.
If you change cadence, change it at a point you can explain, and prefer changing it for all prompts in a topic at once rather than one at a time.
Cadence changes take effect on the next scheduled run, which can be the same day. There is no monthly boundary to protect you from a bill.