Gemini scheduled actions: ten slots, and one detail that matters enormously

What it is

Gemini can run a saved prompt on a daily, weekly or monthly schedule. You write the prompt with the timing in it — "every morning at 7, give me a summary of..." — and Gemini confirms the scheduled action. Results are delivered when the schedule fires.

It's rolling out gradually across personal Google accounts and qualifying Workspace accounts, so availability varies. You can have up to 10 active scheduled actions at a time.

The detail that decides most use cases

⚠ Responses are prepared in advance so they can be delivered on time. Google's own documentation is explicit that as a result, rapidly changing data — stock prices is their example — won't be current when it reaches you.

This is worth sitting with, because it inverts the obvious assumption. A scheduled action is not "run this prompt at 7am". It's closer to "have an answer ready by 7am", and the work may have happened hours earlier. How much earlier depends on your plan: several hours without a paid Google AI plan, within about an hour with one.

So the entire category of time-sensitive monitoring is out. Prices, scores, availability, anything where the value is that the number is current — a scheduled action will hand you a stale figure with no indication that it's stale, which is worse than not asking.

What it's actually good for

Content where a few hours of staleness costs nothing, which is most personal routines:

A daily digest on a topic that moves in days rather than minutes. A weekly roundup. Practice exercises or prompts — language drills, interview questions, writing warm-ups. A morning briefing assembled from your own Google data, where the integration is the real advantage. Creative prompts and recurring nudges.

Within that range it works well, and the tie-in to Gmail, Calendar and Drive means it can reference your actual week rather than generalities.

Setting one up

Every weekday at 7am, summarise my calendar for the day and flag
anything that looks like it needs preparation. Three lines maximum.
If the day is clear, just say "clear day".

Requirements and habits worth knowing: Keep Activity must be enabled or the feature isn't available. Your location at creation time is used for future responses, which matters for anything location-dependent. And as with every scheduled prompt, specify the output shape and give it permission to say nothing, or you'll get an essay every day whether or not there's anything to report.

The ceiling

Ten active actions, daily at the most frequent. That's a generous allowance for personal routines and a hard stop for anything systematic. There's no tier that turns this into a monitoring platform — combined with the prepared-in-advance behaviour, it's explicitly not built for that job.

The cost picture

Bundled into your account, so per-run cost isn't the constraint — the slot count and the freshness behaviour are. Which is the recurring pattern across all three assistants: the limits that bind are structural, not financial.

That's the useful thing to internalise. People assume the reason to write code is to save money on model calls. At personal volume it isn't — daily runs cost pennies a year either way. The reason is that the assistant products are deliberately shaped for a handful of personal routines, and anything beyond that shape needs a different tool regardless of what you're willing to spend.

When you need something else

If you need current data, more than ten watches, sub-daily frequency, or output that's identical enough to feed into something else, you need a scheduled script. It's free, unlimited, runs exactly when you say, and fetches the data at the moment it runs — which, for monitoring, is the entire point.

The overview covers how to choose.