Quick Answer
Gemini Spark supports time-based schedules, Gmail monitors, and topic monitors. A task is the goal, a schedule is the trigger, and a skill supplies reusable instructions or tools. Spark schedules are different from scheduled actions in ordinary Gemini chats.
Use schedules for reviewable, non-urgent work. Google’s help page says event timing can be approximate, usage limits can prevent a run, and topic monitors are not suitable for time-critical events such as buying tickets the moment they go on sale.
The Three Schedule Types
| Type | Trigger | Good fit | Avoid |
|---|---|---|---|
| Time-based | Once, hourly, daily, weekly, monthly, or yearly | Digests, recurring document checks, routine preparation | Deadlines requiring exact execution time |
| Gmail monitor | A new message matching a Gmail filter | Flagging or drafting around bounded email patterns | Unsupervised sending or sensitive account actions |
| Topic monitor | A specified event in news, finance, sports, local events, and other topics | Non-urgent alerts and research updates | Trading, emergency, or instant-purchase decisions |
Google’s June product update also describes monitoring blogs, news, social media, finance, shopping, weather, sports, and email. Treat those examples as supported product categories, not proof of complete or instantaneous coverage.
Current Requirements
Spark’s current help content lists age 18 or older, a personal Google Account, Google AI Pro or Ultra, Keep Activity, and an eligible country. It says Spark schedules can run in the cloud even when the device is off.
The time zone is fixed to the location where a time-based schedule was created. Travel does not automatically move the trigger to the new local time. Edit the schedule when the intended local time changes.
Build a Reliable Schedule
Define all three parts:
- Task: the bounded outcome, such as summarizing five named newsletters.
- Schedule: the time or event that should start the work.
- Skill or method: the sources, filters, format, and actions Spark may use.
Add a stopping rule and review surface. For example: “Every weekday at 08:00, summarize messages with label Project-A into a draft note; cite each message; do not send email, edit calendars, or archive messages.”
Test with Run now, inspect the task work panel, and compare the result with the underlying sources before enabling an ongoing trigger.
Operating Limits and Failure Modes
Google’s August 2 help page documents several practical limits:
- up to 50 active schedules;
- up to 15 tasks running at the same time;
- a schedule may not run when the account’s compute-based usage limit is reached;
- runs may be delayed during high traffic;
- trigger timing is approximate;
- a completed schedule cannot be resumed;
- turning off Spark pauses schedules;
- losing an eligible subscription pauses schedules and turns off skills.
Design the workflow so a missed or late run is visible. Use a dated output, record the last successful source check, and alert on absence when the task matters.
Safety for Unattended Work
Google warns that an offline user may be unable to stop an unintended scheduled action. Avoid schedules that use information or perform actions you consider sensitive. Keep unattended tasks read-only or draft-only until repeated tests show predictable behavior.
Good first schedules produce a digest, checklist, or draft. Riskier schedules delete files, send messages, place orders, change account settings, move money, or publish externally. These need explicit confirmation and an independent audit trail.
For the data source itself, use the Connected Apps and MCP guide. For web tasks, use the Chrome auto-browse safety guide. For scheduled local-file work, review the macOS Connected Folders guide.
Official Sources
- Google Help: Create and manage schedules for Gemini Spark
- Google Help: Use Gemini Spark to manage tasks and workflows
- Google: Gemini Spark monitoring, macOS, and connected-app updates
Source check: August 2, 2026. Verify live limits, eligibility, triggers, and task permissions before depending on an unattended workflow.