What “scheduled tasks” means in each of these tools
A scheduled or recurring task is a standing instruction: check something, or produce something, on a cadence you set, without you opening the app and asking again. Four mainstream AI products now ship this: ChatGPT, Claude, Gemini and xAI’s Grok Bot. Meta’s Muse help page covers reminders and recurring scheduled tasks, including daily check-ins and weekly summaries. It gives fewer details about tasks and integrations than the four products compared below, so Muse stays outside the table.
The differences worth checking before you build on one of these aren’t in the pitch, they’re in the mechanics: how many you can keep active, whether a run can act on your behalf or only tell you something, and what happens the week it fails without an obvious error.
The comparison table
| ChatGPT scheduled tasks | Claude scheduled tasks | Gemini scheduled actions | Grok Bot routines | |
|---|---|---|---|---|
| Create a schedule | Describe the task and time in a chat, or manage it from the Scheduled page in Settings | “Scheduled” tab, then “New task”: describe it to Claude conversationally, or set the name, prompt, approval mode, frequency and model manually | Type a prompt describing the task and how often you want it in Gemini | Ask a Bot conversationally to create a routine, then specify its schedule, input source and approval boundary |
| Active task limit | 3 (Free, Go), 5 (Plus), 10 (Business, Edu), 15 (Pro, Enterprise) | Not published; available on paid plans only (Pro, Max, Team, Enterprise), no free tier | 10 active scheduled actions, regardless of plan | A Bot can own up to 50 routines; the app keeps the 20 most recent run records per routine |
| Frequency | Free: once a day in a flexible window. Paid plans: exact times and hourly repeats | Hourly, daily, weekly, weekdays, or manual trigger | Daily, weekly or monthly, with delivery timed to the hour on a paid Google AI plan and prepared further in advance otherwise | Interval schedules or event triggers (for example, a Slack message or a Git event), set per routine |
| Where results arrive | Push notification and/or email (set in Settings > Notifications), plus the result posted in the chat thread | Posted in the Scheduled tab and the task’s chat; the help article doesn’t specify a separate push or email channel | In-app: the chat is marked unread and, on mobile, a push notification | Posted in the Bot’s conversation; the docs don’t describe an external push or email channel |
| Connected app actions | Supported apps such as Gmail, Slack and GitHub can read data or take permitted actions; some actions pause for approval | Google Workspace connectors can draft and send email, with approval controls available | Scheduled actions can use connected apps; available actions depend on the connected service and permissions | Routines can take actions through configured tools and stop at steps set to require approval |
| On a failed or missing-data run | A task may pause if inactive, if it needs more action, or if its chat is deleted; no retry policy is stated | The scheduling page does not specify a retry policy or failure alert | Inactive actions may turn off; check the task and resume it manually | Routines may pause after inactivity; xAI recommends setting a no-data and stale-data policy and reporting failures |
| Model choice | Supports eligible models; availability depends on the task, account plan and workspace settings. No per-task picker is described | Optional model choice when you set up a task manually | The scheduled-actions page does not specify model selection | Cursor manages model selection; there is no model picker |
| Plan and account access | Free and up, subject to account, app and workspace eligibility; paid plans raise the limit and add exact/hourly scheduling | Pro, Max, Team or Enterprise; no free tier | Personal accounts as the feature rolls out, with Keep Activity on; Workspace access depends on eligibility. A paid Google AI plan changes how far ahead content is prepared | SuperGrok, SuperGrok Plus or SuperGrok Heavy, or eligible Cursor paid plans; not included free |
| Primary documentation | OpenAI Help Center | Scheduling; Google Workspace connectors | Gemini Help | Routines; settings |
Anthropic’s scheduling guide now describes the feature under Claude and lists paid plans. OpenAI says eligible models depend on the task, plan and workspace; its guide does not describe a model picker for an existing task.
Which one acts, and which one only reports
The ability to act depends on the connected app and the permissions you grant. OpenAI says scheduled tasks can use supported apps and take permitted actions; an action that changes external data may pause for approval. Google’s scheduled-actions guide also allows connected apps, whose available actions depend on their permissions.
Claude’s Google Workspace connectors can draft, send, reply to and forward email, with approval controls for actions. A scheduled Monday follow-up can prepare replies for review instead of returning only a summary. xAI says Grok Bot routines can finish multi-step jobs and recommends naming steps such as sending, purchasing, deleting or publishing that should stop for approval.
All four can return a morning update. Each can also interact with connected apps, with the available actions and approval controls varying by product and integration.
How to write a recurring task that survives week six
A recurring task that works on day one tells you little about week six, when a trigger breaks, a source changes shape, or the task covers less ground without saying so. That’s the pattern covered in why recurring AI tasks fail: the common failure is a clean-looking result that covers less ground than it did in week one.
The cited pages do not establish one retry policy shared by all four products. OpenAI says inactive tasks may pause, Google’s guide says inactive actions may turn off, and xAI recommends a no-data and stale-data policy. Build a check into the prompt: ask the task to state what it checked (“checked all 5 vendor pages, no changes”), report when it found nothing, and hold any send, payment, booking, deletion or publication for approval.
When a dedicated always-on operator is the better tool
These four scheduling features live inside a chat product built for conversation first. That’s fine for a single-channel digest you read where you already are. It gets harder once a recurring task needs to reach you in more than one place, run on more than one model depending on the job, or leave a record you can hand to someone else without screenshotting a chat thread.
This is the gap an always-on operator is built for. OperatorNest’s recurring tasks run on the schedule you set, in the channel you asked from, whether that’s WhatsApp, Slack, email or the web, watch for conditions in between runs rather than only firing on a clock, and hold anything that sends, pays, books, deletes or publishes for your approval regardless of which model did the work. Every run is logged with what it checked, what it found and what you decided, so a shrinking scope shows up in the record instead of disappearing into a clean-looking weekly email.
The limits, including ours
Every option here, including a dedicated operator, has limits. ChatGPT’s available models depend on the task, account and workspace, and connected-app actions can pause for approval. Claude can act through its connectors, while its scheduling guide gives no numeric task limit or retry policy. Gemini caps scheduled actions at 10 and relies on connected-app permissions for actions. Grok Bot supports multi-step work and routines, while xAI’s documentation recommends designing for stale or missing data rather than promising a platform retry.
An always-on operator moves these tradeoffs into a different workflow. You get channel and model flexibility and a visible receipt for every run, and you still need to review it; a record that flags a shrinking scope helps when someone checks it before six more weeks pass. A fast, single-tool digest inside a chat you’re already in may be easier to run in ChatGPT or Gemini than to hand to a separate operator.
Which one to use, for what
For a single morning digest inside the chat app you already use, ChatGPT or Gemini keeps the result in that product, and Gemini’s fixed 10-action limit is easy to plan around. For a recurring task that should also draft or send something, compare the connected apps and approval controls each product supports before choosing a schedule. For a task that needs to reach you across more than one channel, use more than one model depending on the run, or produce a record you can point someone else to, consider a dedicated operator alongside a chat product’s built-in scheduler.