Copilot will run the same prompt on a schedule and hand you the result, no reminder needed. Here's how to set one up properly, and ten prompts worth stealing.
Microsoft 365 Copilot’s scheduled prompts feature (generally available since January 2026) lets you save any prompt and set it to run automatically, daily or weekly, across Outlook, Teams, and Office.com chat, with up to 10 running at once. ChatGPT’s Tasks, Claude Cowork’s scheduled tasks, and Zapier’s Schedule trigger all do a version of the same thing, with different limits worth knowing before you pick where to build something. The setup itself takes minutes everywhere. What actually separates a useful scheduled prompt from a useless one is writing it so it still makes sense when nobody’s there to explain what “this week” or “the usual client” means, and being deliberate about which tasks earn a permanent slot at all. Below: the setup, how the other major tools compare, a framework for both of those decisions, ten original prompts to start from (five obvious, five that most lists skip), and what to watch for as an admin.
Every Monday for months, the same forty minutes went the same way. Open Outlook. Scroll back through the weekend’s overflow. Open Teams. Scroll through three channels you actually care about and four you don’t. Try to remember which of the five projects on your plate had something happen in them since Friday. None of it was hard. All of it was the same tedious reconstruction, done from scratch, every single week.
That’s the specific kind of task Copilot’s scheduled prompts feature is built for; not a new capability so much as removing the need to remember to ask for an old one. If you’ve already typed a prompt into Copilot Chat, Microsoft Teams, or Outlook and it gave you something useful, you can tell it to run that exact prompt again automatically, on whatever schedule you set, and hand you the result without you lifting a finger.
The mechanism is genuinely simple. According to Microsoft’s own support documentation, once a scheduled prompt runs, the response appears in your Copilot chat history the same way a manually typed one would, and you can optionally get a Teams or email notification when it’s ready.[1] Microsoft’s admin documentation confirms the feature works across three surfaces specifically: Microsoft Teams, Office.com chat, and Microsoft Outlook.[3] Nothing about what Copilot can see changes when you schedule a prompt instead of typing it live. It’s grounded in whatever your account already gives it access to, the same way any other Copilot chat would be.
This isn’t the same feature that inspired it, either, which matters if you tried scheduled prompts during the earlier public preview and found it clunky. The current version no longer runs on Power Automate under the hood; that older setup still exists for anyone who scheduled a prompt before the switch, but new scheduled prompts run on different infrastructure entirely.[3] If your team wrote this feature off a year ago, it’s worth a second look. We’ve covered a few of Copilot’s other underused capabilities separately if this is the first one that’s caught your attention.
You can’t schedule a prompt before you’ve run it. That’s the one prerequisite worth knowing up front, because it trips people up: there’s no separate “create a scheduled prompt” screen. You write a normal prompt, let Copilot answer it once, and then turn that specific prompt into a recurring one.
Steps as documented by Microsoft Support for scheduling a Copilot prompt. Source: Microsoft Support, “Schedule your most used Copilot prompts” (updated September 2026). [1]
You get up to 10 scheduled prompts running at once, and you manage all of them from one place: the settings menu inside Copilot, under “Scheduled prompts.”[1] From there you can run one immediately instead of waiting for its schedule, edit the timing, pause it, or delete it outright. None of this requires an admin. If your organization has the Microsoft 365 Copilot license and connected experiences turned on, which is the default, this is available to you right now, today, without a ticket.
The license question is worth being precise about, because “Copilot” gets used loosely. Scheduled prompts are part of the paid Microsoft 365 Copilot add-on, which lists at $30 per user per month on an annual commitment, layered on top of a separate qualifying Microsoft 365 plan you’d already need to have.[2] The free version, Copilot Chat, doesn’t include it. If you’re not sure which one your organization has, the fastest way to find out is to just try scheduling a prompt; if the option isn’t there, you’re on Chat, not the licensed add-on.
Once a prompt is scheduled, the same settings menu gives you four things you can do to it, and it’s worth knowing all four exist before you assume you’re stuck with your first draft:
Copilot isn’t the only place this exists, and if your team is like most, someone on it already has ChatGPT or Claude open more than Copilot. The mechanism is nearly identical everywhere it shows up: write a prompt, save it, pick a schedule, review the result somewhere central. The differences that actually matter are in the limits and the fine print, not the basic idea.
| Tool | What it’s called | Active task limit | Where you review it |
|---|---|---|---|
| Microsoft 365 Copilot | Scheduled prompts | 10 per user | Chat history, with an optional Teams or email notification |
| ChatGPT | Tasks | 3 (Free/Go) up to 15 (Pro/Enterprise), depending on plan | The Scheduled page, with an optional push or email notification |
| Claude (Cowork) | Scheduled tasks | No fixed cap published | The Scheduled page; each run opens as its own session |
| Zapier | Schedule by Zapier (a trigger) | Governed by your overall plan’s task allowance, not a separate cap | Zap history, plus whatever notification step you build in |
Confirmed directly against each vendor’s own current documentation, fetched September 2026. Sources: Microsoft [1], OpenAI [4], Anthropic [5], Zapier [6].
A few differences are worth knowing before you decide where to build something. ChatGPT’s free tier caps you at once a day with a loose window like “morning” or “evening”; an exact delivery time and hourly runs need a paid plan.[4] Claude’s scheduled tasks run remotely on Anthropic’s own servers, so most of them keep firing even with your laptop closed, with one exception: a task that needs a specific local file or app on your computer will only run while that computer is actually on.[5] And Zapier’s version isn’t really a scheduled prompt by itself, it’s a trigger you attach an action to, which lets it do something the chat-based tools can’t: check a condition first and only act, or only notify you, when that condition is actually met, like a spreadsheet row matching a search, instead of answering regardless every time it runs.[6]
None of this changes the discipline from the next section. Whether you’re writing the prompt in Copilot, ChatGPT, or Claude, if the instructions only make sense to the person who wrote them, sitting right there to explain what “this week” or “the usual client” means, they won’t survive being run automatically anywhere.
Here’s the part every list of “great scheduled prompts” tends to gloss over: the prompt that worked perfectly when you typed it live, sitting there ready to clarify what you meant, often falls apart the moment it has to run on its own. You knew what “this week” meant. You knew which client “the usual one” referred to. At 7 AM on a Monday with nobody watching, Copilot doesn’t, and it will guess, and the guess will be generically confident and quietly wrong.
The fix isn’t a smarter prompt. It’s a different discipline for writing one. Before you schedule anything, run it through what I’ve started calling the silence test: read your own prompt as if you were handing it to a new hire who has never met you, has no context on your job, and can’t ask a single follow-up question. Would they know exactly what to do?
| Check | Fails when | Fix |
|---|---|---|
| Time window is explicit | You wrote “today” or “recently” | Use a number: “the last 18 hours,” “since Monday,” “the last 7 days” |
| No hidden references | You wrote “the usual suspects” or “that project” | Name the thing, or describe how to identify it without you |
| Output has a shape | You asked for a “summary” with no structure | Ask for a list, a table, or numbered categories, not open prose |
| Scope matches what it can see | You’re asking about a file, spreadsheet, or CRM record | Keep it to email, calendar, and chat, which is what it’s actually grounded in |
| Wrong is survivable | A confidently wrong answer here could cause real harm if acted on | Add a line telling it to say so directly if it finds nothing, instead of padding the answer |
A prompt worth scheduling should pass all five before you save it. The last row matters most for anything financial or people-related.
Write every scheduled prompt for the version of you who isn’t in the room, because at 7 AM on Monday, you won’t be.
A quick before-and-after makes this concrete. “Summarize my week” is the kind of prompt that feels reasonable to type and produces something vague and re-readable every time it runs, because it never specifies a window, a format, or what counts as worth mentioning. “List every email I received in the last 7 days that mentions a deadline, sorted by how soon the deadline is, with the sender and one line on what they need” will produce nearly the same useful output every single week, because there’s nothing left for Copilot to guess at.
Every one of these tools eventually asks you to make the same call: is this worth a permanent slot, or should you just keep typing it manually when you need it? Most people answer that by gut feel, which is how someone ends up with ten scheduled prompts running and can only remember what three of them are for.
Two questions do most of the work. First: how often does this actually repeat? A prompt worth running every week is a very different decision than one worth running once a quarter, and quarterly or one-off tasks are almost never worth a permanent slot; just ask when you need them. Second, and more important: what happens if the automated answer is wrong and nobody notices right away? A morning email brief that’s slightly off wastes two minutes of your time. A summary feeding a financial or people decision, confidently wrong and unreviewed, is a different category of problem.
People who build automated AI systems for a living have a name for what happens when this gets skipped: a silent failure, where the automation quietly starts producing something wrong, generic, or out of date, and nobody catches it because the notification still fires right on schedule, looking exactly as normal as always. The fix isn’t complicated. It’s the same monthly review habit mentioned earlier, plus one more worth stealing from that world: the first time you schedule anything that lands in the “frequent and high stakes” box, read every single result closely for two or three weeks before you trust it enough to stop checking.
These five are the ones I’d set up first, before anything more specific to your role. Each replaces a piece of manual reconstruction most people do by memory every week, and each one passes the silence test above. Copy the prompt text directly; the schedule is a starting suggestion, not a rule.
| Name | Schedule | Prompt |
|---|---|---|
| Morning Triage Brief | Weekdays, 7:00 AM | “List the emails I received in the last 15 hours that need a reply from me today. For each, give the sender, subject, and one sentence on what they’re asking for. Separate anything marked urgent into its own line at the top.” |
| Meeting Countdown | Weekdays, 7:30 AM | “List my meetings for today with time, attendees by name, and one sentence on the likely topic based on the invite title. Flag any meeting back-to-back with no gap before or after it.” |
| End-of-Day Promise Tracker | Weekdays, 5:30 PM | “Look at my sent emails and Teams messages from today. List anything I said I would do, send, or follow up on, who I said it to, and any deadline I mentioned. Put anything due tomorrow at the top.” |
| Monday Blocker Scan | Mondays, 8:00 AM | “Search my email and Teams messages from the last 5 days for anything described as blocked, waiting, or delayed. List what’s blocked, who owns it, and how many days it’s been sitting. If nothing qualifies, say so directly.” |
| Friday Loose Ends | Fridays, 4:00 PM | “Review this week’s emails and Teams messages for anything I said I’d get back to someone on but never actually followed up. List who’s waiting and what they’re waiting for.” |
All five are original prompts written for this article, structured to pass the silence test in the section above.
Notice none of these ask for a stat, a strategic take, or anything requiring judgment. That’s deliberate. A scheduled prompt is closer to a smoke detector than an advisor: cheap to leave running, and its whole value is showing up reliably with a plain factual answer, not an insightful one. Save your own judgment for what you do with the list, not for interpreting whether it’s telling the truth.
Once the obvious five are running, there’s a second layer worth setting up that most prompt roundups don’t mention, because they’re less about summarizing your inbox and more about catching a pattern in your own behavior. These are the ones I’d genuinely call the interesting part.
| Name | Who it’s for | What it catches |
|---|---|---|
| The Delegation Audit | Managers, team leads | Search this week’s email and Teams messages for any question you personally answered more than once for different people. That repetition is the signal a process, template, or delegation is missing, not a personal-inbox problem. |
| The Say-Did Gap | Anyone building a new skill | Compare what you told a colleague, in email or Teams, you were going to learn, try, or practice this week against whether any message since then shows you actually did it. |
| The Batch 1:1 Prep | Managers with 4+ direct reports | Instead of prepping one-on-ones separately, pull recent email and Teams context for all direct reports at once, grouped by person, so prep happens in one sitting instead of five interruptions. |
| The Quiet Client Check | Account managers, consultants | Flag any client you’ve emailed regularly in the past but haven’t heard from in 10+ days. This is a tone signal from your own inbox, not a CRM replacement, and it won’t catch anything CRM-tracked deals already surface. |
| The Content You Promised | Marketing, comms | Search for any commitment made in email to a stakeholder about content or an asset “coming this week,” and check whether a later message confirms it actually went out. |
Five original prompts, written for this article, aimed at catching a pattern rather than summarizing a day.
The Delegation Audit is worth calling out specifically, because it’s the one that tends to actually change how someone works rather than just informing them. Most people who run it for a month find the same two or three questions keep resurfacing from different people, which is usually a sign that something should become a shared doc, a Loom, or a five-minute training moment instead of five individual replies. That’s a workflow fix a scheduled prompt can surface even though it can’t make the fix for you.
If you want more prompt ideas in this vein, particularly if your organization leans further into sales, finance, or HR use cases than the ones here, the “Awesome Microsoft Copilot Prompts” project on GitHub keeps a much longer running list specifically for scheduled prompts, released under a share-alike license.[7] Worth a browse once you’ve got these ten running and want to keep going.
Ten scheduled prompts sounds like a small number until you’re the one deciding what earns a slot. The honest answer is that most people shouldn’t fill all ten immediately, and definitely shouldn’t turn on a notification for all of them, or the feature meant to save you from checking things constantly will just become one more constant interruption.
Each of your ten slots should replace one specific thing you already do by hand, not just sound like a good idea.
If you can’t name the exact manual routine a prompt replaces, it’s probably a curiosity, not a candidate. Before scheduling number six or seven, ask what you’d stop doing manually because this now exists. If the honest answer is “nothing, I just wanted to see what it would say,” leave the slot open.
Only the prompts that would change your next hour deserve to interrupt you. Everything else can wait in your chat history.
| Prompt type | Notify or let it wait? | Why |
|---|---|---|
| Morning Triage Brief, Meeting Countdown | Notify | Time-sensitive; reading it late defeats the purpose |
| Monday Blocker Scan, End-of-Day Promise Tracker | Notify | Directly changes what you do in the next few hours |
| Delegation Audit, Say-Did Gap, Quiet Client Check | Let it wait | Useful on your own schedule, not urgent the moment it lands |
| Anything experimental or newly added | Let it wait | Watch it run a few times before deciding it’s worth an interruption at all |
A rough split across the ten prompts described in this article. Adjust based on which results you’d genuinely act on immediately.
One more thing worth doing, and almost nobody does it: revisit your scheduled prompts a month after setting them up. A prompt that seemed sharp on day one often turns out to need a tighter time window, a narrower scope, or a format tweak once you’ve actually seen four or five weeks of its output. Treat the first version as a draft, not a final answer.
If you’re setting this up for yourself, the main risk is a vague prompt producing a confidently wrong answer nobody catches. If you’re the one who manages Microsoft 365 for your organization, there’s a governance layer worth knowing about before your team starts turning this on broadly.
Scheduled prompts run under a setting called connected experiences, which is generally on by default but can be turned off tenant-wide. As of August 2026, Microsoft moved this specific feature from the “optional” connected experiences category into the standard “connected experiences” policy bucket, according to Microsoft’s own admin documentation.[3] In plain terms, that’s a compliance and policy reclassification, not a change to what the feature actually does; existing scheduled prompts kept running through the change without any action required, and users didn’t notice anything different. But if your organization manages Office privacy settings tightly through Cloud Policy, this is worth a quick check rather than an assumption that everything’s still configured the way it was before.
An admin can disable the feature entirely by turning off the “Allow the use of connected experiences in Office” policy, which also turns off several other Microsoft 365 features that share the same setting, not just this one.[3] Individual users can also opt out for themselves in their own privacy settings, separate from whatever the organization-wide policy allows. If a scheduled prompt someone set up stops appearing, checking this setting is the first place to look, well before assuming the feature broke.
| What the scheduled prompt does | What you still own | How it gets checked |
|---|---|---|
| Pulls the same search across email, Teams, or calendar every time, on schedule | Deciding the prompt is worth trusting in the first place | Read the actual output for the first 2-3 runs before treating it as reliable background noise |
| Formats the result consistently (list, table, or categories) | Acting on what it surfaces | Nothing changes just because a prompt ran; a person still decides what to do with a flagged blocker or a quiet client |
| Runs quietly with or without a notification | Noticing if it stops running or starts producing thin, generic answers | The monthly review described above, not a one-time setup and forget |
A scheduled prompt automates the retrieval and formatting step. It does not automate judgment, and it shouldn’t be trusted to.
None of this is a reason to avoid the feature. It’s a reason to treat “I set up a scheduled prompt” the same way you’d treat “I set up a recurring calendar reminder”: useful, low-effort, and still something a person needs to actually look at and act on, not a system you get to stop thinking about.
A few signs it’s time to revisit one rather than let it keep running unattended:
Any one of those is a reason to edit the prompt, change its schedule, or just delete it and free the slot for something you’ll actually use.
It’s a Copilot prompt you’ve already run once that you set to run again automatically on a schedule you choose, daily or weekly, instead of you having to remember to type it every time. The result lands in your Copilot chat history, and you can turn on a notification so you know when it’s ready.
The same data any Copilot chat you type live would see, grounded in your own account. Microsoft’s own admin documentation confirms the feature runs across Microsoft Teams, Office.com chat, and Microsoft Outlook specifically. Scheduling a prompt doesn’t expand or change what Copilot has access to; it just removes the need to ask it yourself each time.
Up to 10 per user, according to Microsoft’s own support documentation. You manage all of them from one settings menu inside Copilot, where you can run one immediately, edit its schedule, pause it, or delete it.
Yes. Scheduled prompts are part of the paid Microsoft 365 Copilot add-on, which lists at $30 per user per month on an annual commitment on Microsoft’s own enterprise pricing page, on top of a separate qualifying Microsoft 365 base plan. The free Copilot Chat tier does not include it.
A morning email or calendar brief that replaces something you already reconstruct manually every day, like checking what needs a reply before you open your inbox. Pick a prompt with an explicit time window and a clear output format (a list or table, not open prose) so it produces something consistent every time it runs, rather than something vague you have to re-read.
Every vendor documentation source in this piece (Microsoft Support, Microsoft’s own enterprise pricing page, Microsoft Learn’s admin docs, OpenAI’s Help Center, Anthropic’s Claude Help Center, and Zapier’s Help Center) was fetched directly on 7 September 2026, not taken from a secondary summary; the ChatGPT-vs-Claude comparison in particular corrected a secondary-source claim that Claude’s scheduled tasks only run while a computer is awake, which Anthropic’s own documentation does not say. The “generally available since January 2026” timing and the August 2026 connected-experiences policy shift for Copilot are corroborated by Microsoft’s own admin documentation. The idea that unmonitored automation can fail silently is presented as a pattern practitioners who build AI agents describe, which is our own synthesis of that discussion rather than a single measurable statistic, not a cited figure. Every prompt in the two prompt-pack tables was written fresh for this article; none are reproduced from any third-party list. The “Awesome Microsoft Copilot Prompts” GitHub collection, released under CC BY-SA 4.0, is credited once as further reading for readers who want a longer list, particularly for sales, finance, and HR use cases this article doesn’t cover in depth.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.