Your Copilot Scheduled Prompt Runs Every Morning. It Does Not Remember Yesterday Morning.

You set up a scheduled prompt in Microsoft 365 Copilot: every weekday at 8 AM, summarize new emails and flag anything that needs your attention. The first week it feels like a personal chief of staff. By the second week you notice the same unresolved thread showing up Monday, Tuesday, and Wednesday. The prompt reports it every time, as if it had never mentioned it before.

That is not a bug in your prompt. It is how scheduled prompts work. Here is what a scheduled run actually remembers, what it forgets, and what operators do about the gap.

What scheduled prompts actually do

Scheduled Prompts is a Microsoft 365 Copilot feature for automating prompts you run regularly. You run a prompt once, hover over it in your chat history, and select "Schedule this prompt." You set the time and frequency, up to 10 different scheduled prompts per user, and Copilot executes it automatically. Each result lands back in your Copilot chat history, with newly run results shown in bold.

Think of it as a cron job for a prompt. The important word is "prompt": what gets saved and re-executed is the prompt text. Every run is a fresh execution of the same static text at the scheduled time.

This is a genuine scheduled agent: an unattended run, at a set time, doing recurring work. And like every scheduled agent, the interesting question is what it knows when it wakes up.

What a scheduled run remembers: Copilot Memory

Microsoft gives Copilot a personalization feature called Copilot Memory: it remembers relevant details from your conversations and applies them to future interactions. It builds memory two ways: saved memories (you tell it something worth keeping, like your preferred report format) and chat history inferences (patterns it recognizes from your conversations). You can review, edit, and delete saved memories in the personalization settings.

This sounds like the answer to the amnesia problem. It is not. Copilot Memory stores preferences: your writing style, the format you like for reports, recurring responsibilities. It is personalization, not run history. The official line is explicit: Copilot Memory is not the same as Copilot conversation history. Memories include saved preferences and inferred details; they do not include what a specific scheduled run found last Tuesday.

So your scheduled prompt remembers that you like bullet points. It does not remember that it already flagged the vendor thread on Monday and you have ignored it since.

The three things every run forgets

The first is deduplication. A scheduled prompt that summarizes "new emails" has no record of which emails it already summarized. With a relative window like "the last 24 hours," an unresolved item gets flagged Monday and Tuesday, then ages out. The run cannot tell you "I already told you about this" because it keeps no list of what it told you about. For an 8 AM digest, this turns into noise: the same items reappear until they scroll out of the window.

The second is adaptation. The saved prompt text never changes between runs. It cannot say "last run found nothing, so this time look somewhere else" or "the weekly summary already went out, so today do the deep dive." Every run executes the identical text with identical logic. Scheduled prompts are a fixed script wearing a fresh coat of paint each morning.

The third is outcomes. Copilot Memory saves how you like things presented. It does not save what happened: which alert was a false positive, which summary you acted on. Run state is a different category from preferences, and scheduled prompts have nowhere to put it. That is why prompt design guides for unattended runs stress self-contained prompts: relative time windows, structured output, scoped data. The advice exists because the run arrives with no memory of its own history. It has to re-derive everything from scratch.

What operators do about it

The first fix is the documented one: write prompts that assume amnesia. Use relative time windows ("in the last 24 hours," "since Monday") instead of context-dependent references ("the email I sent"). Keep the output self-contained and structured, like priority lists or action items, so it is useful without follow-up questions. This does not give the prompt memory. It makes memory unnecessary for the specific job. For simple digests, that is enough.

The second fix is a completion ledger. Keep a file in SharePoint or OneDrive as a running log, and instruct the prompt to read it at the start of each run and record what it reported. The prompt cannot write run state into its own memory, but it can read a file you both share. The ledger becomes the run history that Copilot Memory was never designed to be. The cost is upkeep: the file only stays accurate if the runs keep it accurate.

The third fix applies when the scheduled prompt is one agent in a larger stack. Most operators have several scheduled agents across several tools: Copilot prompts for the daily digest, n8n workflows for enrichment, a coding agent for reports. Each platform remembers its own preferences and forgets its own runs, and none share state. For that, you need a shared memory layer: one place where all of your scheduled agents read and write run history, so every tool wakes up with the same picture of what already happened.

This is what Vilix AI is built for. It is a cloud-hosted memory layer, so there is nothing to deploy and no infrastructure to manage. The same memory and context follows your agents across every tool and device over MCP: phone apps, laptop, every AI tool you connect. It stores full conversation history, not just distilled facts, so the actual record of what each run did is there when you need it. Headless scheduled agents can connect with an API key as Bearer to the MCP endpoint and get full memory with zero memory infrastructure. It runs on a free plan forever, with a 7-day Pro trial that needs no credit card. Your data is portable: export everything or delete it anytime. One memory, every device, every app.

If you run Copilot scheduled prompts today, keep Copilot Memory for what it is good at, your standing preferences, and the ledger pattern for run-to-run state. When your automation grows past one tool, give every scheduled agent the same shared memory instead of letting each one wake up alone.

The bottom line

Do Microsoft Copilot scheduled prompts remember between runs? They remember the prompt text, your formatting preferences, and the patterns in your chat history. They do not remember what yesterday's run found, what it already reported, or what happened since. For simple recurring digests, amnesia-first prompt design covers you. For anything that needs to know what the last run did, build the ledger; for a multi-tool stack, use a shared memory layer. Your agents forget everything between runs only until you give them one memory.

To give your scheduled agents one shared memory, start Vilix AI free at https://vilix.ai/?utm_source=vilix-blog&utm_medium=article&utm_campaign=copilot-scheduled-prompts-dont-remember-yesterday. Free forever, no credit card, and you can leave with your data whenever you want.