How I Prompted My Way to a More Sustainable Community Roundup

SoftwareOne has a global Copilot initiative that includes monthly Office Hours, live sessions where people across the organization can ask questions, see demos, and get a better sense of what is actually useful in Microsoft 365 Copilot.

To support that broader effort, a colleague and I help manage a shared SharePoint library where people can upload decks, recordings, templates, guides, and other resources related to Copilot and Copilot-adjacent work. The goal is pretty simple: if someone is trying to learn something, prepare for a customer conversation, or avoid reinventing the wheel when delivering a project, there should already be something there to help.

That part I believe in completely.

The harder part is keeping a community resource like that visible and active. Libraries do not stay useful just because they exist. People need reminders that they are there. Contributors need to feel like the things they upload actually matter. Someone has to keep the momentum going.

Which is where the usual problem shows up.

This is not our full-time job. It is community support work layered on top of the rest of what we already do. So while we kept talking about doing a regular roundup, the kind of post that highlights new uploads, thanks contributors, points out gaps, and helps people find useful content, it was also obvious that anything too manual was going to be the first thing to fall off the list.

So I started looking at whether Copilot Cowork could help turn this from a good idea we kept talking about into something we could actually sustain.

Why Cowork

This is not a hypothetical use case. I was not trying to build a flashy AI demo or prove that everything should become a workflow.

I was trying to solve a very ordinary problem. We needed a repeatable way to create update posts around a shared content library without having to manually pull together the same information every time. And because a colleague and I co-manage this content, I also wanted a process that would help keep the posts consistent in structure and voice instead of sounding completely different depending on who happened to draft one that week.

That consistency matters more than people think. Community content can start to feel random very quickly if there is no rhythm to it.

Cowork was interesting because it could work against Microsoft 365 data through prompting. No custom build. No flow to wire up. No extra project around the project. Just a prompt, a source, some iteration, and a clear idea of what I wanted the output to do.

That was enough to make it worth trying.

Starting with the Output

I did not start by asking how to automate a roundup. I started by asking what would make one worth reading.

Nobody needs another bland internal summary full of forced enthusiasm and vague statements about collaboration. If this was going to exist, it needed to do something useful.

So I started with the structure. The sections I kept coming back to were:

• By the numbers, with new uploads, library growth, top focus area, and contributor count
• Contributor shoutouts, because people are more likely to contribute again when their effort is actually acknowledged
• Pick of the week, with one standout item and a short reason it is worth a look
• What was uploaded, in a format people can scan quickly
• Metadata health check, with a friendly nudge where key fields are missing
• Most wanted, to point out thin spots in the library and encourage the right kinds of contributions
• A poll or prompt, to give people an easy way to engage
• Library updates, for any changes to categories, columns, or structure

Once I had that, the prompting got much easier. I was no longer asking Copilot to “write a roundup.” I was giving it a job with a specific format, purpose, order, tone, and set of rules.

That made all the difference.

What Prompting Actually Looked Like

This was an earlier version of the prompt. Good enough to get moving, but still vague in places, which is exactly why the later iterations worked better.

The first few prompts were fine. The later versions were much better.

That was probably the most useful reminder in all of this. Prompting is not magic. It is iteration. The more specific I got, the more usable the output became.

Tone mattered immediately. If you leave it vague, you get the usual polished corporate filler. If you tell it to keep the tone friendly, practical, upbeat, and community-oriented, it gets a lot closer to something a real person would actually post.

I also had to be clear about the purpose of each section. The metadata health check was a good example. I did not want it to read like a compliance report or a scolding email. The point was to make gaps visible in a way that was useful, light, and maybe just social enough that people would fix their own items without being chased down.

I was also explicit about the data. File name, created date, contributor, and the relevant metadata columns from the library. No filler. No vague summaries pretending to be facts. No creative interpretation where none was needed.

If I am using AI against a shared content source, I want the boundaries to be obvious.

Guardrails First

Because this is a shared community library, I was very clear that Cowork was there to read and assemble, not “help.”

• No uploading.
• No editing.
• No renaming.
• No moving files.
• No deleting.
• No fixing metadata.
• No writing anything back to SharePoint.

If a step would require changing the library, stop.

That may sound overly cautious, but I do not think it is. Shared spaces only stay useful if people trust them. The last thing I wanted was for a side experiment to start making changes to content other people contributed.

I also tried to handle the less exciting edge cases up front. What happens if no one uploaded anything that week. What happens if the week was quiet. What happens if there is not much to celebrate except the fact that the library still exists and nobody broke it.

The answer was not to skip the post. It was to make the update still useful by leaning more on gaps, prompts, and reminders about what kinds of content would help.

That mattered too. A roundup should still have a purpose even when the numbers are unimpressive.

What It Produces

Using fictional sample data, Copilot generated a draft roundup in the structure I had defined so I could review the format and show the concept without pretending this was already running as a weekly habit.

The output is a Markdown draft built from the library data and shaped to match the structure I defined.

Once I was happy with it, I saved the prompt and instructions in a Markdown file so the process was shareable and either of us could run it without reinventing it.

That draft still gets reviewed by a human before anything would be posted. I might tweak the featured item, clean up phrasing, or adjust the emphasis depending on what was actually uploaded.

That part is not a flaw in the process. That is the process.

And to be clear, we have not actually started running this as a weekly published roundup yet. What I have built is the approach. The structure is there, the prompt is there, and the workflow is there. I also used Copilot to generate sample output so I could show what this would look like in practice and capture screenshots for demos or write-up purposes without pretending there is already a long-running publication rhythm behind it.

Why This Was Worth Doing

I could have forced this into a more traditional automation path. But for this kind of side-of-desk community work, that would have been too much overhead for the problem I was trying to solve.

What made Cowork useful here was not just that it could pull information from the library. It was that it could help shape that information into something readable, repeatable, and consistent without turning the whole exercise into another mini project to maintain.

It also made co-management easier. When more than one person is supporting a shared content space, having a repeatable prompt and structure helps keep the output aligned. The post does not swing wildly in tone depending on who had the time to touch it. It feels like one stream of communication instead of two people improvising in parallel.

That is a practical benefit, not a flashy one, but it matters.

Because really, this is not about replacing work. It is about protecting the work people are already trying to do.

A library like this only works if people contribute. And people are more likely to contribute when they can see that someone is paying attention, that uploads are appreciated, and that the content is actually being surfaced back to the community. A roundup helps do that. But only if it is light enough to maintain.

That is what I was trying to solve.

What I Learned

Probably the most practically useful section. It made metadata gaps visible without turning the whole thing into a scolding exercise.

The biggest lesson was to design the output before trying to automate anything.

The second was that prompting gets much better the minute you stop being vague and start being specific.

And the third was that consistency is its own value. Not glamorous value. Not “innovation theater” value. Just the very practical value of making it easier to keep a useful thing going.

That is where Cowork helped.

• Not by replacing judgment.
• Not by magically creating a community.
• Not by turning this into some grand content engine.

Just by making a good habit more achievable.

That is enough for me.

If You Want to Try Something Similar

If you manage a shared content space and you are trying to keep it active without turning it into another unpaid part-time job, this is the basic approach I would recommend.

• Start by writing the ideal update by hand.
• Turn that into a detailed prompt with clear sections, tone, rules, and source fields.
• Be explicit about boundaries, especially if the content is shared.
• Keep a human in the loop.

And refine the prompt until the process feels stable enough that someone can actually keep using it.

That is what I built here. Not a magical automation story. Just a practical way to make a useful community habit easier to maintain.