I have a complicated relationship with the phrase "automation."
Every time someone at work says it, I get a little nervous. Not because I'm against it—because I've watched three different automation projects roll through our department, and two of them quietly died after a few months. The third one is still running, but only because one person manually fixes it every week and nobody talks about that.
So when I built my weekly report workflow, I deliberately didn't automate anything.
No scripts. No macros. No connectors or integrations or anything with a subscription. Just a repeatable process that takes about twenty-five minutes and doesn't break when something changes.
It's been running for about five months. Here's the whole thing, including the parts that took me a while to get right.
What the Report Actually Is
Context matters, so let me be specific.
I put together a weekly status report for my manager. It covers project progress, blockers, and anything that needs her attention. It's not a huge report—maybe 400 words. But the information comes from four places: my own notes, two spreadsheets, an email thread or two, and whatever's rattling around in my head.
The old version of this process took me anywhere from forty minutes to an hour and a half, depending on how much I'd let pile up during the week. And it had a bad habit of turning into a research project, where I'd start writing and then realize I didn't actually know the current status of something and have to go find out.
The new version takes about twenty-five minutes, and most of that is me reviewing rather than assembling.
The Workflow
Five steps. Same order every week. I do it Thursday afternoon, because the report goes out Friday morning and I want a day of buffer if something's off.
Step One: Collect the Raw Material (5 Minutes)
I gather everything into one place. Not write anything—just collect.
Notes app: everything I jotted down during the week
Spreadsheet A: project status tracker
Spreadsheet B: the numbers sheet
Email: anything flagged or unresolved
I don't organize any of it yet. I just make sure it's all in front of me.
The reason this step matters: the old version of this process had me hunting for sources while writing, which is how a twenty-minute task becomes an hour. Collect first, write second.
Step Two: Sanitize (3 Minutes)
Same rule as always. Before anything goes near ChatGPT, I strip identifying details.
Client names become "Client A," "Client B"
Coworker names become "Person A," "Person B"
Account numbers, dollar figures tied to specific accounts, and anything marked confidential get removed entirely
Project codenames get replaced with generic descriptions
I do this in a scratch document, not in the original files. Never edit the source.
Three minutes is about right for this. If it's taking longer, that's a signal the report contains more sensitive material than I should be pasting anywhere, and I do it manually that week.
Step Three: The Assembly Prompt (4 Minutes)
Here's the prompt. I've been refining it for months, and this version is the one that finally stuck.
"Below are raw materials for a weekly status report. Turn them into a draft with these sections:
Progress This Week — what actually moved forward
Blockers — anything stuck, with what it's waiting on
Needs Attention — items requiring my manager's input or awareness
Next Week — what's planned, only if it's already decided
Rules:
Keep the whole thing under 400 words.
Do not invent status, dates, owners, or numbers. If something is unclear, write 'UNVERIFIED' rather than filling it in.
Do not include routine items that require no action from anyone.
Do not use the phrase 'moving forward' or 'circle back.'
If the same item appears in two sources with different status, flag it under Needs Attention.
Format: Short sentences. No bullet points longer than one line."
Two rules in there are doing more work than the rest.
"Do not include routine items that require no action from anyone" — this one cut my report length roughly in half. The first drafts ChatGPT produced were padded with things like "attended standing meeting" and "reviewed inbox." Technically progress. Useless to a reader.
"If the same item appears in two sources with different status, flag it" — this catches discrepancies between my notes and the spreadsheet, which happens more than I'd like to admit. It's the same contradiction-detection logic I use for email threads, and it's saved me from sending a report with two different timelines in it.
Step Four: Review and Verify (10 Minutes)
This is the biggest block, and it's non-negotiable.
I read the draft with the source materials open next to it and check every claim. Anything marked UNVERIFIED gets resolved or deleted. Anything in Needs Attention gets a second look—does my manager actually need to know this, or am I just escalating?
Ten minutes. It's the most important ten minutes in the whole workflow.
Step Five: Final Format (3 Minutes)
I copy the approved draft into the email, add the subject line, and check the attachment. Then it sits as a draft overnight.
I don't send it Thursday. I let it sit, glance at it Friday morning, and send. That overnight gap has caught more errors than any prompt tweak I've ever made.
The Test Card
Task: Produce a weekly status report from scattered notes, spreadsheets, and email threads.
Prompt or workflow: Collect all sources, sanitize identifying details, run the four-section assembly prompt, verify every claim against the source, let it sit overnight before sending.
Starting conditions: Notes app entries from the week, two spreadsheets, one or two email threads, and no existing draft.
Time before ChatGPT: 40–90 minutes depending on how much material had piled up, with occasional multi-hour delays when a status needed to be researched mid-write.
Time after ChatGPT: About 25 minutes total—5 collecting, 3 sanitizing, 4 prompting, 10 verifying, 3 formatting.
Editing required: Moderate. The draft is usually structurally correct but needs verification work and some trimming.
What went wrong: Early versions padded the report with routine items and occasionally stated a status that contradicted a source. Both fixed with explicit constraints. The overnight delay was added after I nearly sent a report with a stale number in it.
Privacy notes: All names, clients, accounts, and confidential project identifiers replaced before pasting. Sanitization happens in a scratch document. Weeks where the material can't be meaningfully anonymized are done manually.
Who should not use this method: Anyone whose report contains information that can't be described generically. Anyone in a workplace that prohibits external AI tools. Anyone who won't do the ten-minute verification step—an unverified status report is worse than a late one.
Why I Didn't Automate It
I want to explain this, because "I built a workflow" and "I didn't automate it" sound like they contradict each other.
They don't. Here's my reasoning.
Nothing here is stable enough to automate. The report format changes. The sources change. Every few months, my manager asks for something different. An automation built around last quarter's format becomes a liability this quarter.
I need to read the material anyway. To write a good report, I have to actually know what happened this week. An automated system would produce a report I didn't fully understand—and the first time someone asks me a question about it, I'd be stuck.
The value is in the verification. The ten minutes I spend checking claims against sources is where the quality comes from. Automating the assembly while keeping the verification would save maybe four minutes. Not worth the maintenance.
Automation breaks silently. I don't. When my process fails, I notice. When a script fails, it produces a wrong report and nobody catches it for three weeks.
So I built something repeatable instead. Same steps, same order, same time every week. No software, no subscriptions, nothing to maintain.
What Actually Changed
The honest version.
It stopped being a research project. The old process had me writing and researching at the same time, which meant every report was a surprise. Now the collecting happens first, and by the time I sit down to write, I already know what I'm working with.
I stopped dreading Thursdays. This is the one I didn't expect. I used to put off the report all week because it felt like a big task. Twenty-five minutes doesn't feel like a big task. So I just do it.
Fewer questions come back. My manager used to send follow-ups asking for clarification on two or three items per report. That's down to maybe one every other week. The "Needs Attention" section is doing the work there—I'm flagging things before she has to ask.
The overnight gap is the real hero. I can't overstate this. Reviewing a draft the next morning is qualitatively different from reviewing it five minutes after writing it. I catch things I'd never catch in the moment.
The Bottom Line
You don't need automation software to make a recurring task manageable. You need a repeatable process, a prompt that stops the most common failure modes, and enough verification that you actually trust the output.
Mine is five steps and twenty-five minutes. No scripts. Nothing to maintain. It works because it's boring, and it's boring because I've stopped trying to make it clever.
I tried it at my desk so you don't have to. No magic—just a workflow that survived Monday, and Thursday, and every Friday for five months running.
