For the last thirty days, I kept a log.
Not a productivity experiment with rules and a spreadsheet and a big conclusion at the end. Just a simple note on my phone where I wrote down every time I used ChatGPT for something real, what the task was, how long it took, and whether it actually helped.
I did this because I've been writing about ChatGPT for a while now, and I realized I had no idea what my actual usage looked like. I had impressions. Impressions are useless. I wanted data.
Thirty days later, here's what the data says. Five things stayed. Four things failed. Three things I dropped entirely.
I'm including the failures, because that's the part nobody writes about.
Why I Did This
Two reasons.
First, I wanted to know if I was actually saving time or just feeling like I was. Those are different things, and the second one is much easier to achieve.
Second, I wanted to write an honest post about practical use. Not a "30 days of AI changed my life" post. Not a "AI is overhyped" post. Just: here's what happened when an ordinary office worker with two kids used this thing for a month and wrote it down.
The log is about 40 entries. Some days had three uses. Some days had none. That distribution itself is worth noting—this is not a tool I use constantly. It's a tool I use in specific moments.
What Stayed
These are the five uses that survived the month, showed real value, and are now part of my routine.
1. Meeting Notes to Action Items
Used: 11 times in 30 days
Average time saved: 20–25 minutes per use
Verdict: Still the single highest-value use.
This one isn't close. Eleven meetings produced messy notes, and eleven times the four-section prompt turned them into something usable in about twelve minutes instead of thirty-five.
The reason it works is that it's mechanical. Converting fragments into structured items is a task with a clear right answer, and it's exactly the kind of thing I'm slow at and ChatGPT is fast at.
The failures were all the same kind: it invented an owner or a deadline that wasn't in my notes. Caught them every time in review, but they happened often enough that the "do not invent" constraint is non-negotiable.
2. Email Rewriting
Used: 14 times
Average time saved: 8–10 minutes per use
Verdict: Reliable for tone, unreliable for judgment.
I used this constantly. Drafting an email, pasting it, asking for a cleaner version. It worked well almost every time.
The catch: it's good at how something is said and useless at whether it should be said. Twice I caught myself about to send something that was well-written and strategically wrong—too direct for the relationship, or too soft for the situation. ChatGPT doesn't know your office politics. It doesn't know that this particular manager reads terseness as respect and that one reads it as hostility.
Use it for the words. Keep the judgment.
3. Document Explanation
Used: 6 times
Average time saved: 25–30 minutes per use
Verdict: Genuinely excellent, with a big caveat.
Six documents I would have read badly and understood partially. Six times I got a plain-English explanation in about 15 seconds.
This was the biggest surprise of the month. I expected it to be useful. I didn't expect it to be this useful.
The caveat is the same as always: never paste confidential material, check your company policy, and take the flagged ambiguities to a human. The tool explains. It doesn't advise.
4. Task List Reset
Used: 18 times
Average time saved: 10 minutes per use
Verdict: Small benefit, big psychological effect.
Eighteen times I did my end-of-day brain dump and had ChatGPT sort it into categories. The time saved is modest—maybe ten minutes—but the effect on how I leave work is bigger than the number suggests.
The reason: it closes loops. Vague worries become specific questions. Things I was carrying around in my head get written down and sorted. I go home with an empty head instead of a full one.
That's not a productivity gain. It's a quality-of-life gain. Worth noting separately.
5. Family Scheduling
Used: 4 times
Average time saved: 15 minutes per use
Verdict: Useful, but only with iteration.
Four Saturdays planned. Four times the first draft missed something real—drive time, a repair window, a kid's energy level—and the second version worked.
The pattern is consistent: first draft is a mirror that shows you what you forgot to mention. Second draft is the one you use. If you're only doing one pass, you're not getting the value.
Privacy note stands: kids referred to by age, no names, no schools, no addresses.
What Failed
Four things that didn't work, despite my wanting them to.
1. Writing From Scratch
Attempted: 3 times
Result: Gave up on all three
Every time I asked ChatGPT to write something from nothing—a report section, a proposal paragraph, a message with no existing draft—the output was generic. Not wrong. Just empty. It read like it could have been written for anyone, which means it was written for no one.
The pattern is clear: ChatGPT is excellent at transforming existing material and mediocre at creating it. Give it something to work with. Don't ask it to start from a blank page.
2. Research
Attempted: 4 times
Result: Unreliable every time
I asked about a vendor's background, a technical specification, and two current-events-adjacent questions. Every time, I got confident answers that I couldn't verify and didn't fully trust.
This isn't a knock on ChatGPT. It's a statement about how I use it. If I can't check the answer, I shouldn't be relying on it.
I've stopped asking it factual questions. If I need to know something, I look it up.
3. Creative Brainstorming
Attempted: 2 times
Result: Nothing usable
I asked for ideas twice—once for a family activity, once for a way to present information at work. Both times I got lists of obvious options I'd already considered.
The problem is that brainstorming works best when the ideas are specific to your situation, and ChatGPT doesn't know your situation. So it produces the generic version. Which is fine, if you've never thought about the problem. Useless if you have.
4. Anything Time-Sensitive
Attempted: 3 times
Result: Wrong or outdated twice
Current rules, recent changes, anything that happened in the last year or two. Two out of three attempts produced outdated information presented with full confidence.
I've stopped using it for this entirely.
What I Dropped
Three things I used early in the month and then stopped.
1. Asking It to Make Things "Better"
"Make this better" produces vague, padded output. Every time. If I want something improved, I have to say what specifically—shorter, warmer, clearer, more direct. Without a specific direction, ChatGPT adds words instead of improving them.
2. Long Multi-Part Requests
"Summarize this, pull out action items, and suggest next steps" produces three mediocre outputs instead of one good one. Every time I tried it, I ended up redoing the work in separate passes.
One task per prompt. Always.
3. Casual Conversation
I tried it twice for no real reason. Both times it was fine and pointless. If I want to talk to someone, I'll talk to Jamie.
The Test Card
Task: Determine which ChatGPT uses actually earn their place in a daily workflow.
Prompt or workflow: Log every use for 30 days—task, time spent, time saved, verdict. Review at the end.
Starting conditions: Ordinary office work, family logistics, and no formal productivity system.
Total entries: About 40 across 30 days. Roughly 35 of them in the five categories that stayed.
Time before ChatGPT: Not measured across the month, but per-task estimates range from 10 minutes (email) to 35 minutes (meeting notes).
Time after ChatGPT: Per-task savings of 8–30 minutes depending on the task. Total monthly savings roughly 8–10 hours, though I'd treat that number loosely.
Editing required: Consistently moderate. The output is a draft, not a finished product, in every category that worked.
What went wrong: It invented details in meeting notes (caught in review). It produced outdated information on time-sensitive questions (stopped using it for those). It wrote generic content when asked to start from scratch (stopped doing that).
Privacy notes: Same rules throughout the month—names replaced with roles, client identifiers removed, financial figures generalized, no family names. Checked against company policy before using external tools on work material.
Who should not use this method: Anyone in a workplace that prohibits external AI tools. Anyone who won't do the review step—every failure mode I found was caught in review, not by the tool. And anyone hoping this replaces judgment; it doesn't, and the month made that clearer than ever.
What I Learned From the Log
Three things, and they've changed how I talk about this tool.
The Value Is in Transformation, Not Creation
Every single high-value use involved taking something that already existed and reshaping it. Notes into action items. A draft into a cleaner email. A dense document into plain English. A messy head into a sorted list.
Every failure involved asking it to produce something from nothing.
That's the pattern. I'd bet it holds for most office work.
The Savings Are Small and Frequent
I didn't save three hours in one afternoon. I saved 20 minutes on Tuesday, 10 minutes on Wednesday, 25 minutes on Thursday. The total is meaningful. The individual instances are not dramatic.
This is why the "ChatGPT will change everything" framing doesn't match my experience. Nothing changed dramatically. A lot of small things got slightly faster.
Review Is the Whole Safety Net
Every failure I caught, I caught in review. The invented deadline, the outdated rule, the email that was well-written and strategically wrong—none of those were flagged by the tool. All of them were flagged by me reading carefully.
Which means the value of this tool is directly proportional to how much attention you pay to its output. Use it carelessly and it's a liability. Use it carefully and it's a modest, reliable time saver.
The Bottom Line
Thirty days. Five uses that stayed, four that failed, three I dropped.
The honest summary: ChatGPT is a good tool for reshaping things that already exist, a bad tool for creating things that don't, and an unreliable tool for anything current. It saves me somewhere around eight to ten hours a month in small pieces, and none of those pieces feel dramatic.
That's not a headline. It's just what the log says.
I tried it at my desk for thirty days so you don't have to. Useful beats impressive—and after a month of writing it down, I finally know which is which.
