ChatGPT dot: What I Delegated and What I Still Reviewed

Sunday gave me a reason to keep working with my ChatGPT dot. I had breakfast with my family, was out for the morning, and had lunch. Across that day, work kept moving. A fantasy football lineup was updated, and two versions of a reel came back. I gave direction and stayed involved. I just wasn't sitting at the desk doing every part myself.

I'd been using dot for about two weeks. This was the point when I could look at several actual results and see how I wanted to use it.

The useful part was coming back to work I could review.

What I handed to my ChatGPT dot

Let's start with fantasy football. Important business.

I wanted help with my lineup. It checked my lineup, researched injuries and possible replacements, and made the approved changes. That included adding and dropping a player and rearranging the lineup.

I can't promise that made me a better fantasy football manager. But I could check whether the task had been carried out. The requested change and the result were both visible.

That's a useful distinction when you're choosing your first task. If you can't describe what a finished result should look like, you're going to have a hard time judging what comes back. A small job you understand gives you a better place to start than a broad request to organize your entire business.

For a real estate professional, that might be turning approved meeting notes into a short list of follow-up actions. You already know the meeting. You can check whether the list preserves what was actually agreed. Keep that first attempt inside the conversation rather than asking it to send the follow-up.

An edit gives me something to react to

I had a video for my AI community. dot helped with the script, and after I recorded it, I handed the video back for editing. It returned two versions, one for Facebook and one for YouTube Shorts.

The Facebook version was about 28 seconds. The Shorts version was about 27. They came back with captions, tighter cuts, and an ending. I had actual files to watch.

That changes where my work starts. I can look at an edit and say which part to keep, what needs changing, or whether the ending feels right. I still decide what gets published. Having a version in front of me makes that decision much easier to work through.

It also helped put together the script and slides for the video accompanying this article. We changed the direction along the way. I didn't like the angle of the first script, so I pushed back and it changed.

The assistant could help produce another version. It couldn't supply my experience or decide what I meant. That was my contribution, and it stayed my contribution through the revisions.

An unfinished project can still move forward

The CRM work was more involved. I ran the first round of testing myself and made the first changes. Then I asked my ChatGPT dot to complete the remaining testing overnight.

By morning, I had a list of what worked, what didn't, which changes had been made, and which changes still needed to be made.

The project wasn't finished. It still needed direction from me, and there were things to work through.

CRM implementation is one of the services we offer clients. Getting help with testing let me come back to the results, make decisions, and keep the project moving.

For a property management owner, the same approach could start with property onboarding. Work through your existing steps with the assistant, then try the documented process on a fictional example before using live client information. Check the result with the person who owns that process. That's a proposed starting task, not a claim that an assistant can safely change your live workflows without supervision.

If you're choosing a process to try, my article on the READY framework for property management gives you a way to keep the experiment small. Use the information your organization allows, and keep control of changes to live systems.

Be careful with the time-saved story

I estimated the time these three jobs saved me at about 30 minutes for football, one hour for reel editing, and four hours for CRM testing. That adds up to five and a half hours. The work on this video's script, slides, and editing was extra; I didn't include it in that total.

Those are my estimates of what the jobs would've taken me, not measured savings. The CRM testing happened overnight, and my direction, review, and revisions still took time.

The useful result I can point to is more concrete: approved changes happened, edited files arrived, and the business-system work moved to another review point. Those are things I can inspect.

If you want to compare the time on your own task, include setup and checking the answer. Include the corrections, too. Try it more than once before deciding what the savings are.

I don't need a bigger claim to be interested. Coming back to something I can evaluate is already a different way to work.

Give your first task a finish line

You don't need to begin by connecting every account you use. Pick one task you understand, with a small input you're allowed to share. Your own non-confidential outline or fictional meeting notes can be enough for a first exercise.

Tell the assistant what you want back. A five-item action list is easier to check than "help me get organized." Name the source it should use, the length you want, and where it should stop. If you're asking for a draft, say to keep it in the conversation and not send it.

Then compare the result with the source. If the notes name a meeting date, did the assistant turn that into a deadline nobody agreed to? Did it give an unassigned task an owner? Did completed work come back as unfinished?

Those are specific corrections you can make. Ask it to fix the error and check the rest of the result for the same problem. Keep the corrected brief so your next attempt starts with what you've learned.

I wrote earlier about using my dot for meeting preparation. This weekend showed me a broader use: I can hand over a bounded piece of work, step away, and return to something worth reviewing. My judgment is still part of the job.

Check that your account has dot access before planning around it. Availability and the tools you can connect vary. OpenAI's current documentation is the place to check the product details.

I've put together a free guide called Getting Started with Dots. It includes a short interview to choose your first task, a reusable brief, a fictional practice exercise, and checks for the result. Get the guide, choose one job you already understand, and see what comes back. Start small enough that you can tell whether it helped.