5 ways I actually use ChatGPT Work in my small business

What happened

Here at ZDNET, we've been diving into ChatGPT Work and its cousin, Claude Cowork, the agentic multi-step AI tools that are providing substantial productivity benefits to the folks who use them. Over the past few months, I've been putting both tools to work. In this article, I'll spotlight five ways I use ChatGPT Work in my business.

I'd only have to tweak some colors and formatting. ChatGPT Work took 13 minutes to do the job, while I did some chores around the house. That's a big help. Of course, ChatGPT does have some excellent visualization tools. So when I turned back to ChatGPT Work and asked it to use its image features outside of PowerPoint, it gave me a much more attractive (and ZDNET-style accurate) slide.

This use of ChatGPT Work was pretty astounding. I'm working on several Apple apps that will eventually be offered on the App Store and use in-app purchasing. Developing a pricing strategy is fairly complex, especially since my apps run on all Apple platforms (except the Vision Pro). You have to decide when to present offers, how, what pricing to use, what those entitlements include, and more.

Plus, you really need to understand what the app does to understand its user base and features. In my case, instead of writing a feature spec for ChatGPT Work to use as a starting point, I merely pointed the AI at the source code folder, told it to avoid making any changes, and read it.

I told it to learn all about what the app does from the source code itself. This approach works. I've used it for other projects. In each case, ChatGPT Work derives solid marketing feature lists directly from the code. The process saves a huge amount of time and ensures I don't miss anything important when writing feature summaries.

I've also used ChatGPT Work to identify comparable apps in the App Store and Google Play Store. I had the AI summarize the various pricing strategies the comparables use. Work went back and forth with me until we had a pricing strategy I liked. Then I had ChatGPT Work create a set of instructional prompts telling Claude Code about the pricing strategy, allowing Claude to implement it.

Why it matters

Keep coming back to ZDNET, because over the coming weeks, I'll also be doing deeper dives on individual projects using both ChatGPT Work and Claude Cowork. I use ChatGPT Work a lot for email support. Initially, connecting Gmail to ChatGPT Work (and Claude Cowork) freaked me out, but I soon settled on a safe pattern. I only enable the connection before I give the AI a fairly large assignment.

Then, I disable the connection immediately after. This routine creates added friction, but it helps me sleep better at night. One way I use ChatGPT Work with email is to find conversational history. For example, I get a tremendous amount of PR-related email about AI and about OpenAI specifically. I have conversations with the OpenAI team, but I also get their press releases.

I also get messages from all the other PR peeps who want to pitch quotes or perspectives from their clients. Digging through and finding relevant parts of conversations can be quite time-consuming. Getting help with this task is a strong use of the agentic workflow. One thing to note is that you often have to tune your prompts.

I started with this prompt: Unfortunately, ChatGPT Work limited the results to emails from OpenAI personnel who were in PR. OpenAI mostly uses an external PR agency, so it missed the bulk of my conversations. But a revised prompt provided more details about what I was looking for and worked well. Notice the red "Gmail" at the beginning.

That's the Gmail connector, which is active when you type @gmail at the beginning of the prompt. I've found that ChatGPT Work is fairly good at converting raw data into PowerPoint slides. The AI is strong with data but struggles with slide aesthetics. Most of the recent slides I've used this process on contain confidential data, so I can't share them here.

For this example, I pulled information from one of San Francisco's public datasets, the public record of fire incidents. That dataset is big, with almost 750,000 entries. A San Francisco fire battalion is an operational unit in SFFD that groups several stations under a single commander. For this example, I wanted to show the top 10 battalions with the most incidents.

After a bunch of permission requests, the AI produced this slide. To its credit, the AI understood that ZDNET is green, but then it used the wrong green. It also scaled the chart with really large margins inside the slide. But don't dismiss that output as the AI not working. It did the hard work. The tool took a web page and turned it into an aggregated chart.

What to watch

OpenAI's Codex could have done the same thing, but I use Claude with my Apple coding and Codex with my WordPress coding. All told, the AI did a huge amount of work and helped me plan an approach for a fairly challenging project. I'm overloading my battery backup devices in the Fab Lab. As 3D printers get bigger and more powerful, they take more power.

I have reached the point in the Fab Lab where my UPSs can't stand the load. So, I reached out to technology specialist Jackery. My colleague Adrian Kingsley-Hughes previously spoke highly about Jackery devices on ZDNET. I do most of my 3D printer coverage on my YouTube channel, but I always give Adrian's recommendations the highest priority.