5 ways I actually use ChatGPT Work in my small business to save time and money
AI News Desk
·
ZDNet
··
9 min read
Follow ZDNET: Add us as a preferred source on Google.
Follow ZDNET: Add us as a preferred source on Google.
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.
Also: How to use ChatGPT Work - and my top 10 tips for getting started with agentic AI
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. 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.
Also: 'I can't stop': 80% of developers find AI coding more addictive than helpful
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:
Summarize the last 5 messages from OpenAI's PR people and capture decisions, open questions, and what I should follow up on next.
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.
Also: This AI expert says the job apocalypse isn't coming, even if you're a coder - here's why
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.
Also: 75% of developers I surveyed prefer Claude Code - here's why they choose it over Codex
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. 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.
Also: I loved ChatGPT Desktop until OpenAI gutted it to make room for Codex and Work
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.
Also: I used Claude Code to vibe code a Mac app in 8 hours, but it was more work than magic
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.
Also: A low-tech solution from the past may be your best defense against AI deepfakes
Then I had ChatGPT Work create a set of instructional prompts telling Claude Code about the pricing strategy, allowing Claude to implement it. 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.
When Jackery indicated a willingness to be a project partner, the next step was to send them an assessment of my electrical infrastructure. Dammit Jim, I'm a computer engineer, not an electrician. My expertise tends to stop at 12 volts DC. House power has a lot more juice.
Also: You can earn your Google Al Professional Certificate for free - and I highly recommend it
I described each printer I have in the Fab Lab, then described the room's circuit infrastructure. I had previously audited those circuits , so I had a very clear picture of how much capacity they could handle.
But I was unclear about the aggregate load on UPS devices when the 3D printers were in operation. ChatGPT Work pulled in the specs for each of my printers. The AI also scanned the Jackery site for their devices to ascertain which devices would work best for my situation. Then it provided a detailed recommendation on how to replace my UPSs and why certain Jackery devices would be much better substitutes.
Jackery was kind enough to send four units. I've filmed the B-roll, and I'll install them in the lab this week. ChatGPT Work helped me understand and spec out what I needed, which will make the fabrication workshop much more functional.
A few months ago, we received a bill consolidation offer from what seemed like a trusted brand. One of my small company's best practices is financial scrutiny and observation. We're always looking at ways to keep the business on solid financial footing.
Like all companies, we have a fair amount of debt, scattered across a small selection of vendors. This loan promised to consolidate the debt and reduce our aggregate interest rate, which theoretically could save us a fair amount of cash.
Also: This new Claude skill saves you from bad contracts - and costs less than a lawyer
But we never take these offers at face value. Anytime bankers and lawyers get together in a room, you know there will be shenanigans. I fed ChatGPT Work this prompt:
Analyze this document very carefully. Look for any indications that it's fraudulent or, alternatively, any indications that make it clear that it is legitimate.
The offer wasn't fraudulent. It was from a credible company. But wowzer-bowzer were the terms ungood. They seemed favorable from the main copy. But the terms and conditions meant this loan had wildly unfavorable conditions. The AI caught the issue before we made a very innocent and potentially very costly mistake.
Another financial use was to have ChatGPT Work analyze our spending patterns with one of our vendors. I gave it this prompt:
Go through the last 12 months of emails and find all bills or receipts from [my vendor]. Provide me a report on what's been spent with [my vendor] and for what.
This process involved using email again, as well as data aggregation. When ChatGPT Work finished, I got a detailed, category-by-category analysis of all purchases and trends related to the vendor. We found a few things we could cut, but were also happy to determine that most of our expenses were reasonable and as expected.
Also: 'Specialists aren't required' anymore: How to stay valuable in an AI agent workplace today
I could have done this task myself, but ChatGPT Work turned what would have been a day of tedious and potentially error-prone work into a 15-minute run with solid results.
ChatGPT Work (and, likewise, Claude Cowork) can be incredibly powerful tools. As I've learned to use the technology over the past months, I keep finding more and more things to delegate to them.
But use caution. If you're giving the AIs access to sensitive information, keep a close eye on those permissions. But also keep an eye on what the tool produces.
I recently gave ChatGPT Work a deep-research assignment to dig into some public records in my town. The AI returned with a comprehensive report, except it juxtaposed the name of one of the city council members with the city manager. These aren't their real names, but if the city council member was named Ellen Ripley and the city manager was named Sarah Connor, the report referred to the city manager as Ellen Connor. At that point, the entire analysis became suspect.
My point is this: ChatGPT Work is powerful, but it makes mistakes. Don't trust AI implicitly. Use it as a tool, but know that it's not perfect. If you keep that attitude, you'll probably get good results.
You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter , and follow me on Twitter/X at @DavidGewirtz , on Facebook at Facebook.com/DavidGewirtz , on Instagram at Instagram.com/DavidGewirtz , on Bluesky at @DavidGewirtz.com , and on YouTube at YouTube.com/DavidGewirtzTV .
