8 Insane ChatGPT Work Use Cases That'll Save You Hours

One of these jobs took 1 hour, 7 minutes and 44 seconds, and I wasn't at my desk for any of it. Another rewrote 130 YouTube descriptions in a single run, each one with its own tracked link. A third built a live reporting dashboard out of a screenshot I'd found on Google Images.

They're the eight ChatGPT Work use cases I keep coming back to, and every one is a job I actually ran, filmed from start to finish, with the bits that went wrong left in.

They all run in ChatGPT Work, which is the version of ChatGPT that lives in the desktop app rather than your browser. It can see the files you're already working on, it has a browser of its own, and it can go and do things inside the tools you use every day. Instead of handing you an answer to copy out, it does the job and gives you the finished thing.

How to Set Up Your ChatGPT Work

There are two settings worth sorting out before you try any of this, because several of the use cases below simply won't run without them.

Use the dedicated ChatGPT desktop app rather than the web version. On desktop it can reach the files and apps on your own machine, so some of what follows simply isn't available in a browser tab, and one of the jobs below involves 35 megabytes of spreadsheet exports that the web version would refuse outright.

Then, once it's open, check that ChatGPT is selected in the top left rather than Codex, and use the toggle at the top to switch from Chat to Work.

The toggle is worth pausing on. Chat mode answers your questions, whereas Work mode goes off and does the job itself, and everything below assumes you're in the second one.

There are a couple of terms coming up that are worth defining now, because they're easy to confuse and the difference between them changes how you'd set a job up. A plugin is a capability you switch on for a particular job, like spreadsheets or data analytics. A skill is a set of written instructions you save once so you don't have to explain the same job every time. Where a tool has no official connection to ChatGPT Work, you can write a skill that teaches it how to connect, which is exactly what I did for our YouTube channel and our link shortener.The eight use cases

1. Building a live reporting dashboard from a screenshot

This is the one that surprised me most while I was testing. I wanted a dashboard so the team and I could see how our YouTube videos were performing, without anyone having to open Studio.

I started by searching Google Images for "YouTube analytics dashboard" and finding a layout I liked, then pasted that screenshot straight into ChatGPT Work and asked it to rebuild the design as a live site pulling real numbers from our channel and from BigQuery, which is where we keep the rest of our YouTube data. It came back with our channel logo, the headline stats, daily view breakdowns, average view duration, a per-day view showing which release days actually perform for us, and a table of recent videos.

It didn't get it right first time, and that's worth saying. In the first version, hovering over the charts showed no numbers, so I told it so and it worked for another four minutes and fixed it. Compare that to the old route: a ticket to the BI team, who have a backlog from every department, and a two to three week wait if you're lucky.

2. Downloading and organising files from a website

ChatGPT Work has its own browser built into the desktop app, which you get to from the side panel at the top right.

We use a consistent illustration style for the feature images on this blog, and I wanted a batch of them saved locally so I could later teach ChatGPT Work to make more in the same style. Rather than right-clicking twenty images one at a time, I gave it the blog link and asked it to download the first 20 feature images, save them to a folder, and build a CSV pairing each article title with its image.

The interesting part was watching how it did it. Rather than clicking each image like a person would, it wrote itself a short piece of JavaScript and downloaded the lot in one go. A few minutes later I had all 20 images and the CSV.

That's a small example, but it points at something bigger. This is how you get at tools that have no official connection to ChatGPT at all. Put it on a schedule, have it log into that platform once a month when the invoices land, download them and file them where they need to go.

3. Building an Excel cohort model with real formulas

Nobody enjoys building cohort analyses, so I gave it a genuinely awkward one. I prepared five exports from five different systems: member stats from our Circle community, their activity log, data from the platform that runs our live events, our Stripe transactions so it knew who was paid and who was free, and an export from the tool we used before we migrated.

The instruction that made the difference was insisting on formulas. No hardcoded numbers anywhere, because a hardcoded number is one you can't check. Every cell had to be a real Excel formula I could open and trace.

It worked for 1 hour, 7 minutes and 44 seconds while I did something else. The exports came to around 35 megabytes, which the web version would have rejected outright. What came back had tabs for all the raw data, a retention tab with the cohort analysis, the same view split by source so I could compare YouTube against LinkedIn ads against organic, and a findings tab with editable graphs pulling live from the model.

One thing worth knowing if your company already has a spreadsheet it trusts. Tag the spreadsheets plugin and you can give it a template, so instead of inventing a model it fills in yours.

4. Researching 25 companies with verified buying signals

I wrote a skill that teaches ChatGPT Work how we find companies who might want AI training, so I don't have to explain the criteria every time I run it.

When I run it I give it the region, the signals to look for, and how many companies I want. The signals we use are things like currently hiring for AI, automation or ops roles, or having recently announced an AI initiative, at companies between 50 and 500 people. It also has our full client list so it knows who to exclude.

It ran for over 20 minutes, and the reason it took that long is that it split the job up. It created sub-agents, which are essentially copies of itself working in parallel, one covering Berlin and Hamburg, another on Munich, Vienna and Zurich, another on the UK and Ireland, and one more just checking we weren't about to contact existing clients.

What came back was a spreadsheet of 25 companies, each with a named decision maker, their title and their LinkedIn profile. The part I'd insist on for any research job is that every single signal came with the exact URL behind it, so I could click through and verify the claim myself rather than taking its word for it. One Berlin company had just opened a principal AI transformation role to set company-wide targets, which is about as clear a moment to make contact as you'll get.

5. Filing receipts and logging expenses automatically

This one is small and it runs every week without me.

I keep a folder on my computer called receipts. Whenever I spend something I photograph the receipt and drop the picture in. A skill picks it up, reads the image, pulls out the amount, the date, the supplier and the category, adds a row to our expenses spreadsheet, then moves the file into a processed folder sorted by the date on the receipt rather than the date I photographed it.

The whole thing takes a photograph and a few seconds of my attention, and come tax time everything is already filed in the order an accountant wants it.

6. Editing 130 YouTube descriptions in one run

A lot of our website traffic comes from the links in our YouTube descriptions, and we wanted to add a new one across the whole back catalogue. That's over 130 videos, and every link we use is a custom short link with tracking parameters so we know which video sent someone to the site.

Doing that by hand means creating 130 individual tracked links and then editing 130 descriptions in the YouTube dashboard. Realistically I'd never have done it.

First I had it list every public video and save all the existing descriptions to a file. That step did two useful things: it learned the exact structure we write in, and it gave us a copy to restore from if the live edits went wrong. Then I wrote the rules. If a description already starts with a lead magnet link, put the new one underneath. If there's no call to action at all, make it the first line. If it starts with a subscribe prompt, replace that entirely.

Our link shortener has no official connection to ChatGPT Work, so I wrote a skill teaching it how to create the links itself, with the video ID passed in so each link tracks separately. I tested it on the 20 to 30 videos we published this year before letting it near the rest, which is the step I'd recommend to anyone doing bulk edits on something live.

7. Generating on-brand images inside a workflow

You already know ChatGPT can make images. The useful part is having it do that inside a job you needed doing anyway.

Having downloaded our last 20 feature images in use case two, I asked it to study them carefully, note the exact format and resolution we use, and then produce images for three new articles. I asked for three options each rather than one, because image generation rarely lands first time and three gives you a decent chance of liking one.

It worked for 12 minutes and returned nine images, all normalised to our standard format and matching the house style closely enough to use.

The detail I didn't ask for and now rely on: it saved every prompt it had used into a separate file. So when an image goes in a direction I don't like, I can read the prompt that produced it and see exactly where it went wrong, rather than guessing.

8. Starting a job at your desk and finishing it on your phone

The last one is the built-in remote feature, and it changes how the rest of these fit into a working day.

In the ChatGPT app on your phone, go to Remote in the left-hand menu, which sits under More if you can't see it. It asks you to authenticate, Face ID in my case, then loads the ChatGPT Work sessions you started on your computer.

I used it on the session that was rewriting our YouTube descriptions. I'd kicked it off at my laptop, and while I was out I could see it had finished, so I added a note asking it to run for all our 2025 videos as well and sent it. The task started running on the laptop back at the office.

These jobs are long. Several of the ones above run for twenty minutes to an hour, which is far longer than you'd sit and watch. Your computer does need to be switched on and awake, since the desktop version is the one with access to your files, but you don't need to be sitting at it.

What these eight have in common

Read back through them and the pattern is the same every time. None of these are jobs where I asked a question and copied out the answer. In each one I handed over a whole task, went and did something else, and came back to a finished thing I could check.

That gap is where most companies are stuck right now. McKinsey's 2026 survey found 44% of organisations have AI scaling across the enterprise, which means the majority still don't, even though four in five report individual productivity gains. Plenty of people are getting value out of asking questions. Far fewer have moved to handing over jobs.

The other pattern is that I still had to check everything. The dashboard was wrong first time. The lead research is only useful because every claim has a URL attached. The bulk edit was tested on 30 videos before it touched 130. Handing over the work doesn't mean handing over your judgement.

What to hand over first

Don't start with the most impressive job on this list. Start with the one you redo most often, where you already know what a good result looks like, because that's the one where you'll spot a mistake immediately.

Give it your context so it isn't guessing, run it once while you watch, then build a skill for it so you don't have to explain it from scratch next time. That's the on-ramp, and from the following week it should be running without you.