I’m always tinkering with AI tools, looking for ways to make my work faster or better. It’s both helpful and satisfying to hit on a strategy that works.
But a lot of that tinkering is trial and error, and that takes time.
A while ago, I tried to get ChatGPT to review entire webpages—text, images, formatting, the whole thing. It’s something a person can do just by opening a browser and viewing the site, so it didn’t seem like it would be a hard problem. But because of how ChatGPT retrieves and processes web content, the results were inconsistent. Sometimes it worked, and sometimes it didn’t.
I spent a long time iterating, trying different inputs and prompts to get ChatGPT to behave the way I wanted. Partly because I thought it should be possible, and partly because I can be pretty determined to make things work the way I want them to.
I should have stopped to ask myself, “Is this a good use of AI?” and “Is this a good use of my time?”
When does using AI actually save time?
These questions can help you decide whether using AI will actually save you time on a task or workflow:
Is this simple enough that I could do it myself faster than I could create a decent prompt?
If the task is quick and straightforward, the overhead of explaining it to AI may outweigh any time savings.
Is this a one-time task, or something I do repeatedly?
It’s usually only worth investing time in prompts or workflows if you’ll reuse them. Occasional tasks rarely justify the setup.
Am I going to have to significantly edit or reformat the output before I can use it?
If the output needs heavy cleanup or doesn’t fit your format, you may just be shifting the work, not reducing it.
Will I need multiple rounds of prompting to get something workable?
A few iterations can be fine, but if it takes many tries to land on something usable, the time adds up quickly.
Will using AI for this task create more work downstream?
Fixing inconsistencies, aligning with other materials, or rechecking decisions can extend the workflow.
How much time will I need to spend verifying the output?
If accuracy matters and you need to double-check most of it, the added review can take as long as doing the whole thing yourself.
Has someone else already figured out a reliable way to do this with AI?
If you’re building from scratch, the learning curve can be steep. Starting from an existing approach can make the difference between saving time and losing it.
Sometimes the best move is to step back
Often, it’s not a matter of “use AI or don’t.” It’s more like, use it for one part and not the rest. For example, maybe it helps you get unstuck or sketch an outline, but you still do the actual writing.
It’s also worth noticing what you actually like doing. If you delegate the parts you enjoy to AI, that’s not much of a win.
Sometimes the right move is just to stop. It’s easy to keep going because it feels like you’re close, or because you think it should work. I’ve definitely lost time that way. But I’ve found that pushing it usually costs more than it saves.
I’ve moved away from assuming AI can make anything better. Instead, I check whether it’s actually helping me move faster. My time is valuable, and I want to get the most out of what these tools have to offer.