How I Learned to Stop Wasting AI Credits on My Website
Think once. Prompt once. Publish once — and save the fuel for the road.
I’ve been building NixDrones.com using ChatGPT Sites, and I’ve discovered something rather interesting along the way. AI credits are a bit like fuel. It’s not necessarily the journey that empties the tank. It’s all the unnecessary running around. When I first started making changes to the website, it was very easy to work like this: create a blog post. Then add these photographs. Then change the layout. Then add the sharing button. Then publish it. It feels perfectly logical. But every time I send another instruction, the AI potentially has to go back into the website, understand what’s there, work out what needs changing and perform another job. So I’ve been looking into how ChatGPT’s credit system actually works and, more importantly, how I can make NixDrones.com without chucking credits into the digital furnace.
The answer is surprisingly simple
Plan first. Build second.
Instead of giving ChatGPT Sites five separate instructions, I’m going to try giving it one properly thought-out instruction describing exactly what I want the finished result to be.
For example, when I want to publish a new Nix Story, the job shouldn’t need to become a redesign of the entire website. The article can be written first. The photographs can be selected. The title and structure can be decided. Then Sites gets one instruction.
Create the Story. Add the media. Use the existing design. Publish it. Don’t alter anything else.
Done. Or at least that’s the theory!
Why bother?
The important point isn’t simply how many words are being written. Writing 500 words isn’t necessarily the expensive part.
Asking an AI agent to explore an entire website, understand how everything works, change something, check it, change something else and repeat the process potentially requires considerably more work.
That led me to a new way of working.
My new Nix rule is simple: ChatGPT writes. Sites installs.
I’ll use ChatGPT to work out what I actually want before asking Sites to touch the live website. For a new Nix Story that means getting the words, photographs, title and requirements sorted first. Then I’ll send one complete instruction to Sites.
I’m also going to stop asking it to reinvent things that already work. If the Stories design works, use it. If the navigation works, leave it alone. If the homepage isn’t involved in the job, don’t touch it.
Sometimes the most useful instruction you can give an AI isn’t what to change. It’s what not to change.
Sometimes the most useful instruction you can give an AI isn’t what to change. It’s what not to change.
Less tinkering, more travelling
There’s a pleasing irony here. I’ve spent years trying to make computers faster, smarter and more capable, and apparently I’ve now reached the stage where I need to tell one: ‘Just do this bit and leave everything else alone.’
Progress.
But if it means fewer credits disappearing into repeated website edits, more time actually creating stories, taking photographs, flying drones, riding Torque and heading off in Vanilla, I’m all for it.
NixDrones was never supposed to become a full-time web-development project. It’s supposed to record the adventures.
So that’s the new plan: think once. Prompt once. Publish once.
And save the fuel for the road.