It has been with much difficulty that I’ve been able to wrap my head around elements and frankly, I haven’t fared too well… Now I am using ChatGPT MCP to try and complete the project I had started back sometime ago. Before I got a quarter of the way down the homepage I found out I had ran out of tokens (or whatever) on my ChatGPT plus account. It looks like a pro account at $100 per month is the next step.
My question is, what is the most economical way to use AI to complete my 34 page (I think it is) website?
I think the token usage is something they might be still working on. I was about to open a discussion about the same thing. I was using the built-in ai linked to my Claude api and asked it to design a header section just to see how it would do and burned through 2 million tokens and it couldn’t even finish because it ran my account dry.
Have you looked into running your own local LLM? Sort of depends upon your Mac, particularly how much RAM you have, but LMStudio running Gemma (a Gemini variant) or QWEN (Chinese open source) seems to work fine.
This, by the way, is why everyone was scooping up Mac minis stuffed with RAM but not much SSD prior to the current model. You’d spend maybe US$1000 to own your own AI server.
How are you using AI in Elements? With the MCP option, you are in your AI app (Claude, ChatGPT, etc.) telling it what to do. That utilizes your AI subscription usage limits (Plus, Pro, etc.).
The MCP option should be less costly than the in-app AI Assistant.
The AI Assistant that you chat with in the app can be a bit more costly as it uses an API key you generate with your AI provider, and it pulls from some account credit top-ups you added at your AI provider (which are separate from your AI subscription costs). If you are using a more expensive model in the selector, costs can add up quickly.
There is a helpful comparison chart here for OpenAI Compare models | OpenAI API . You can compare the costs between the different models.
You can switch back and forth between the models depending on your task if you are using the AI assistant in the app. For example you can use the lower cost models for stuff that’s not too complex, and then the more intelligent models for some of the harder work.
The problem you’d have with that is that you’ll have to use smaller models. So instead of using Qwen 3.8 27B you’d have to use Qwen 3.5 9B. The B is “billion”, as in the number of parameters that are being handled as the model navigates the tokens. That might make a difference in how well the AI does the jobs you ask it. That’s a variable thing, as different asks will have different results.
Not for me. I want as many panels open and as big a WYSIWYG display in the middle as I can get. So once I open Page/Resource/Page Layout on the left and the Inspector on the right, I’m out of panel space. If you have your AI running by itself as the only other program, you can quickly Command-Shift-Tab between the Elements and AI window.
Thank you for the information! I think I prefer to use an AI assistant for the website so I will dig in and choose my best option. What I’ve seen so far it’s very helpful!