The process of making my new website using GPT 5.6 sol.

Why did I need a new website?

My website looked like AI slop and needed massive improvements. With the new release of GPT 5.6 sol ultra thinking, I decided a perfect task for it, was to improve my website.

As someone who has used 5.3 billions codex tokens in the past few months, I think I've used enough AI to have a fair say on the quality of GPT 5.6 sol of making my website.

How do you make a GOOD looking website using AI?

Making a good website is not as easy as just asking AI for a website, it takes a while of meticulous planning, testing and tweaking slowly.

My steps

Step 1: Making a new skill.

I first started by asking AI to make a new skill for building my website. I asked GPT 5.6 sol to do this asking it to search sources online for signs of AI websites to make an anti-ai-slop website. It worked for 57m and 40s, making a short skills called build-distinctive-websites. The size of the skill was disappointing, however after reading it, it was high quality and had a very specific style, it also made an example site, and it looked fairly good. The prompt I used was

Search the inherent to find the best online sources for signs of AI websites, colour themes, button styles, literally everything, search for a good while and create the ultimate skill for how to build a website that DOES NOT LOOK AI GENERATED [$skill-creator] then install it once done using [$skill-installer]. It's not just website HTML and CSS, it's also the actual designs of website that make it clear if something it AI, so please take that into account when creating this skill.

Step 2: Re-design plan.

Since I already have a working website with content, I didn't need to create a new website from scratch and it was just a case of modifying the existing website. It's important when doing any large task with AI all at once, that you use plan mode, although it seems like you are sacrificing speed and token consumption for quality, but that's completely wrong. In reality, not only will the AI build faster (because it knows what it's doing before hand), but it also uses less tokens as it knows in what file things are present so it doesn't need to re-check. There are no downsides to just asking AI to plan out first. The prompt I used for designing was basic, but got the point across. With older AI models, I would have first uploaded the prompt to ChatGPT to optimise the prompt further, however GPT 5.* has generally been good enough, and I was fairly confident I could put my faith into GPT 5.6 sol being the most capable model.

Please completely re-design the home page of my website please, generate an image of what the styles should look like, don't necessarily need to change text or content much, just a re-design, it should be light mode, new colours, new layout, new everything, literally yoink delete all the html and css currently and re-design the whole thing, sick animations, transitions etc and most importantly use the [$build-distinctive-websites] skill please, and only make a design plan, don't make the website changes yet please.

As you can see. The prompt is not very well optimised, but this is the stage we are at with AI, all the points I wanted came across and since we are planning, it means it's less important to get it perfect.

Working for 23m and 18s, the AI was able to make this plan. Previous models such as GPT 5.5 or 5.4 would have only worked for on average for less than 5 minutes on this task, the sheer time increase is crazy. The plan made was short and includes an AI generated screenshot of what it might look like. It looked good enough. Not AI slop that's for sure.

Step 3: Implementation.

At this point, I was happy with the screenshot so I just said Yes, implement this plan, however if there was something that I didn't like, then I would have mention in this message to implement but do this different. If there were many things wrong, or if I just wanted to do something completely differently, I would ask for the plan to be improved.

51m 1s later, we had a working website. It did NOT look like the plan. So now it's time for some tweaking.

Step 4: Tweaking.

Tweaking is very important. There were many small things that needed changing, I uploaded screenshots of what I didn't like, told it straight up, and even went and modified the cod myself. The first tweak took 1h 20m 14s (longer than initial implementation).

The next tweak I did was to change and fix the contact system. This was not mentioned in the original prompt, so it's less of a tweak and more of a new feature. This took an additional 52m 3s to implement.

27m 8s for the next 3 tweaks combined time. And I was finally happy with the website. Many changes by hand were made as well, changing code by hand is often much better than using AI, however I only spent short times on these.

Step 5: Testing.

Testing happens in parallel with the previous tweaking step. It just means you (and the AI if you want) goes through and test all features fixing all the small inconsistencies and checking that forms submit, links work, checking edge cases. Generally with modern AI doesn't make many mistakes, but it's important to check.

The final website.

The total time that AI ran for was almost 6h. It might seem like a while, but remember that you don't sit there waiting for it to complete. In total it would probably have taken under 1h of work from me including test, deployment, prompt writing. So far, writing this has taken longer than the time I spent actually making the website, yet it is exactly how I would have done it. It took a total of 350million tokens which if payed for via the OpenAI API, could cost as an estimate of $2,144 with the following prices:

• Uncached input: $5/million
• Cached input: $0.50/million
• Cache writes: $6.25/million
• Output/reasoning: $30/million

and the following estimate

• 85% input, 15% output/reasoning
• 70% of input successfully read from cache
• 25% normal input
• 5% cache writes Thats the estimate. That is mad. GPT 5.6 sol massively overthinks. Obviously I did't spend $2,144 on updating my website, it used a small portion of my weekly Codex Credits on the 5x Pro plan which costs $100 per month.

Conclusion

GPT 5.6 sol is very good if used well. It can make decent looking websites, but it is slow, and it uses far too many tokens.