As my research progressed and following my proposal presentation, my dissertation idea had pivoted to concentrate on current AI design tools. This would involve investigating how they behaved in use and what results they produced.
The first stage involved selecting the tools to be tested and developing a methodology to ensure fair comparisons between tools.
A wide selection of AI design tools are available to designers today, a variety of additional functionality within existing products, such as Framer AI and Figma Make, as well as standalone tools like Lovable and UIzard.
Due to the time I had available to write my dissertation, I would not be able to test every tool available with enough rigour to develop educated opinions. Therefore, I needed to select three tools that represented the spectrum of current tools.
A key issue initially was paywalls; while all the tools offered the ability to generate designs from a prompt, some require payment to go any further. UIzard, for example, created designs with screens that I could not access without paying a subscription; for this reason, UIzard was not part of my test group.
In the screenshot above, I have highlighted the parts of the generated design I would need to pay for to access and how even a simple app quickly exceeds the number of free components available.
A second issue became apparent while I was using Framer AI; it simply could not develop the designs I requested. It was clear that the Framer AI model was only able to tackle certain design tasks, so when my prompt was to generate a web app, it generated a landing page for the app rather than the app itself.
With these two tools unable to meet my needs, I was left with my final three for testing:
These all offered enough functionality for me to test for free and could be tested with automated accessibility testing software. This would allow fair and comparable testing to be conducted.
I now needed a methodology that would enable me to generate designs that could be tested consistently to achieve fair results.
To gain a breadth of experience with each tool, each was used to generate two applications, one using a single prompt and the other using an iterative, step-by-step approach, recommended by the tool’s creators.
I chose this approach because many people believe AI can do everything immediately. By giving it a simple prompt and analysing the results, I will be able to see how close the AI can get to achieving this.