UX Testing with AI
An important step or just hot air?
Predictive attention models, trained on eye-tracking data, users simulated by large language models (LLMs), or agent-based interaction tests have been the subject of the most heavily discussed topics in ux groups since 2025. Providers such as Attention Insight and Neurons demonstrate that these new methods appear to work, citing accuracy rates of over 90%. As a result, these methods produce heat maps, clarity scores, click test results, or focus distributions. But how relevant are these insights? We wanna show you which of these methods are effective and where they fall short.
Especially when reviewing wireframes or planned click paths, saliency models and interaction tests can provide insights that would otherwise only have been obtained through time-consuming user tests. This not only saves effort but also leads to a more accessible result, since more iterations can be carried out. In this regard, innovative predictive UX tests clearly stand out for their speed and clarity.
While heatmaps clearly measure what catches the eye and what doesn’t, they fall short when it comes to evaluating user intentions. A real user never visits a website without a purpose. Rather, they have a certain level of prior knowledge, a goal, or an aversion to certain things. It is precisely this context that is completely missing from heatmap results. At the very least, heatmaps are hardly suitable for measuring user intent or for assessing whether users achieve their goals.
Another problem with for example modern fixation models, is that they assess in complete isolation whether a button is visually visible to the user. While this is certainly relevant, the model clearly lacks the comprehension component. It does not take into account whether users actually understand the button. It just measures pure perception.
The rates promised in advertisements and benchmarks are based on the results of laboratory studies conducted with small groups of users. These are not representative of older or users with disabilities, and thus fail to account for the very target audience that WCAG is intended to address.
FUF Insight
The clear strength
The predictive UX tests mentioned above are a valuable tool in the otherwise time-consuming process of accessible design. At FUF, we use these methods primarily for the iterative review of wireframes and early design concepts.
This allows us to efficiently evaluate visual hierarchies, interaction patterns, and potential weaknesses during the development and design phases without having to conduct extensive user testing for every iteration.
However, these methods do not fully replace accessibility walkthroughs, user tests, magnifier walkthroughs, or clickpath evaluations. Especially for a product’s final version, testing with real-world usage scenarios remains a crucial component of a comprehensive accessibility review.