We are witnessing a rise in the use of AI for architectural visualisations, reflective of a global shift with implications for visual representation and design pedagogy. This change is prompting questions about how discernible and how effective AI-generated images can be in portraying architectural spaces convincingly. In this paper, we present the initial findings of a wider ongoing study at the University of Salford, investigating perceptions of AI-generated images. We focus on the ability of 56 participants to discern between AI-generated architectural images and human-created CGIs. To our knowledge, this is the first study of its kind to systematically evaluate human perceptions in the use of AI-generated imagery in architecture. A mix of students and construction professionals with varying levels of experience and from across 2 continents participated in two sets of tests: Task 1, where the images presented individually and Task 2, where participants needed to identify the image in a pair of two. We then asked participants if the AI images could meet the criteria of the Lovelace test, a benchmark for assessing whether an artificial agent can produce creative output on a par with a human. Findings indicate significant differences in perception relative to experience levels, with less experience surprisingly yielding more accuracy. We also find that while generated images are received as highly creative, they are perceived as of less value and originality. The results suggest that the experience with AI does not equal better accuracy. We conclude that AI visualisations are at a threshold in producing work that is considered creative, but perception of these images is influenced by doubts about meaning, value and authorship.