Building Information Modelling (BIM) has been increasingly integrated with Artificial Intelligence (AI) solutions to automate building construction processes. However, the methods for effectively transforming data from BIM formats, such as Industry Foundation Classes (IFC), into formats suitable for AI applications still need to be explored. This paper conducts a Systematic Literature Review (SLR) following the PRISMA guidelines to analyse current data preparation approaches in BIM applications. The goal is to identify the most suitable methods for AI integration by reviewing current data preparation practices in BIM applications. The review included a total of 93 articles from SCOPUS and WoS. The results include eight common data types, two data management frameworks, and four primary data conversion methods. Further analysis identified three barriers: first, the IFC format’s lack of support for time-series data; second, limitations in extracting geometric information from BIM models; and third, the absence of established toolchains to convert IFC files into usable formats. Based on the evidence, the data readiness is at an intermediate level. This research may serve as a guideline for future studies to address the limitations in data preparation within BIM for AI integration.

Design errors carried over into construction lead to rework, material waste, and avoidable embodied carbon. Rule-based compliance checking in Building Information Modelling (BIM) can identify explicit design issues. However, current rule-based engines struggle to capture complex spatial arrangements and contextual geometric relationships among surrounding components. Artificial intelligence (AI) offers a promising alternative by learning tacit, geometry-dependent conventions of expert practice directly from BIM data. Yet existing AI approaches either learn from 2D representations or rely on handcrafted geometric features that presuppose the target rules. We conduct a controlled study on a parametric benchmark derived from a single apartment plan. A graph transformer predicts where a masked door belongs along a known host wall. It learns from BIM topology and pose-conditioned neural signed distance function (SDF) embeddings, without explicit coordinate features. On unseen rigid transforms, the model achieves 79.1% side accuracy. A graph attention baseline reaches 68.9%, and a message-passing baseline stays at chance. On a second plan with a different topology, it reaches 68.2%. Doubling the training data does not raise this. Without retraining, the first plan’s model reaches 74.0% on the second. Performance on an unseen room-semantic configuration falls to chance. This is a necessary step towards learning tacit design conventions.