The paper aims at improving the efficiency of Artificial Intelligence implementation as part of modern computer-aided design systems. In order to optimize the process of human-computer interaction and stimulate engineering creativity, there is proposed a model for marking-up engineering data based on a concept of “affordance”. As part of the study, an original experiment was conducted, which consisted of determining the functional purpose of vintage and antique devices based on their images by technical specialists, representatives of related professions, and various generative models of Artificial Intelligence. Analysis of the guessing performance under the control of visual activity hit maps demonstrated the advantage of pre-labeling images based on the affordance model both for humans and artificial neural networks. Affordance-oriented image annotation enhances the efficiency of prompt engineering by reducing task completion time and improving the accuracy of functional interpretation of objects by human engineers and Large Language Models.