In manufacturing, verifying a product’s compliance with specifications is a fundamental step to ensure quality and reliability. In assembly, verification is not limited to the finished product, but includes checks distributed throughout the entire process. When inspection is predominantly visual, variability may arise due to human factors and observation conditions; therefore, increasing objectivity, repeatability, and traceability is a relevant goal. This work evaluates the effectiveness of a multimodal LLM in supporting such verification by comparing a CAD-derived geometry reference (STL mesh) with a photograph. The approach aims to reduce the reliance of the judgement on subjective interpretations, making the outcome justified, structured, and verifiable.