Restaurant menus rarely provide nutrition information, making it difficult for consumers to identify options that may be high in saturated fat or salt before ordering. Although packaged foods commonly display nutrition labels, restaurant menus typically contain only dish names and brief descriptions, despite out-of-home meals contributing substantially to dietary fat, sodium and overall energy intake (1). Existing methods for estimating nutritional quality include ingredient-based inference and image-based estimation. However, ingredient-based methods generally require structured ingredient statements (2), while image-based methods depend on food images that are often unavailable before purchase (3). These limitations highlight the need for methods that can infer nutritional characteristics directly from menu text. This study introduces SmartMenu, a keyword-based nutritional inference method for translating unstructured menu language into interpretable high-saturated-fat and high-salt cues. SmartMenu treats menu descriptions as implicit nutritional signals and applies a deterministic two-stage inference process. First, dish-level analysis evaluates dish names and cooking methods for explicit cues. When no cue is identified, keyword-level analysis examines ingredient-related keywords, estimated keyword prominence, and cooking methods to infer nutritional risk. A keyword database was developed using the United States Department of Agriculture FoodData Central database as the reference source. Keywords were mapped to high-saturated-fat and high-salt cues using the United Kingdom front-of-pack traffic light nutrition labelling criteria. Face validity of the generated cues was assessed by nutrition experts. Overall, 73.9% of high-saturated-fat cues and 76.5% of high-salt cues were rated as nutritionally valid. Among invalid classifications, overestimation was the predominant error, accounting for 18.2% of generated high-saturated-fat cues and 17.4% of generated high-salt cues. This pattern suggests that SmartMenu more frequently adopted a conservative approach by flagging potential unhealthy menu attributes rather than failing to identify them. Expert feedback indicated that uncertainty surrounding ingredient quantities was the main factor affecting cue validity, reflecting the limited detail typically provided in restaurant menus. These findings demonstrate that keyword-based analysis can generate plausible nutritional cues from sparse menu descriptions, although performance depends on the specificity of menu language and the presence of nutritionally informative keywords. SmartMenu provides a lightweight, transparent, and interpretable alternative to approaches that require detailed ingredient data, curated dish repositories, food images, or labelled menu-to-nutrition datasets. By operating directly on menu text, it offers a practical foundation for real-time nutritional guidance and population-level monitoring of out-of-home food environments.