Disguised Ad
The Disguised Ad pattern is a visual manipulation technique where promotional content is engineered to mimic native interface elements, such as navigation menus, editorial content, or primary action buttons (e.g., a fake “Download” button). This pattern deliberately weaponizes Gestalt phenomena to blur the boundary between figure and ground, causing the advertisement to seamlessly blend into the host interface's contexture. By doing so, it exploits users' acquired “banner blindness”—the cognitive heuristic where users subconsciously filter out traditional ad formats—inducing erroneous interaction with the promotional material under the false premise of operating the native application.
1
Cross-Origin Action Masking
Condition 1: Cross-Origin Action Masking
Given
The most adversarial variant of this pattern occurs when an ad intercepts a user's functional intent. We define as a node styled as a primary action button, bearing an NLP-extracted text label that matches high-intent native tasks (e.g., {“Download”, “Start”, “Play”, “Next”}). We compare the primary domain of the web application, , against the destination domain resolved by inspecting the href attribute or intercepting the click event. The feature triggers if a visually prominent action button falsely masquerades as a native function but structurally routes the user to an external advertising domain:
2
Morphological Similarity
Condition 2: Morphological Similarity
Given
To quantify visual mimicry, we define as the set of visual feature vectors (encompassing computed background color, typography, border radius, and aspect ratio) extracted from legitimate, primary action nodes on the page via image analysis. We contrast this with , representing the feature vector of a structurally distinct third-party node, such as an <iframe> or a container injected by an external ad network. By calculating the cosine similarity between these vectors, the algorithm triggers if the ad element is styled to structurally and visually mimic native elements beyond a predefined similarity threshold :
3
Semantic Mimicry of Native Action Labels
Condition 3: Semantic Mimicry of Native Action Labels
Given
To establish a semantic baseline for Disguised Ad, the algorithm compares the text label of third-party advertising nodes against a corpus of native functional labels (e.g., “Download,” “Next,” “Play”). Using cosine similarity over LLM-generated semantic embeddings, the feature triggers if an ad's button label achieves a semantic similarity score exceeding with any native function label, indicating deliberate linguistic impersonation: