How Does Your Brain Choose Between Words and Images in Ads?

Neuromarketing is the science that reads the brain while watching an advertisement. In a recent study, 22 volunteers wore an EEG cap, a device that records brain waves, while viewing verbal, visual, or mixed ads. Before the experiment, participants were classified as "verbalizers" or "visualizers" using the SOP questionnaire, which measures whether they prefer processing information with words or images.

During viewing, the EEG captured five frequency bands (Delta, Theta, Alpha, Beta, Gamma). Researchers then trained three algorithms – Support Vector Machine (SVM), decision tree, and k‑Nearest Neighbors – to recognize, from brain waves, which group each person belonged to. All models achieved accuracy between 86% and 93%, but SVM proved more stable when data were imbalanced.

A statistical analysis highlighted that the Theta band (4‑8 Hz) differed significantly between the two groups, suggesting this frequency is linked to how the brain processes words or images.

These findings open the door to more personalized advertising: in the future, content could adapt to your cognitive style, making the experience more engaging and less intrusive.

EEG analysis relies on decomposing the signal into frequency bands, each linked to specific cognitive states. The Theta band (4‑8 Hz) is traditionally associated with working memory and multimodal integration processes, explaining its relevance in distinguishing verbalizers from visualizers. In this study, raw data were filtered to remove ocular and muscular artifacts using ICA (Independent Component Analysis) algorithms, ensuring that analyzed components reflected genuine neural activity.

For classification, the SVM algorithm was configured with a radial basis function (RBF) kernel and optimized via grid search over C and gamma parameters, maximizing the margin between classes. The decision tree and k‑Nearest Neighbors, while showing comparable accuracy, suffered from overfitting when one class had fewer samples, highlighting sensitivity to imbalanced datasets.

Evaluation metrics included the area under the ROC curve (AUC) and sensitivity/specificity for each class. The SVM’s average AUC was 0.94, indicating excellent discriminative ability. Moreover, independent t‑tests revealed statistically significant differences (p < 0.01) in Theta power, but not in other bands, suggesting a potentially robust neurophysiological marker for processing style.

Study limitations include the small sample size (N=22) and lack of external validation on more heterogeneous populations. Future work could integrate deep‑learning techniques to capture more complex non‑linear patterns and combine EEG with eye‑tracking for a fuller mapping of visual attention and verbal processing. Practical application will also require ethical considerations regarding neural data privacy and transparency in using such information to personalize advertising content.