For the experiment, I analyzed Frankenstein; Or, The Modern Prometheus by Mary Wollstonecraft Shelley using two different digital tools: Voyant Tools and Google Gemini 2.5 Pro. I downloaded the novel from Project Gutenberg and removed the boilerplate text so that only the story itself remained.
Using Voyant Tools, I began with the Cirrus word cloud, which highlighted the most frequent words—man, life, father, shall, and eyes—after excluding stop words. The Reader and Trends tools traced how certain terms appeared throughout the text, while Links tool visualized relationships between commonly co-occurring words. These features provided an immediate visual overview of the novel’s vocabulary and structure. However, most of Voyant’s tools emphasize frequency and proximity rather than meaning. Although the patterns were intriguing, the analysis largely remained at surface level, reducing the novel to lists of recurring words rather than deeper interpretations.
Gemini produced a distinctly different type of analysis. Following the prompts from Anastasia Salter’s instructions, it generated a list of significant terms—father, life, death, creature, and spirit. Then, it identified Frankenstein as a hybrid of Gothic, Romantic, and early Science Fiction genres. Gemini also mapped out the social networks of the story. It described Victor and his Creature as two central figures connected primarily through violence, and it highlighted recurring phrases such as ‘poor creature’ and ‘my dear victor’, reflecting the novel’s emotional tone and its recurring themes. However, it was unable to produce visualizations such as word clouds or networks when prompted.
Gemini’s output resembled a human-written essay, interpreting rather than merely describing. Because Frankenstein is a widely studied novel and Gemini is a trained AI model, its response was likely shaped by exposure to existing analyses in its training data. This raises important questions about authorship and bias in AI-assisted interpretation: how much of the analysis arises from user input, and how much from prior data embedded in the model? Such influences suggest that different AI tools may produce divergent readings of the same text depending on their design and dataset.
Overall, Voyant and Gemini demonstrate both the potential and the limitations of computational reading. Voyant provides quantifiable insights into linguistic patterns, while Gemini provides interpretive connections informed by prior knowledge. Yet both reveal that data-driven tools cannot fully capture the complexity of literature. In the Age of AI, digital analysis expands how we read, but meaningful interpretation still depends on human reflection and awareness of the biases embedded in the tools we use.
