ABSTRACT We propose a demand estimation approach that leverages unstructured data to infer substitution patterns. Using pre‐trained deep learning models, we extract embeddings from product images and textual descriptions and incorporate them into a mixed logit demand model. This approach enables demand estimation even when researchers lack data on product attributes or when consumers value hard‐to‐quantify attributes such as visual design. Using a choice experiment, we show this approach substantially outperforms standard attribute‐based models at counterfactual predictions of second choices. We also apply it to 40 product categories offered on Amazon.com and consistently find that unstructured data are informative about substitution patterns.