Synthetic Spanish Text Dataset for Research on LLM-Generated Content

This dataset comprises synthetic news descriptions, entirely focused on Spanish-language content. Each description was generated from a real newspaper headline through a controlled Large Language Model (LLM) pipeline employing two distinct configurations: Retrieval-Augmented Generation (RAG),and generation without contextual retrieval (NO-RAG). For both configurations, three different temperature values were used during generation to control variability in the outputs. Headlines were collected over a two-week period from two prominent Spanish newspapers, referred to in this dataset as Newspaper A and Newspaper B. Each description generated in RAG mode uses a context derivated from a one-month history of past news descriptions from the same newspaper. These past news are embedded and stored in a vector database to enable semantic retrieval.

The dataset contains 18,236 synthetic descriptions in total (5,716 from A and 12,520 from B). Regarding configuration, 9,120 descriptions were generated by NO-RAG, and 9,116 using RAG.

To provide a clear overview of the dataset, the key details are summarized in the table below:

Headlines collection period

26 October 2025 – 11 November 2025

RAG knowledge base period

25 September 2025 – 25 October 2025

Language

Spanish

LLM model

mistral:7b-instruct

Temperature values

1, 0.75 and 0.5

The dataset is delivered as a collection of JSON files organized in a hierarchical folder structure for clarity and ease of use. The folder structure is as follows:

dataset/

  + 01. llm_news_NO_RAG/

      - llm_news_NO_RAG.A.json

      - llm_news_NO_RAG.B.json

  + 02. llm_news_RAG/

      - llm_news_RAG.A.json

      - llm_news_RAG.B.json

   The subfolders distinguish between the generation configurations:

llm_news_NO_RAG contains descriptions generated without contextual retrieval.

llm_news_RAG contains descriptions generated using retrieval augmented generation.

Each JSON file corresponds to a specific newspaper (A or B) and includes all generated descriptions for that source.