The advent of AI-powered chatbots has sparked its use in teaching and learning languages. With respect to English as a foreign language, the impact of these tools has inspired research on its impact, particularly to enhance language skills. Text quality assessment has been boosted since chatbots like Chat GPT appeared a few years ago.

This paper brings into play DeepSeek, a similar AI tool, and aims to explore quantitative differences in the assessment of narrative texts written by thirty-eight (N=38) first-year pre-service teachers (PSTs) in a Spanish higher education context. Participants handwrote a narrative and self-assessed and peer assessed their texts, once they finished their marking, they used AI-driven evaluation using DeepSeek. Text quality was measured by using an analytic rubric, with a 1 to 10 scale, which assessed content, text organisation, vocabulary, grammar, and spelling.

A Kruskal-Wallis test was run, and significant differences were found in all features, except for Spelling. The Dwass-Steel-Critchlow-Fligner post-hoc test brought about significant differences only between PSTs peer assessment and DeepSeek’s in those aspects mentioned above.

The findings suggest that DeepSeek assessed PSTs’ texts in a similar way to humans. However, PSTs showed to have been more indulgent to their peers than to themselves as their scores for all aspects but Spelling were higher when compared to DeepSeek’s assessment.

Keywords: Writing assessment, DeepSeek, English as a foreign language, text quality, Pre-service teacher education.