DeepSeek’s Feedback on Short Essay-Writing in EFL: Pre-service Primary School Teachers’ Perceptions

Oct 15, 2026

Uploaded by Ophelia

Generative AI tools are increasingly used to support language learners’ writing, yet their role in pre-service teacher (PST) education remains underexplored, especially regarding DeepSeek. This study examines PSTs’ perceptions of DeepSeek’s feedback after a task-based teaching intervention. Thirty-nine Spanish primary education PSTs (A2–B2) wrote a short essay and provided data via pre/post mixed-methods, including open-ended questionnaires and focus groups. Results show a statistically significant decline in some items and reduced mean scores. Participants still reported improvements in grammar, vocabulary, and spelling, suggesting limited effectiveness for content and organisation for this task.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages11

Reading Time28 min

AccessFree to download

Abstract

Keywords

1. Introduction

2. Theoretical framework

The EuroCALL Review Volume 32, No. 2, 2025, 38-48 ISSN: 1695-2618 [https://doi.org/10.4995/eurocall.2025.23951](https://doi.org/10.4995/eurocall.2025.23951) RESEARCH PAPER DeepSeek’s Feedback on Short EssayWriting in EFL: Pre-service Primary School Teachers’ Perceptions Abraham Cerveró-Carrascosa *; Sofia Di Sarno-García ** * Universitat de València, Spain; ** Universidad Politécnica de Madrid, Spain *[abraham.cervero@uv.es](abraham.cervero@uv.es |)[ |](abraham.cervero@uv.es |) **[s.disarno@upm.es](s.disarno@upm.es) How to cite this article: Cerveró-Carrascosa, A., Di Sarno-García, S. (2025) . DeepSeek’s Feedback on Short Essay-Writing in EFL: Pre-service Primary School Teachers’ Perceptions. The EuroCALL Review, 32(2), 38-48. [https://doi.org/10.4995/eurocall.2025.23951](https://doi.org/10.4995/eurocall.2025.23951) Abstract Over the past few years, there has been a growing use of generative Artificial Intelligence (AI) tools to support the development of writing skills among language learners. However, limited attention has been given to the potential of AI in pre-service teacher (PST) education, and specifically, the tool DeepSeek has yet to be examined. This study, therefore, aims to investigate PSTs’ perceptions of the usefulness of DeepSeek's feedback on the production of a text following a task-based teaching intervention. Thirty-nine primary education pre-service teachers took part, they were first-year students from a Spanish university whose English proficiency ranged from A2 to B2 . The task required them to write a short essay. Data were collected using a pre-and post-test with openended questions adapted from Di Sarno-García and Argüelles-Álvarez (in press), along with a focus group discussion, thus adopting a mixed-methods approach that triangulates quantitative and qualitative data. The results of the Wilcoxon signed-rank test indicateda statistically significant decline in some items, as well as a slight decrease in mean scores from pre- to post-test which are aligned with the participants’ contributions from the open-ended questionnaire and focus group. These findings suggest that DeepSeek may not have been as effective as anticipated in providing feedback on content and text organisation for this particular task. Nevertheless, participants reported that they found the tool helpful for improving grammar, vocabulary, and spelling in their texts. Keywords Artificial Intelligence; DeepSeek; writing skills; perceptions; pre-service teacher education; English as a foreign language 1. Introduction Education, and language teaching in particular, has been transformed by the recent proliferation of Generative Artificial Intelligence (GenAI) (Di Sarno-García & ArgüellesÁlvarez, in press), especially after the launch of ChatGPT back in November 2022 by OpenAI, which is based on the generative pre-trained transformer (GPT) (Law, 2024) . According to Callanan (2024), AI has the capacity to independently select information from external sources to achieve a desired outcome, which has not only generated curiosity but also ethical concerns and challenges. Despite this, AI should not be banned from the classroom but rather integrated within teaching and learning practices. For this reason, authors such as Liu et al. (2024) encourage following a menu approach according to which only those AI tools that might be useful for the task should be employed, rather than using AI to do everything for us, in the same way that we would not order everything from a restaurant menu. Different AI chatbots have now been launched (e.g., ChatGPT by OpenAI and Copilot by Microsoft), the most recent being DeepSeek, which was launched in January 2025 by the Chinese company DeepSeek. The main differences between ChatGPT and DeepSeek are that, by the time this teaching intervention took place, the latter could answer an unlimited number of queries and is based on open coding, as opposed to the former. Its release r

FAQs

What is the main aim of the study on DeepSeek’s feedback?

To investigate pre-service primary school teachers’ perceptions of how useful DeepSeek’s feedback is for producing a text after a task-based teaching intervention.

How was data collected and analyzed?

Data were collected using a pre- and post-test with open-ended questions and a focus group discussion, adopting a mixed-methods approach. A Wilcoxon signed-rank test was used to assess changes.

What feedback aspects did participants find helpful even though results declined?

Participants reported finding DeepSeek helpful for improving grammar, vocabulary, and spelling in their texts, despite declines in some other measured items.

Oct 15, 2026

Uploaded by Ophelia