AI evidence extraction

Generative AI and Students’ Voice in Academic Writing

Harshita Vaghela, Sonali Mahadu Dharade, Nilima Bhalke, Ashok Ghuge · Journal of Humanities and Education Development · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

7.33

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.22161/jhed.8.2.10

Methodology & findings

Study design

Interpretive qualitative research design using purposive sampling of 20 undergraduate students.

Sample

N = 20, 3 groups

Primary method

Inductive thematic analysis; codes grouped into candidate themes such as 'Efficiency vs. Integrity' and 'Human-AI Partnership.' Analysis interpreted through Activity Theory lens examining AI as a mediating artifact. No quantitative statistical tests, p-values, or inferential statistics reported.

Main result

The study found that students perceive generative AI primarily as "an instrument of productivity and overcoming the previous paradigm of writing," with participants utilizing AI for cognitive load reduction, linguistic refinement, and time saving. However, a critical tension emerged: "The first dilemma for me is the choice between sounding better or sounding like me when integrating AI into the decision-making process," highlighting an authenticity paradox where AI-refined writing may override students' unique voices. The analysis revealed that "students who engage in a highly interactive, iterative process (rather than linear copy-pasting) achieve higher performance scores in their final drafts," suggesting that agency in the prompting loop is essential.

Reports effect sizes.

Research paradigm

Interpretivist/Qualitative

Author conclusions

The authors conclude that "generative artificial intelligence has revolutionized the academic writing process, making it not an individual thinking process but a collaborative interaction between the learners and a machine-learning process." They further state that "higher institutions of learning have to change their approaches related to academic integrity. Instead of focusing on detection software, which is often unreliable with prone errors, universities ought to work on developing critical AI literacy. This implies the establishment of official training programs that teach students to use AI tools as tactical editors and brainstorming partners and at the same time maintain independent control of their arguments."

Risk of bias

Selection bias: Purposive sampling of volunteers who already use AI may skew toward favorable attitudes; Researcher interpretation bias: Thematic coding from qualitative data is subjective and depends on coder reliability (not reported); Temporal bias: Single point-in-time data; no longitudinal tracking of writing development; Cultural/contextual bias: Sample composition and geographic context not fully specified; generalizability to non-English-speaking contexts unclear; Social desirability bias: Participants may underreport cheating concerns or overstate ethical awareness in interviews; Selection bias: Purposive sampling limits generalizability to self-selected participants willing to discuss AI use; Small sample size (n=20) may not represent diverse undergraduate populations; Potential social desirability bias in self-reported ethical concerns and AI usage patterns; Researcher interpretive bias in thematic coding (no inter-rater reliability reported); Lack of control group or comparison condition; Artifact elicitation may privilege articulate participants; Selection bias: Purposive sampling of volunteers who actively use AI in writing; Social desirability bias: Participants may underreport ethical concerns or overstate benefits in interviews; Limited diversity: Sample appears concentrated in non-native English speakers in higher education contexts; Researcher interpretation bias: Qualitative coding susceptible to analyst bias without explicit inter-rater reliability reported

Limitations

The paper states that "Future studies should look more closely at how students change their writing over time when they use chatbots, and also how different cultures or contexts affect these experiences." Additionally, the study acknowledges reliance on qualitative data from a limited sample without explicit discussion of generalizability constraints or saturation testing, and notes that "the quantitative dynamics of the AI adoption are fairly well-studied, whereas a more comprehensive and nuanced insight into the role played by the student perceptions in their real writing practices and results is still required."

Open questions raised

Limited longitudinal research on how students' writing evolves over time with sustained chatbot use
Insufficient investigation of how different cultures and contexts affect AI-assisted writing experiences
Need for more nuanced qualitative insight into subjective experiences and real writing practices beyond quantitative adoption statistics
Absence of comprehensive policy frameworks at universities regarding ethical limits of AI assistance
Lack of critical AI literacy training programs in higher education institutions
The authors identify that "quantitative dynamics of the AI adoption are fairly well-studied, whereas a more comprehensive and nuanced insight into the role played by the student perceptions in their real writing practices and results is still required." They also recommend future studies examine longitudinal changes in student writing and cultural/contextual differences in AI experience.

Extracted from: pdfAgreement 66%

Explore related topics

AI Manuscript Writing Academic Writing Qualitative