The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2773
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T17:21:53.619468Z
status: published
lens: g_space
sector: education
headline: Impact of Artificial intelligence tools on learning motivation in English instruction: a network meta-analysis
dek: Abstract The rapid expansion of English-Medium Instruction (EMI) programs and the growing demand for English language proficiency have introduced significant motivational challenges for EFL learners who face linguistic barriers. Artificial intelligence (AI) tools offer a potential solution by providing personalized scaffolding and interactive support. This study conducts a network meta-analysis (NMA) of 16 empirical studies (N = 1,923) to compare the effectiveness of three AI interventions, including Generative…
gain_title: In a network meta-analysis of 16 controlled studies with 1,923 English-learning students, AI language apps (g = 0.907), chatbots (g = 0.824) and writing assistants (g = 0.692) each raised learning motivation more than traditional instruction.
problem_title: (none)
trace_subject: (none)
gain_reading: In a network meta-analysis of 16 controlled studies with 1,923 English-learning students, AI language apps (g = 0.907), chatbots (g = 0.824) and writing assistants (g = 0.692) each raised learning motivation more than traditional instruction.
gain_evidence: All three AI intervention categories demonstrated statistically significant positive effects on learning motivation compared to traditional instruction
problem_reading: (none)
problem_evidence: (none)
quick_read: A network meta-analysis by Liwei Hsu and Yu-Chun Wang pooled 16 experimental and quasi-experimental studies covering 1,923 K-12 and university students learning English as a foreign language. Every AI category beat traditional instruction on learning motivation: language apps (g = 0.907, 2 studies), generative AI chatbots (g = 0.824, 11 studies) and AI writing assistants (g = 0.692, 3 studies).

The effects are medium to large and exceed the 0.40 benchmark the authors cite for meaningful educational effects. Sensitivity analyses that dropped three higher-risk studies changed the estimates by less than 0.06, and Egger's test found no sign of publication bias.

The ranking of the AI tools against each other is much less certain. No study compared AI categories directly, so those comparisons rest on indirect evidence, and none of the pairwise differences reached statistical significance. The authors call the rankings preliminary.
limitation: All 16 studies compared AI with traditional instruction and none compared AI tools with each other, so rankings among AI types rest on indirect evidence. The median intervention lasted eight weeks and effects may be inflated by novelty and the Hawthorne effect. Nine studies were from China.
tag: Evidence-backed gain
key_points: The analysis covered 16 studies published 2021 to 2025 and 1,923 students; sample sizes ranged from 50 to 412, with a median of 85. | Eleven studies tested chatbots (N = 1,268), two tested language apps (N = 443) and three tested writing assistants (N = 212). | Nine of the 16 studies were from China; the rest came from Iran (3) and one each from the UAE, Algeria, Nigeria and the United States. | Interventions lasted 6 weeks to one semester, with a median of 8 weeks; 14 studies used traditional instruction as the control. | Heterogeneity was moderate (I2 = 42.3%), and effects were slightly smaller for studies longer than 8 weeks (g = 0.78) than for those of 8 weeks or less (g = 0.86). | University studies (g = 0.81) and K-12 studies (g = 0.89) did not differ significantly, and East Asian and Middle Eastern or African studies gave similar results (g = 0.83 vs 0.82). | The authors propose a three-level framework that pairs apps, writing assistants and chatbots with different support needs, and say AI should complement teacher guidance.
rundown: The authors searched Web of Science, Scopus, ERIC, PsycINFO and Google Scholar for peer-reviewed studies from January 2015 to December 2025. Of 2,156 records, 138 full texts were assessed and 16 met the criteria, which required a comparison group and quantitative motivation measures. Two reviewers screened independently with high agreement (kappa 0.87 and 0.92).

Study quality was rated with Cochrane's RoB 2 for randomized trials and ROBINS-I for quasi-experimental studies. Blinding was generally not feasible, leaving most studies at moderate risk of bias in performance-related domains. A frequentist random-effects model estimated Hedges' g for each AI category against traditional instruction.

The authors suggest chatbots work partly by reducing speaking anxiety and apps by gamified progress markers that support competence, but these mechanisms are interpretation drawn from other research rather than tested here. They also warn that over-reliance on AI feedback could weaken independent learning, and recommend fading AI support as proficiency grows.

Their recommendations include assessing students' proficiency before choosing a tool and using chatbot practice before oral presentations. They also call for head-to-head randomized trials and multi-semester studies.
sources:
- peer_reviewed | Asian-Pacific Journal of Second and Foreign Language Education | https://doi.org/10.1186/s40862-026-00421-9 | 2026-07-16
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

46194a07bb62d7ed9ae5d0d0ba80249ebbba442f5960a474f751d63fa6afaa67

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000