Interest tailoring was structurally present in AI-generated readings: logs averaged 52.24 tailoring markers per 10,000 words and showed moderate divergence (cosine similarity 0.50–0.61)¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Automated analysis of a 1/5-page sample of reading logs B–K found a mean of 52.24 tailoring markers per 10,000 words (SD = 13.73), ranging from 38.74 to 74.29. → Sidorkin 2026 q2 i? TF-IDF cosine similarity across logs ranged from 0.50 to 0.61 on average, indicating a shared instructional core with substantial sector- and role-specific divergence. → Sidorkin 2026 (2)

Evidence¶

Sidorkin 2026¶

Sidorkin, A. M. (2026). From One-Size Texts to Tailored Readings: Student Experiences with AI-Generated Course Materials. Open Praxis, 18(3), 506–522. https://doi.org/10.55982/openpraxis.18.3.1141

q2 · i? · design · r2

Artifact analysis using dictionary-based automated counts on a systematic 1/5-page sample of logs B–K (samples 13,728–22,034 words). The article reports "a mean of 52.24 markers per 10,000 words (SD = 13.73)", showing customization language was frequent but not uniform.

"the overall rate of tailoring markers was substantial, with a mean of 52.24 markers per 10,000 words (SD = 13.73), and a wide observed range from 38.74 (Log K) to 74.29 (Log B)"

Sidorkin 2026 (2)¶

Sidorkin, A. M. (2026). From One-Size Texts to Tailored Readings: Student Experiences with AI-Generated Course Materials. Open Praxis, 18(3), 506–522. https://doi.org/10.55982/openpraxis.18.3.1141

q2 · i? · design · r2

TF-IDF cosine similarity computed across sampled log text in Table 3, with robustness checks (stopword removal, bigram vectors) shifting averages by no more than 0.03. The article interprets this as "a common instructional backbone across logs, with meaningful contextual overlays".

"Average cosine similarity across logs ranges from 0.50 to 0.61"

Discussion¶

Related Claims¶

• Comprehension-level tailoring occurred: AI responses following comprehension prompts showed 3.4x to 8.7x higher definitional density than non-triggered responses — related
• LLM-simulated student ideas moderately overlap with teacher- and student-generated lesson ideas (mean cosine similarity 0.52), with larger LLMs slightly outperforming Mistral-7B — related
• Students showed bounded trust and uneven verification behavior: 23 of 837 student turns (about 2.7%) contained explicit risk-aware moves — related
• AI-generated readings lacked internal traceability: only about 0.80% of pages contained APA-style citations, URLs appeared on about 0.53%, and DOI strings were essentially absent
• Educators broadly view AI-generated text customization to student interests and reading levels as a desirable opportunity, particularly for ELs — a broader claim this one bears on