From Output to Inquiry: The Limits and Possibilities of AI-Assisted Primary Source Analysis in Readings in Philippine History

John Edward Alfonso

National University, NU Clark, Philippines.

How to cite: Alfonso J. E. (2026). From Output to Inquiry: The Limits and Possibilities of AI-Assisted Primary Source Analysis in Readings in Philippine History. In: 12th International Conference on Higher Education Advances (HEAd’26). Valencia, 15-18 June 2026. https://doi.org/10.4995/HEAd26.2026.21471

Abstract

This paper examines how ChatGPT can assist in the teaching and learning of Readings in Philippine History (RPH), a Tertiary-Level General Education course in the Philippines that emphasizes the context and content analysis of primary sources. Using a qualitative pedagogical case study, the researcher provided ChatGPT with Antonio Pigafetta’s account of Magellan’s voyage and an RPH-style analysis guide, then evaluated the AI-generated output against the expectations of source analysis. Results show that ChatGPT can quickly produce organized, template-aligned responses; however, its analysis often remains broadly contextualized, relying on summary rather than text-grounded justification, and may include omissions or inaccuracies when traceability is required. The study concludes that AI is most valuable as a scaffold for drafting and organizing ideas, while meaningful historical analysis still relies on human verification, evidence tracing, and synthesis, supported by disclosure norms and authentic assessment.

Keywords: AI-assisted Learning; Readings in Philippine History; Historical Source Analysis; ChatGPT; Historical Thinking; Context and Content Analysis.

1. Introduction

Generative artificial intelligence (AI) tools, particularly large language models (LLMs) like ChatGPT, have rapidly become an integral part of everyday academic life. Within higher education, instructors and students now encounter AI not as a distant “emerging technology,” but as an accessible system that can draft, summarize, reorganize, and explain information on demand. In history education, this shift has elicited mixed reactions, with optimism about efficiency and new learning opportunities alongside concerns about academic integrity, bias, and the potential weakening of core critical thinking skills. As Gossard (2024) observes, history educators increasingly find themselves at a crossroads because generative AI can support learning but also raises serious questions about historical inquiry and student learning.

These concerns are particularly relevant in the Philippine tertiary-level General Education course Readings in Philippine History (RPH), which is designed to train students in disciplined interpretation of the past through primary sources rather than reliance on textbooks (Commission on Higher Education [CHED], 2013). RPH exposes learners to written, oral, visual, audio-visual, and digital sources and expects them to analyze these materials through contextual and content analysis. Contextual analysis examines the circumstances of production (time and place), the author’s background and intent, and the source’s relevance to the present; content analysis focuses on identifying arguments and claims, comparing viewpoints, detecting bias, and evaluating claims based on evidence (CHED, 2013). In other words, RPH does not merely require students to state an interpretation; it requires them to justify interpretations through sourcing and evidence-based reasoning.

However, the growing accessibility of generative AI complicates this goal. A developing body of research suggests that AI can produce outputs that appear highly competent. Sometimes, it even outperforms human writing in perceived quality. Tirado-Olivares et al. (2023), for instance, found that preservice teachers rated a ChatGPT-generated historical argumentative text higher than a peer-produced text on most historical thinking dimensions. Such findings highlight a classroom risk: AI outputs can look “complete” and persuasive even when they remain shallow, incomplete, or inaccurate. This makes source-based courses like RPH especially vulnerable to shortcut learning unless AI use is intentionally guided and assessed.

This paper addresses this issue through a qualitative pedagogical case study examining an AI-assisted evaluation and analysis of Antonio Pigafetta’s account of Magellan’s voyage using an RPH-style analysis guide. The study examines the accuracy, depth, and educational value of AI-generated insights and identifies where the AI output becomes overly general, insufficiently grounded in the source, or inaccurate, particularly when traceability to specific parts of the text is required. Consistent with history pedagogy, which treats AI output as material for critique rather than as an authority to be accepted (Washburn & McCutchen, 2024), the case also informs practical strategies for responsible AI-assisted learning.

Guided by this framing, the study addresses three questions: (1) What kinds of contextual and content analyses does ChatGPT generate when asked to analyze Pigafetta using an RPH-aligned guide? (2) To what extent are these outputs accurate and evidence-traceable in relation to the primary source? and (3) What responsibilities and classroom practices are necessary for instructors and students so that AI supports learning rather than replaces it?

2. Methodology

2.1. Design, Materials, and Data

This study employs a qualitative pedagogical case study design to examine how ChatGPT performs in a core RPH task: contextual and content analysis of a primary source. It presents an exploratory pedagogical case study that illustrates how a generative AI tool behaves when completing an RPH-aligned source analysis task. It does not measure student learning outcomes and is not intended to be generalizable. The AI-assisted analysis was conducted using ChatGPT 4.0 on the ChatGPT platform at the time of the interaction. The case centers on Antonio Pigafetta’s account of Magellan’s voyage, paired with an RPH-style source evaluation and analysis template aligned with CHED’s framing of RPH source work (CHED, 2013). The primary data for analysis consist of (1) the Pigafetta excerpt used in the activity, (2) the guide questions/template provided to ChatGPT, and (3) the full transcript of the researcher-ChatGPT interaction (captured via screenshots and exported transcript). The interaction transcript and prompts may be made available upon reasonable request. The findings are presented as illustrative evidence of strengths and limitations in AI-generated source analysis and as a basis for practical instructional recommendations.

2.2. Procedure and Basis of Evaluation

The researcher provided ChatGPT with the source text and the RPH-aligned guide, then prompted the system to generate an evaluation and analysis. The interaction was documented to preserve exact prompts and responses. The researcher then used iterative follow-up prompts to test and refine the output, including: requesting a more detailed historical background; redirecting the analysis toward what the excerpt reveals about the Philippines; requiring traceability by requesting paragraph-level references; and correcting inaccuracies when they appeared. As a validation step, claims were tested by requiring paragraph-level references, then checking whether the cited locations matched the source excerpt; discrepancies were documented in the transcript and prompted correction. The analysis in this paper is based on these observable adjustments and improvements documented in the transcript – specifically, whether the AI output (a) moved from generic context to more specific contextualization, (b) shifted from summary claims to text-grounded justification, and (c) produced errors (e.g., incorrect paragraph references) that required human verification and correction. This approach is consistent with scholarship emphasizing that AI outputs become educationally valuable when they are treated as draft material, subject to critique and verification, rather than as final answers (Gossard, 2024; Washburn & McCutchen, 2024).

2.3. Analytical Framework

AI outputs were analyzed using an RPH-aligned framework drawn from CHED’s emphasis on contextual and content analysis. Responses were categorized according to: (1) contextualization specificity (generic vs. source- and period-specific), (2) evidence grounding (summary-level vs. traceable claims anchored in the text), and (3) accuracy/traceability (presence of omissions or incorrect references, particularly when paragraph-level support was required). These categories were applied across the transcript as the researcher prompted revisions, enabling a transparent account of what improved, what remained generic, and what required human correction.

3. Case Presentation and Results

In the Pigafetta case, ChatGPT was first provided with the primary source excerpt and an RPH-style guide, and it produced an immediately organized, template-aligned evaluation and analysis, but the initial contextualization remained broad, and the supporting points were largely summary-based rather than clearly grounded in textual evidence.

The researcher then prompted improvements by supplying a specific reference to enrich the historical background and by redirecting the analysis toward what the excerpt reveals about the Philippines, which led to a more detailed context and a clearer interpretive focus organized into thematic claims.

However, when required to demonstrate traceability by providing paragraph numbers for each supporting detail, the AI initially produced inaccurate references that had to be corrected through human intervention, underscoring that even with the text and guide provided, acceptable RPH-aligned analysis still depends on evidence-checking, verification, and the learner’s critical judgment.

4. Discussion

The Pigafetta case illustrates a central tension already flagged in the literature: generative AI can produce outputs that look academically competent, yet still fall short of disciplinary standards for historical thinking. In the transcript, ChatGPT immediately generated a structured RPH-style response (historical background, author/purpose, main idea, supporting details, and credibility). This “template-fitting” strength helps explain why AI-generated writing can be evaluated favorably by readers: Tirado-Olivares et al. (2023) found that preservice teachers rated a ChatGPT-generated historical argumentative text higher than a peer-generated text across most historical-thinking dimensions, suggesting that coherence and structure can strongly shape perceived quality. In the Pigafetta exchange, the same pattern appears: the response was organized and plausible, but its strongest value was as a first-pass scaffold rather than a final analytical output (cf. Gossard, 2024).

At the same time, the transcript shows that AI-generated “contextual analysis” may remain broad and generic unless the user provides more specific constraints or additional references. The initial historical background focused on general themes (European maritime expansion, spice trade competition) and became more detailed only after the researcher directed the system to a specific external reference and requested refinements. This supports Ma’s (2024) “chance or challenge” framing: ChatGPT can assist with drafting and explanation, but it can also produce confident narratives that are not sufficiently situated unless users actively guide, verify, and delimit what counts as acceptable historical context. The implication for RPH is straightforward: a tool that generates polished prose quickly can encourage premature closure, even when contextualization is incomplete, precisely the habit RPH aims to correct through contextual analysis (CHED, 2013).

The most pedagogically significant limitation emerged when the task required traceability to the primary source. When the researcher asked ChatGPT to identify paragraph-level support for claims, the AI initially provided paragraph references that were later flagged as inaccurate, then “corrected” after human intervention. This moment matters because it demonstrates a practical weakness relevant to CHED’s (2013) RPH goals: RPH is not only about stating an interpretation; it is about linking interpretation to evidence through disciplined reading, sourcing, and evaluation of claims. In other words, AI can generate an interpretation rapidly, but it may fail when asked to show its work in a way that is verifiable in the text. This aligns with Philippine-oriented cautions that generative AI use can encourage shallow engagement or “intellectual laziness” if students rely on outputs without verifying accuracy (Gagaza, 2025). It also reinforces the discipline-specific recommendation that a productive way to use AI in history education is to treat AI output as something to be critiqued and checked rather than accepted (Gossard, 2024).

This is where the literature offers a clear instructional pathway that matches the observed pattern in the transcript: use AI as an object of critique to strengthen historical thinking. Washburn and McCutchen (2024), working in the context of teaching American Indian history, emphasize activities where students evaluate AI-generated summaries/reviews and identify what is missing, weak, or incorrect. The Pigafetta case supports this approach because the AI output improved only through iterative prompting and correction: acceptable analysis depended on human judgment, close reading, and revision. In short, the case demonstrates how AI can “supplement rather than subvert” learning (Washburn & McCutchen, 2024) when the task design forces students to (a) verify, (b) revise, and (c) justify with evidence.

These findings also point toward assessment and ethics measures that prevent AI from replacing effort. Recent scholarship in higher education argues that educators must adjust assessment formats because conventional outputs can be convincingly produced by generative AI (Hellany et al., 2025). For RPH, the transcript suggests one concrete safeguard: require evidence traceability (quotations, paragraph/line references, and a brief justification of why the cited portion supports the claim). This single requirement exposes inaccuracies quickly and shifts attention back to the source. In parallel, the ethical guidance literature stresses transparency and declared use: rather than treating AI use as automatically illegitimate, instructors should require disclosure and set clear boundaries on what counts as acceptable assistance (Cao, 2025; Morato et al., 2025). Such principles align with the practical reality shown by the case: AI can be useful, but only when students remain responsible for checking, interpreting, and synthesizing.

Taken together, the Pigafetta case supports a concise pedagogical position with direct relevance to RPH: AI can accelerate organization and drafting, but it does not reliably produce RPH-level historical analysis without human verification and source-grounded justification. This is consistent with the broader teaching stance that technology improves learning only when aligned with content goals and pedagogy, rather than used as a drop-in shortcut (Jimenez & Albela, 2024). For RPH specifically, the most defensible classroom approach is therefore not “AI or no AI,” but AI-with-requirements: students read and draft first, use AI to refine or explore overlooked angles, and submit work that is demonstrably anchored in the primary source, with transparent disclosure and opportunities for classroom processing and authentic assessment.

5. Pedagogical Implications

5.1. For students: AI use must be evidence-based, not answer-based

The findings suggest students should be taught a simple rule for RPH tasks: read first, think first, then consult AI. AI can help organize ideas and surface possible angles, but students must still do the core work of historical thinking: close reading, identifying bias, and grounding claims in the source (CHED, 2013). This aligns with the broader view that AI can assist learning but becomes harmful when it replaces effort and critical engagement (Gagaza, 2025).

Practical student protocol:

• Read and annotate the source first then draft initial answers

• Use AI to refine, not replace: ask for clearer structure, alternative interpretations.

• Verify every historical claim and every “specific detail” back against the text.

• Synthesize: final output should reflect the student’s idea, not AI’s voice.

5.2. For instructors: design tasks where AI cannot “finish the work” alone

The case indicates that generic prompts allow AI to generate plausible but thin analyses. To protect RPH learning goals, instructors should design activities that require traceability and justification, and treat AI output as something students must critique.

Two strategies strongly supported by the literature are:

• Critique-based activities: students evaluate AI-generated responses, identify inaccuracies, and revise it using the source (Gossard, 2024; Washburn & McCutchen, 2024).

• Disclosure and ethical use norms: require students to declare AI use and treat AI as a tool with limits (Cao, 2025; Morato et al., 2025).

5.3. Assessment redesign: shift from product-only to process + authenticity

Because AI can produce polished outputs quickly, traditional take-home templates can be vulnerable to “finished-looking” but unauthentic work. Recent discussions in higher education emphasize that assessment in the era of generative AI must evolve toward formats that reveal thinking and learning (Hellany et al., 2025). For RPH, this can include:

• In-class source analysis workshops (students build answers with the text present).

• Oral defense (“Show me where in the text you got this claim.”)

• Process-based submissions: annotated source + draft + AI prompt log + revision notes.

These approaches keep AI in a supportive role while making the student’s reasoning visible, consistent with the position that effective technology use depends on pedagogy-content alignment rather than tools alone (Jimenez & Albela, 2024).

6. Conclusion

This study examined the usefulness of ChatGPT for RPH through a focused case: an AI-assisted evaluation and analysis of Antonio Pigafetta’s account of Magellan’s voyage. The findings show that ChatGPT can quickly produce an organized response that resembles a completed source analysis, which may appear credible and “submission-ready” at first glance. However, a closer review reveals that the output often remains broad in its contextual framing, relies on summary rather than text-grounded justification, and may contain omissions and inaccuracies, especially when the task requires traceability to specific parts of the primary source. In short, AI can support drafting and organization, but it does not reliably meet RPH’s core standards of contextual and content analysis without human verification, correction, and synthesis.

The implications are practical for teaching. Rather than banning AI or allowing unrestricted use, RPH instruction can adopt an “AI-with-requirements” approach: students do an initial reading and draft first, use AI to refine or explore alternative angles, and then verify all claims against the text through evidence-traceable citations. This can be strengthened by clear disclosure rules and more authentic assessment formats that make student thinking visible (e.g., in-class source workshops, oral defense, annotated drafts). Such measures keep AI in a supportive role while protecting the course’s central aim – forming students who can evaluate sources, recognize bias, and construct evidence-based interpretations.

This study analyzes a single tool-task instance using one historical text and a researcher-led interaction; it therefore describes AI behavior in a specific analytic task rather than student learning effects. Findings should be interpreted as exploratory and illustrative, with future work needed to include classroom implementation, student outputs, and multiple sources for comparison. Ultimately, the educational value of AI in RPH depends less on the tool’s fluency and more on how teachers and students use it. AI can organize information, but only learners can turn it into wisdom. Our role is to teach students to use it as a partner to their curiosity, not a replacement for their effort.

References

Cao, B. (2025). Don’t be sorry, just declare it: Pedagogical principles for the ethical use of ChatGPT. In Proceedings of the International Conference on Higher Education Advances (HEAd’25). https://doi.org/10.4995/HEAd25.2025.20092

Commission on Higher Education. (2013). CHED Memorandum Order No. 20, series of 2013: General education curriculum; holistic understandings, intellectual and civic competencies. Commission on Higher Education.

Gagaza, M. A. (2025). Reimagining Philippine education in the era of Education 5.0: Empowering teachers through generative artificial intelligence and design thinking. The Normal Lights, 19(2). https://doi.org/10.56278/tnl.v19i2.3386

Gossard, J. M. (2024). Teaching history in the time of generative AI. Teaching History, 49(1). https://doi.org/10.33043/5zbfza2bc

Hellany, A., Heinitz, A., McAlpine, I., Yevenes, K., Stevens, M., & Shrestha, S. (2025). Assessments—Search for an answer in the era of generative AI. In Proceedings of the International Conference on Higher Education Advances (HEAd’25). https://doi.org/10.4995/HEAd25.2025.19985

Jimenez, J. V., & Albela, E. J. A. (2024, June). Teaching history with technology and its challenges in the context of the current system of Philippine education. In 17th DLSU Arts Congress Proceedings (Vol. 8, No. 1, pp. 97–108). De La Salle University. https://16thdlsuartscongress.com/17th-dlsu-arts-congress-proceedings-volume-8/

Ma, G. (2024). Chance or challenge: The role of ChatGPT in history teaching and historical research in higher education. In G. Guan et al. (Eds.), Proceedings of the 2023 3rd International Conference on Education, Information Management and Service Science (EIMSS 2023) (Atlantis Highlights in Computer Sciences, 16, 870–874). Atlantis Press. https://doi.org/10.2991/978-94-6463-264-4_101

Morato, R., Peral, M., & Ferrández, A. (2025). Ethical use of generative artificial intelligence in higher education: Establishing principles for students. In Proceedings of the International Conference on Higher Education Advances (HEAd’25). https://doi.org/10.4995/HEAd25.2025.20046

Tirado-Olivares, S., Navío-Inglés, M., O’Connor-Jiménez, P., & Cózar-Gutiérrez, R. (2023). From human to machine: Investigating the effectiveness of the conversational AI ChatGPT in historical thinking. Education Sciences, 13(8), 803. https://doi.org/10.3390/educsci13080803

Washburn, J., & McCutchen, J. M. (2024). AI meets AI: ChatGPT as a pedagogical tool to teach American Indian history. Critical Humanities, 2(2), Article 2. https://doi.org/10.33470/2836-3140.1037