My colleague Dr. Johanathan Woodworth and I developed a framework for AI integration that we call BEARA. The name comes from its five domains: Build, Evaluate, Apply, Reflect, and Adapt. We have used the framework with pre-service teachers in our university courses, and I am happy to share that one of the papers growing from this work has now been published.
The paper introduces BEARA Reading, an adaptation developed for AI-mediated reading. Reading deserves this attention because AI can now summarize a paper, explain a difficult passage, translate unfamiliar language, and offer an interpretation before the reader has formed one. The framework sits naturally within my wider work on critical AI literacy and asks a practical question: how can we use this support while keeping the reader responsible for meaning?
Summary of the Study
We analysed coursework produced by 37 K–12 in-service educators enrolled in two sections of a graduate educational technology course. Participants first completed an independent close reading of a scholarly text. They then used a generative AI tool and examined how its responses clarified, reinforced, changed, or distorted their understanding. Their submissions included documented AI exchanges and were organized through the five BEARA domains. Two research assistants independently coded the de-identified work after grading was complete.
Participants found AI useful for orientation, clarification, structural overview, and comparison. It could unpack difficult concepts or make dense prose more accessible. They also found omissions, oversimplification, misplaced emphasis, and interpretations that extended beyond the source.
The most difficult problems were not always obvious hallucinations. Some outputs remained close enough to the text to sound trustworthy while quietly flattening qualifications or presenting an inference as though the author had stated it. This is why reading AI critically requires more than checking whether an answer sounds reasonable.
The participants’ work helped us recalibrate the five domains around the act of reading. Build protects the reader’s initial encounter with the text. Evaluate requires AI claims to be checked against the source. Apply uses AI to extend analysis while human judgement remains central.
Reflect examines how the tool influenced trust, certainty, attention, and understanding. Adapt turns these observations into better routines for the next reading task. Several participants arrived at a sequence I find especially useful: close reading first, AI second. That sequence is a guiding principle, not an inflexible rule. A multilingual or struggling reader may need carefully chosen support earlier to gain meaningful access to the text.
The study does not claim that BEARA Reading improves comprehension or validates the framework independently. The assignment itself was organized through BEARA, and the analysis used those same domains. We can show that the structure supported meaningful reasoning about AI-mediated reading and required a more reading-specific form.
We cannot show that the domains would have emerged without prompting. The evidence also comes from educators’ written coursework in one graduate course, not direct observation of reading or assessment of K–12 students. The paper provides conceptual and practice-based support for BEARA Reading. Testing its effects across readers, subjects, text types, and educational settings comes next.
Practical Insights
BEARA Reading can be used as a sequence for classroom reading activities or as a planning tool for teachers.
- Build: Ask students to read, annotate, and record a provisional understanding before consulting AI. What is the main argument? Where did comprehension break down? What remains uncertain?
- Evaluate: Let students compare an AI explanation with the original text. They can mark each claim as supported, inferred, incomplete, or unsupported and attach textual evidence to their decision.
- Apply: Use AI to generate an alternative interpretation, identify assumptions, clarify a difficult idea, or raise questions for further reading. Students still decide whether the response is warranted.
- Reflect: Ask what made the AI response feel trustworthy. Was it the evidence, or its fluent language and confident tone? Did it resolve a genuine misunderstanding, confirm an existing belief, or introduce certainty the text did not provide?
- Adapt: Have students write a short reading protocol for their next assignment. They might delay AI until after annotation, write their own summary first, trace every AI claim to the source, or keep a record of where their interpretation and the AI response diverged.
The framework also clarifies assessment. Teachers need to decide what kind of reading they want to assess. Unaided comprehension, AI-supported comprehension, critical comparison, source verification, and regulation of a human-AI reading process are different achievements.
A polished summary cannot provide evidence for all of them. When AI is permitted, students can submit independent annotations, selected AI exchanges, claim checks, or a short explanation of how their understanding changed.
For researchers, the next stage requires direct measures of comprehension, source verification, transfer, and critical reading. Studies should include students across age groups and compare what readers can do with AI, without it, and after assistance is removed. Think-aloud protocols and reading logs would also reveal decisions that finished coursework cannot capture.
For Johanathan and me, this publication is an important step. It moves BEARA from a general model of pedagogical AI integration into the specific work of reading, interpretation, and critical literacy. The central principle remains the same across both versions: AI use should be guided by informed human judgement and revised in response to evidence.
Reference
Woodworth, J., & Kharbach, M. (2026). From BEARA to BEARA Reading: Adapting a pedagogical AI integration framework for AI-mediated reading. https://journals.calstate.edu/ijaipil/article/view/7728/5798
