Prompt Engineering

Prompt engineering — the practice of designing and refining inputs to large language models to achieve desired outputs. In education, prompt engineering serves dual roles: as a learner skill (students must learn to prompt effectively) and as a system design lever (developers craft prompts that shape AI tutoring behavior).

Questions to Consider

• You've likely typed a prompt into an AI tool recently. Now consider this: the way you phrased it isn't neutral — it may reveal how you planned, thought, and allocated your effort. What might your own prompting habits say about how you approach problems?
• A study found that users who phrase requests skillfully systematically get better output than those expressing the same intent less adroitly. If you accept that 'prompt privilege' is real, is fair access to AI best fixed by teaching everyone to prompt better, or by redesigning the system to not demand that skill — and what are the trade-offs of each?
• Is prompting a 'trick' to be memorized, or a genuine intellectual skill? One line of research treats it as professional judgment within a discipline (journalism, law, medicine); another treats it as a core of AI literacy. Which view aligns with your own experience of what actually separates good prompts from bad ones?
• Well-designed prompts can scaffold student thinking, while poorly used ones can encourage cognitive offloading. Can you recall a moment when an AI answer did the thinking for you? What about the prompt — or your intent — made that happen, and could it have been designed to do the opposite?
• Prompting is both a learner skill and a system-design lever: some tutors now automatically route and select prompts for the user. As prompting moves from the user to the system, what do students lose — and what do they gain?
• Set a small goal before you read: after learning about prompt engineering, decide on one concrete way you'll change how you write prompts in your own work, and what result you'll check to know it worked.

Introduction

Prompt engineering is central to effective Generative AI use in education. Unlike traditional programming interfaces, LLMs respond to natural language — but the quality, accuracy, and pedagogical value of those responses depend heavily on prompt design. Research in this knowledge base reveals that prompting is not a neutral act: it reflects how students think, plan, and allocate cognitive effort. Miles, Haber-Curran and Arar (2026) sharpen what the term covers by distinguishing prompt engineering, the technical optimization of inputs for performance, from prompt literacy, the rhetorical, ethical and reflective work of clarifying purpose, reading output critically and revising with stated reasons. Their Prompt Literacy Cycle (Clarify Purpose, Craft the Prompt, Engage with Output, Refine the Prompt, Reflect) and a sample process rubric make the distinction teachable, and they argue that instruction which optimizes outputs alone leaves the ethical and epistemological dimensions of LLM use untouched.

How prompt engineering appears in the research

• Prompting as cognitive trace: Misiejuk et al. show that prompt patterns reveal cognitive offloading — high-quality work uses context-rich, polite, and instructional prompts; low-quality work shows reactive disagreement without domain grounding

• Depth improves the product, not the retention. In 22 postgraduates, the share of explanation-seeking ("why/how/explain") prompts predicted independently marked task quality (β = 6.27) beyond baseline knowledge and prompt volume, yet showed a null association with immediate recall (Tsiligkiris (2026)).

• Competence is the stratification variable, and it is trainable. In a survey of 380 Chinese undergraduates, prompt-engineering competence correlated strongly with perceived academic gains (r = .793) but only weakly with perceived educational inequity (r = −.131), and family digital capital worked largely through competence rather than alongside it. A within-subject experiment then showed a five-component scaffold — Role, Goal, Input, Process, Output — lifting rated output quality from 3.318 to 4.603 across four model families. The equity reading matters more than the effect size: if the benefit runs through a skill, then who receives prompt instruction is a distribution decision (Li et al., 2026).

• Prompt cognition tracks the discipline, not the student. Chang and Li (2026) classified 60,087 prompts from 116 courses and found Bloom-level profiles differed by discipline — STEM Apply-prevalent (20.8%), social science Create-prevalent (33.8%) — with course-level variance exceeding student-level variance.

• Prompting as literacy: Tracing GenAI literacy and K-12 prompting literacy research frame prompting as a core AI Literacy component

• Prompt literacy is a perceived capability, and it still predicts engagement. In 829 design students across 40 intact classes, self-reported prompt literacy — six adapted self-report items, not observed prompting — predicted learning engagement at β = 0.539, 95% CI [0.473, 0.595] (Han and Cao, 2026). The association held regardless of functional dependence (p = 0.396) or emotional dependence (p = 0.941). Only cognitive dependence, substituting AI judgment for one's own, clearly thinned it.

• Novices default to trial and error, and blame the model. In a text-linguistics seminar, ten genAI novices refined prompts by trial and error, rarely reached for in-context examples, and overwhelmingly attributed poor outputs to the LLM rather than their own prompt formulation (Brocca & Garassino (2026)).

• Vague prompts leave a trace in the dialogue. The lower-quality group's signature was a self-repeating clarification loop (IC→IC, Z = 3.25), traced to vague prompts and AI misreadings, and carrying the study's largest frustration gap (2.12 vs 3.82); prompt debugging on authentic logs targets it as it happens (Fang et al. (2026)).

• When the instructor models the prompt, students reuse it verbatim. Among 310 recorded prompts from twelve middle-school STEAM groups, exact copying of the instructor's instruction was the most frequent student standpoint at 47.1%, ahead of spontaneous inquiry at 28.7% (Zhao & Li (2026)).

• A regulatory cycle beats a prompt formula. In a quasi-experimental pilot with 42 undergraduates, students taught the IDEA cycle (Intent, Deconstruction, Expression, Adaptation) produced higher-quality prompts and outputs than peers taught Role–Task–Context–Format prompting in all five task categories (adjusted prompt gains of +11.77 to +29.19 points) (Wang et al., 2026).

• Prompting as system design: CoTAL uses human-in-the-loop prompt engineering for formative assessment scoring; anchor-based prompting improves automated essay scoring

• Prompting is the entry skill, not the discipline. Gorsky (2026) frames AI Literacy for software professionals as the ability to manage agents rather than to prompt them, naming framing, specification, context engineering, verification, multi-agent orchestration and auditability as the skills a curriculum must assess (Gorsky (2026)).

• Adaptive prompt routing: Learning to Prompt treats prompt selection as part of the tutoring system itself — subject-aware prompt routing over 14 pedagogical features, where a stochastic router selects the best prompt per conversation. This shifts prompting from a learner skill into an adaptive system-design lever, improving engagement and efficiency (28.1% vs 19.6% exercise conversion in a real-world A/B test).

• Prompt modalities: Voice vs. text input research examines whether prompting modality affects learning outcomes

• Prompting beyond text — scientific illustration. Prompting can generate molecular and physical-chemistry figures quickly, but a locally persuasive rendering can still be wrong or carry representational bias, so students must interrogate AI-generated visualizations against chemical principles rather than trust them (Li et al. (2026)).

• Scaffolded prompting: Guided LLM scaffolding and critical engagement scaffolding teach structured prompting as a learning intervention

• Coach prompting with fading support: in ARPG+, a real-time coach diagnosed prompt quality across six dimensions and faded support as competence grew, lifting final prompt quality to 7.82 against 5.95 for static templates and 4.52 with no assistance (Ye et al. (2026)).

• Task decomposition has an optimum: generating tutor-training lessons in three segments produced the highest-rated lessons (mean 14.67) while a single pass scored lowest (10.67) and five segments fell back below three — moderate decomposition beats both extremes (Lin et al. (2025)).

• Prompt privilege and equity: Jin et al. show prompting expertise is unevenly distributed — users who phrase requests skillfully systematically get better output than those expressing the same intent less adroitly. Their Prompt Equity Transformer shifts prompt optimization from the user to the AI system, arguing that equitable output should be engineered into the model rather than demanded of novices.

• Prompt refinement plateaus; fine-tuning takes over from there. Iterative prompt design yielded diminishing item-quality gains in L2 listening Assessment, but fine-tuning GPT-4.1 on the optimized prompt — the prompt held constant — produced more contextually grounded and balanced items, isolating model adaptation rather than prompt craft as the next lever (Aryadoust & Wong, 2026).

• Prompting as situated professional judgment. Beyond literacy and system design, prompting can be framed as a disciplinary practice. The Dierickx et al. taxonomy for journalism treats task definition and prompting as a form of professional judgment exercised within a domain's epistemic and ethical norms — translating journalistic work into explicit tasks (newsgathering → sensemaking → editing → publication/distribution) makes assumptions, priorities, and ethical considerations visible, and turns prompting into a pedagogical tool for critical AI literacy. Its logic transfers to other knowledge-intensive professions (law, medicine, public policy).

• Prompt design as instructional specification. Neto and colleagues (2026) find in their systematic review of GenAI in healthcare education that prompt design functions as a form of instructional specification, encoding the cognitive targets and quality criteria implicit in expert authoring — yet only 34.8% of studies aligned generated content with instructional frameworks and only 34.8% reported prompting in enough detail to reproduce. Looi, Liu, and Sun (2026) further show how prompt architecture can embed pedagogical rules (correctness gates, anti-spoiler boundaries, goodbye gates) to constrain Large Language Models (LLMs) tutoring behavior in procedural domains.

• Rubric-guided and role-aware prompting. Yaşar et al. (2026) showed that rubric-guided prompting — treating the rubric as a semantic interface between human pedagogical intent and machine inference — drove LLM–human agreement on student design work from 54.75% to 81.25% (Cronbach's Alpha 0.393 → 0.798). Rubrics engineered for LLMs must balance precision and flexibility: too vague invites free interpretation, too rigid reduces the model to pattern-matching. Role-aware prompting — evaluating the same artifact under instructor, peer-reviewer, and grant-reviewer prompts — produced qualitatively distinct, epistemically different feedback, showing that prompt design shapes not just accuracy but the evaluative stance of the output.

• Prompt scaffolding can move a model further from the teacher. Almousa et al. (2026) had seven models produce paragraph-level writing feedback under three prompting strategies and found that adding category names or worked examples increased divergence from the teacher distribution for most of them, improving only Mistral-7B (0.2695 to 0.2398). The zero-shot baseline stayed closest, so focus-type instructions are something to test rather than assume.

• Context-aware prompting for assessment. Context-aware prompting of pre-trained language models automates the coding of collaborative problem-solving skills from process data, modeling dependencies between behavior codes and fusing cognitive and social abilities. This enables structured CPS analysis at scale and in real time, overcoming the labor intensity of manual coding schemes.

• Role-based templates and quality rubrics for teacher planning. Luo and Tahir (2025) empirically develop a prompt framework for children's STEAM arts lesson planning that pairs a Role (R) – Instructions (I) – End Goal (E) template (adapted from RISEN) with a "four points and one line" optimization rubric — standardized, practical, engaging, and complete, plus an extension dimension. Applying the rubric to critique and refine prompts kept generated plans acceptable to practicing art teachers (mean ratings above 4/5) while exposing recurring gaps (personalization, child-safety constraints, cultural bias) that plain one-shot prompting left unaddressed — showing prompt templates plus explicit evaluation criteria function as a quality-control scaffold for classroom generation.

• A five-part scaffold, tested across four model families. Role–Goal–Input–Process–Output prompts raised rated output quality from 3.318 to 4.603 (Δ = +1.285) over 400 matched pairs, with every family improving and the weakest baseline gaining most (Li et al. (2026)).

• Role and constraint design as the independent variable. Wang et al. (2026) compare two agents built on the same model and platform at temperature 0.3 whose only difference is how the prompt specifies role, skills, and constraints: an expert teacher agent answering from a bounded textbook knowledge source versus an empathic student-centered agent scripted to diagnose Misconceptions about AI and check comprehension. The role difference alone shifted learning performance, cognitive load, flow experience, and perceived empathy, showing that role specification is an instructional-design decision with measurable effects rather than a stylistic flourish (Pedagogical Agent).

• Prompting as assessment design. Xiong et al. (2026) map Evidence-Centered Design's student, evidence, and task models onto the parts of a scoring prompt, deriving five principles — intentionality, traceability, accountability, iteration, and equity; their ECD-built prompts scored narrative essays more accurately than zero-shot, few-shot, and chain-of-thought baselines.

• Model identity, not prompt wording, drives LLM–human coding agreement. In Keith et al. (2026) controlled decomposition, prompt type was nonsignificant (F = 1.77, p = 0.136, partial η² = .03) while the model factor carried partial η² = .19. Adding definitions or detailed instructions impaired listing detection even as it lowered errors on correct-definition judgments, and price did not track quality: Gemini 2.5 Pro cost $178.13 yet underperformed GPT-4.1 Full at $56.77.

• Inject the theory's name, not its full texts. Zambrano et al. (2026) show that named-theory prompting beat supplied papers on usefulness (4.40 vs 3.20) and concreteness (3.80 vs 2.40), while full texts won only theoretical alignment (4.60 vs 3.60).

Connections to broader concepts

Prompt engineering connects to Scaffolding — well-designed prompts can scaffold student thinking rather than bypass it. It intersects with Metacognition and AI Literacy, as effective prompting requires understanding both the AI's capabilities and one's own learning goals. The Cognitive Offloading research directly links prompt quality to whether AI use supports or undermines learning.

• Writing skill drives prompting, and both predict Vibe Coding success. In a preregistered CHI 2026 study (N=100), Thorgeirsson, Weidmann & Su found that written-communication proficiency predicted GUI-oriented vibe-coding performance (r = .29), with human-graded prompt quality mediating the link — response-process evidence that clear, structured prose translates into better natural-language programming prompts. Both writing skill and CS achievement were independent predictors, and CS achievement (r = .39) carried roughly twice the unique variance, so improving prompting alone is unlikely to fully substitute for programming fundamentals in LLM-native development.

• Prompting strategy predicts performance. An empirical study of 128 engineering students found that AI Query Efficiency (clear, well-structured prompts) and AI-Driven Problem Solving (strategic integration of AI output into reasoning) were the strongest predictors of academic success — even after controlling for GPA — indicating prompting is a teachable skill that shapes how effectively students learn with AI.

• Prompting style, not just prompt quality, tracks outcomes. Traits derived from 1,540 tutoring sessions linked conceptual questioning to exam performance while task-delegation behaviors correlated negatively — but the same traits failed to replicate the following semester, so they are behavioral patterns rather than stable skills (McNichols, Du and Lan (2026)).

• A usable taxonomy, and which prompt categories actually pay off. Jacobsen et al. (2026) translate technical strategies into the 3K model (Kontext, Kernauftrag, Klarheit — context, core task, clarity): eleven practice-oriented categories, each with a good/average/suboptimal rubric, and each tested as an experimental variation on feedback generated for pre-service teachers' learning goals. Domain-specific technical language was the decisive category — replacing subject terminology with everyday paraphrases significantly reduced feedback quality across three models (β = −0.412). Adding concrete examples and removing the chain-of-thought instruction produced no significant difference from the baseline in the first study; examples did help once the analysis was rerun with the best-performing model-prompt combinations (β = 0.52). Prompt quality and model choice together explained 42.8% of the variance in rated feedback quality, which is the paper's case that prompt engineering is a measurable and teachable competency rather than a stylistic preference — and that its categories are not interchangeable in effect size.

• Better prompts reduced iteration without raising output quality. Teachers given AI training before a lesson-preparation experiment wrote longer, more detailed prompts that needed less refinement, while untrained teachers used shorter prompts and required more subsequent iterations, yet blinded expert ratings of the resulting materials were equal (Appel et al., 2026). Prompt craft here bought fewer correction rounds, not a stronger first draft.

• Students name intent alignment as the difficulty, not syntax. Surveying 234 Chinese university students, Fan and Liu (2026) found refinement after a poor output was the most endorsed capability (72.2%) while setting output constraints was the least (66.2%, with 22.2% neutral). Strategy preferences barely moved between a research assignment and a creative brief: of four matched pairs only output-format specification differed after correction (0.26 points, dz = 0.233, BH-adjusted p = .002). A fixed role-task-format template drew the weakest support of any general strategy at 59.4%. Among the 12.8% who answered an open question, prompt-intent alignment appeared in 40.0% of comments, and adding more detail did not reliably fix a stuck exchange. All of it is self-report from one questionnaire.

Common Misconceptions

• "Prompt quality is a student attribute." Prompt cognition tracks the discipline, not the student: Chang and Li (2026) classified 60,087 prompts from 116 courses and found the discipline predicted prompt sophistication. Teaching students to prompt better addresses a variation that sits in the subject matter.
• "Students develop their own prompting practice." When the instructor modeled the prompt, students reused it verbatim — among 310 recorded prompts from twelve middle-school STEAM groups, exact copying of the instructor's instruction was common. Modeling produces reproduction, which is why it is a system-design lever rather than a transferable skill.
• "Prompting is a learner skill first and a system property second." This shifts prompting into an adaptive system-design lever, with measured effects on engagement and efficiency (28.1% vs 19.6% exercise conversion). The interface's prompting affordances moved behavior more than instruction in prompting did.

Connected Concepts

• Pedagogical Patterns — The scaffolding layer inside structured-use sequences
• Vibe Coding
• Guardrails
• Scaffolding
• AI Literacy
• Agentic AI
• Metacognition
• Curriculum Design
• Cognitive Offloading
• Writing
• K-12
• Generative AI
• Learning Design
• CS Education
• Higher Education
• Technologies — Umbrella: AI technologies and techniques (models, LLM training, robotics, RAG, agentic)

Connected Articles

• Who benefits from LLMs? Prompt engineering competence and equity in higher education — prompt competence predicts gains and mediates family digital capital; a five-part scaffold is trainable

• Generative artificial intelligence prompting strategies among Chinese university students across academic tasks — 234 students on what they find hard: intent alignment, not prompt syntax

• Comparing teacher-centered and student-centered agents based on prompt engineering: Effects on learning performance, cognitive load, flow experience, and empathy perception in physics learning — Agent role and constraint prompts as the design variable in physics learning (Wang et al. 2026)

• How to train your dragon: Evaluating prompting and fine-tuning for GPT-based item generation in L2 listening assessment — Prompting vs. fine-tuning for GPT-based L2 listening item generation (Aryadoust & Wong 2026)

• What students ask matters: LLM interaction depth, task quality, and immediate recall in higher education — What students ask matters: LLM interaction depth, task quality, and immediate recall (Tsiligkiris 2026)

• ARPG+: a simulation-based study of real-time coaching for educational LLM prompting — ARPG+: real-time coaching for educational LLM prompting

• A Taxonomy of LLM-Supported Tasks for Critical AI Literacy in Journalism — Task-based taxonomy of LLM tasks for critical AI literacy in journalism

• Same Question, Different Answer? Measuring and Mitigating Prompt Privilege for Equitable AI Access — Prompt Privilege: measuring & mitigating accessibility disparities in LLM access

• Principal Trait Analysis: Towards Deriving "Skills" in Human-AI Collaboration — Principal Trait Analysis: data-driven traits of human-AI collaboration

• LLMs in text linguistics teaching: An exploratory study with genAI novices in higher education — LLMs in text linguistics teaching

• The IDEA Framework for Metacognitively Regulated GenAI Use in Higher Education: Development and Exploratory Pilot Evidence — The IDEA framework for metacognitively regulated GenAI use

• Automatic Large Language Models Creation of Interactive Learning Lessons — LLM generation of interactive tutor-training lessons (Lin et al. 2025)

• Learning to Use AI for Learning: Teaching Responsible Use of AI Chatbot to K-12 Students Through an AI Literacy Module

• ASE-26: A Curriculum for Agentic Software Engineering as a Discipline

• Anchor Is the Key: Toward Accessible Automated Essay Scoring with Large Language Models Through Prompting

• Beyond Access: Guided LLM Scaffolding for Independent Learning in Undergraduate Statistics

• Learning to Prompt: Improving Student Engagement with Adaptive LLM-based High-School Tutoring

• Cognitive offloading in student–AI collaboration: A longitudinal analysis of prompting strategies

• Tracing GenAI Literacy: Student-AI Interaction Patterns in Academic Writing

• Translating UNESCO artificial intelligence guidelines to chemical education and its intersection with sustainable development goals — UNESCO AI guidelines translated to chemical education; epistemic drift

• An empirical study of ChatGPT use in engineering education: Prompting and performance — Prompting behaviors predict engineering student performance

• Chat as Learning: Student-AI Conversations as Discipline-Associated Cognitive Engagement Patterns — Discipline-associated Bloom-level cognitive engagement in student-AI conversations (Chang & Li 2026)

• From evaluation to emulation: LLMs as agents of iterative pedagogical design — LLMs as agents of iterative pedagogical design

• ChatGPT-assisted lesson planning for children's STEAM arts education: An experimental study on benefits, challenges, methods, and a prompt framework

• Prompts to Practice: A Pedagogical Framework for Human-Centered AI Engagement — Prompt engineering vs prompt literacy: a five-phase human-centered GenAI engagement framework with a five-step Prompt Literacy Cycle (Miles, Haber-Curran & Arar 2026)

• AI Literacy of Teachers: Prompt Engineering and Model Selection as Predictors of AI-Feedback Quality — Prompt engineering and model selection as predictors of AI-feedback quality (Jacobsen et al. 2026)

• Evaluating feedback focus and pedagogical adaptivity in LLM-generated feedback on student writing — Evaluating Feedback Focus and Pedagogical Adaptivity in LLM-Generated Feedback on Student Writing

• Exploring Middle School Students' Interactions and Perceptions of Using Generative AI in STEAM Learning Environments — Middle-school STEAM groups copied the instructor's instruction verbatim in 47.1% of prompts

• Using Evidence-Centered Design to Guide Prompt Engineering in Measurement: An Illustration with Automated Essay Scoring — Using Evidence-Centered Design to Guide Prompt Engineering in Measurement: An Illustration with Automated Essay Scoring

• Human–AI collaborative patterns in GenAI-supported instructional design among pre-service teachers: an analysis of interaction sequences and cognitive load — Vague prompts produced self-repeating clarification loops with the largest frustration gap

• Who benefits from LLMs? Prompt engineering competence and equity in higher education — Prompt engineering competence as an equity pathway, with a cross-model scaffolding experiment (Li et al. 2026)

• Learning engagement and creative self-beliefs in generative AI-supported design education: self-reported prompt literacy and dimensions of AI dependence — Prompt literacy as a perceived capability that still predicts engagement