Educators and institutions across the U.S. have begun developing frameworks to guide pedagogically sound uses of generative AI in high school classrooms. These models emphasize instructional design that keeps learning at the forefront, providing actionable strategies for teachers to implement AI tools in ways that build student skills (like critical thinking, inquiry, and creative production) rather than undermine them. Below, we summarize several prominent frameworks and models – all tested or implemented in real educational settings – and then compare the common strategies they promote (such as human-AI collaboration, cognitive scaffolding, and project-based learning).

Washington’s Human-Centered AI Approach (OSPI’s H-AI-H and SHIFT Frameworks)

One comprehensive example comes from the Office of Superintendent of Public Instruction (OSPI) in Washington State, which has issued human-centered AI guidance for K–12. At its core is a “Human-AI-Human” (H-AI-H) philosophy: educational AI use should always start with human inquiry and end with human insight and empowerment, ensuring that students and teachers remain central at the beginning and conclusion of any AI-aided learning process. This approach is reflected in multiple tools and frameworks from OSPI:

• SHIFT Framework: OSPI’s guidance includes a “SHIFT” framework to scaffold students’ critical thinking about AI outputs. The acronym stands for five steps that teachers can coach students through when using AI:

  • S – Start your curiosity engine: Consider what intrigues you about the AI’s output or response.

  • H – Hone in on a detail: Identify specific details the AI got right or wrong, and explain how you know.

  • I – Identify your context: Relate the AI’s information to the bigger picture of the assignment or topic at hand.

  • F – Frame it from a new perspective: Think of a different perspective or angle – could AI help uncover an alternative viewpoint?

  • T – Talk about what’s missing: Reflect on the limitations or gaps in the AI’s answer (what did it not or cannot address?).

    This framework explicitly cultivates a questioning mindset and deeper inquiry into AI-generated content. By prompting curiosity and scrutiny, students practice analyzing the “intricacies and implications” of AI applications (e.g. checking for biases or inaccuracies), connect AI outputs to broader societal or ethical contexts, and exercise creativity by considering new perspectives. The final step ensures students acknowledge what AI cannot do, reinforcing metacognition and responsible use of AI in their future endeavors. In short, SHIFT turns AI use into an opportunity for developing critical thinking skills rather than passive consumption of AI answers.

• 5-Level Scaffolding of AI Integration: In addition to SHIFT, Washington’s model describes a graduated “AI Integration Scale” for classroom tasks, from Level 1: No AI up to Level 5: AI as Co-Creator. At Level 1, students rely on their own knowledge with no AI assistance. By Level 2, AI-assisted brainstorming is allowed (AI can help generate ideas). Level 3 permits AI-supported drafting of initial versions, but any AI-generated material must be significantly revised and the final content must be the student’s own work. Level 4 involves AI-collaborative creation, where AI-generated content can be included so long as the student critically evaluates and edits the AI’s contributions. Finally, at Level 5, AI acts as a co-creator in extensive content generation – yet even here the student must ensure original analysis and thought, cite the AI’s role, and adhere to academic integrity. This scaffolded framework has been applied in real lesson plans to gradually build students’ capacity to work with generative AI. For example, in earlier levels a history student might only use AI to brainstorm essay topics, whereas at higher levels they might co-write portions of an essay with an AI and then refine it deeply to add their own insights. Teachers play a critical role at all levels – setting guidelines (e.g. requiring AI use to be transparent and cited) and moderating AI’s influence so that it “augments rather than replaces” human thought. The key goals served here are to maintain academic honesty and to use AI as a learning aid that can be incrementally introduced as students develop stronger self-regulation and evaluative skills.

Together, OSPI’s human-centered approach and frameworks like SHIFT emphasize critical inquiry, ethical use, and structured skill-building. Students learn to treat AI as a tool to be questioned and examined, not an oracle. By starting and ending with human thought (H-AI-H loop) and by explicitly teaching steps to analyze AI output, this model promotes critical thinking, scientific inquiry (through constant questioning and evidence-checking), and even creative thinking (framing problems in new ways). It has been piloted in Washington schools and aligns with broader efforts to prepare students for an AI-filled world while safeguarding the human elements of learning.

Decision Tree for Student AI Use (Stauffer & Gold)

Another actionable framework comes from educators Jen Stauffer and Jonathan Gold, who created a “Decision Tree” model to guide high school students’ use of AI tools. Featured on Edutopia in 2024, this model was born out of classroom experience and aims to help students make responsible, metacognitive decisions about when and how to leverage generative AI. It consists of a series of key reflective questions (a flowchart-like sequence) that students should work through whenever they consider using an AI application for an assignment. The framework’s steps ensure that AI is used only in pedagogically appropriate ways that genuinely support learning. The main decision points include:

1. Permission – “Am I allowed to use AI for this task?” Students first check the class/school policy or teacher’s instructions about AI for the specific assignment. Rather than assuming they can use ChatGPT undetected, the model encourages open discussion of AI use. Teachers are advised to set clear “red light, yellow light, green light” rules for AI (e.g. green = free to use with citation, yellow = use with caution or partial use, red = not allowed) and to build trust by explaining why certain uses are or aren’t permitted. This step reinforces digital citizenship and ensures that student use of AI begins within ethical and allowed boundaries.
2. Enhancement – “Will using AI enhance my learning?” Students must consider whether AI will augment their learning process or simply do the work for them. This prompts a metacognitive pause: Is the AI tool helping me achieve the learning objectives (for example, by clarifying a concept or inspiring a new approach), or is it short-circuiting the learning (for example, just giving me an answer I don’t understand)? The decision tree pushes students to use AI as a supplement, not a substitute for thinking. Teachers can support this by clearly communicating the goals of each assignment – when students grasp the purpose of a task, they are better able to judge if an AI’s help would truly deepen their understanding or if it would undermine the intended skill-building. This aspect of the framework targets the goal of critical thinking, training students to align tool use with learning goals.
3. Iterative Use – “Am I using AI iteratively (PROMPT and EDIT)?” Rather than treating a single AI query as a final answer, students are encouraged to engage in an iterative process with AI. Stauffer and Gold introduce two acronyms to scaffold effective use: PROMPT (Purpose, Role, Organize, Model, Parameters, Tweak) helps students design better prompts for the AI, and EDIT (Evaluate, Determine, Identify, Transform) guides them in analyzing and refining the AI’s output through multiple rounds. This essentially teaches a form of cognitive scaffolding for AI: students learn to plan what role to assign the AI (e.g. “act as a tutor and explain this step-by-step”), set parameters or examples, then critically evaluate the response and adjust their prompt to improve it. The model thus embeds “critical metacognitive friction” into AI use, ensuring students remain actively engaged. Not only does this improve the quality of AI assistance they receive, it builds skills in prompt engineering and self-directed learning. (Notably, this step aligns with other work like the “AI roles” framework by Ethan and Lilach Mollick, which suggests giving the AI a clear role like tutor or coach to get better results – see next section for more on that.)
4. Transparency – “Can I show how and why I used the AI?” The framework strongly emphasizes that students be transparent about their AI use. If they do use an AI tool, they should be prepared to explain their process, share AI conversation logs or screenshots, and properly cite or attribute any AI-generated content used. This is presented as a crucial step to prevent unethical use and to shift focus to process over product. In practice, teachers using this model have students annotate their assignments, highlighting where AI contributed and describing how it helped. There are even emerging citation formats (MLA, APA, Chicago) for AI which the framework points to for formal academic honesty. By making students document their AI interactions, the decision tree fosters accountability and ethical use, and it deters misuse (since students know they must disclose AI help, they’re less likely to copy answers blindly).
5. Reflection – “Am I reflecting on my use of AI?” In the final stage, the student steps back and reflects on how using the AI tool affected their learning. Did it help them achieve the learning goals? How did it influence their understanding or workflow? What would they do differently next time? This closing reflection reinforces that AI is not a magic shortcut but part of an ongoing learning process. It encourages metacognition, aligning with the goal of making students “not just tech-savvy but tech-wise” thinkers. Teachers can facilitate class discussions or journals about these reflections, turning individual experiences into shared learning moments about the role of AI in education.

Overall, Stauffer and Gold’s decision-tree model serves the educational goals of critical thinking, ethical inquiry, and student ownership of learning. It has been applied in classrooms to build students’ AI literacy – defined as the ability to thoughtfully and responsibly use AI tools – which the authors note is a “vital component of digital literacy and citizenship” today. By guiding students through permission, purpose, process, and reflection, this framework ensures that human judgment and learning remain central. It’s a clear example of how teachers can proactively integrate AI in day-to-day instruction while mitigating the risks of over-reliance or academic dishonesty.

AI as Collaborator and Tool: Seven Roles Framework (Mollick & Mollick)

At the university level, Professor Ethan Mollick and researcher Lilach Mollick at the University of Pennsylvania’s Wharton School have proposed a versatile framework that is influencing K-12 educators as well. In their 2023 paper “Assigning AI: Seven Approaches for Students, with Prompts,” they outline seven distinct roles or models for integrating AI into learning activities. Each role frames the AI in a different pedagogical capacity, with specific benefits and cautions, and the authors provide example prompts for each to demonstrate how students and teachers can practically implement them. The seven roles are:

• AI Tutor: The AI acts as a personal tutor, providing direct instruction or explanations on a topic. For example, a student might prompt an AI to “explain the causes of the Civil War in simple terms” or “walk me through this algebra problem step by step.” This can customize learning to the student’s pace and level (mimicking one-on-one tutoring). The benefit is on-demand clarification and teaching; the risk is that students might accept explanations uncritically or bypass struggle that is pedagogically useful. Teachers using an AI tutor would need to ensure students still practice problem-solving and confirm the AI’s information is correct.
• AI Coach: The AI provides feedback, hints, or encouragement as a student works through a task – more like a writing coach or a debate coach than a lecturer. For instance, a student writing an essay could ask the AI to “give me feedback on my thesis statement and suggest improvements,” or a coding student might have the AI debug their code and give hints. This role supports iterative improvement and skill refinement. It keeps the student in charge (the student is doing the work, the AI is guiding), but requires the student to critically evaluate the AI’s suggestions. The AI Coach’s value is in personalized feedback; the caveat is that feedback might be incorrect or one-dimensional, so students must be trained to verify and compare with human feedback.
• AI Mentor: In this role, the AI offers broader guidance or advice, perhaps on open-ended projects, creativity, or even academic/career questions. For example, a student could consult the AI on “How can I approach designing an experiment for my science fair project?” or “What are some strategies to improve my study habits?” The AI Mentor is meant to stimulate reflective thinking and goal-setting. It can expose students to new ideas and strategies (acting like a knowledgeable advisor), but there are clear warnings – AI systems don’t truly understand personal context or ethics, so their advice must be weighed carefully. The pedagogical goal is to give students another avenue for brainstorming and perspective, while teachers ensure that any guidance aligns with sound practice and student well-being.
• AI Teammate: Here the AI becomes a collaborator in group work or problem-solving. Students treat the AI as if it were another member of a team. For instance, a group might use an AI chatbot to generate possible solutions during a project (“Our AI teammate suggests we try X approach – let’s evaluate that”). In writing or media projects, a student might alternate writing paragraphs with an AI to co-create a story or presentation. The AI Teammate model gives experience in human-AI collaboration, mirroring how professionals might work with AI tools in the workplace. It can enhance creativity (by injecting new ideas) and help manage tasks (the AI can take on routine parts of a project). However, the framework cautions that students must practice “active oversight” – continually critiquing the AI’s contributions – so that they don’t blindly accept what the AI produces. The educational aim is to improve teamwork skills and creative problem-solving by leveraging AI, all while learning to critically filter AI input.