The rise of generative AI in education has created a need for common practices around annotation, disclosure, and validation. Without clear guidance, stakeholders may struggle to understand how GenAI contributes to content or whether the necessary safeguards are in place. Interest in developing this guidance members raised concerns about tracking and compliance, transparency for trust and accountability, ownership, and ethical and security considerations. Members and 1EdTech agreed there was a clear need for a shared framework, leading to the work on this collaborative effort. This document is designed to serve multiple contexts for institutions and providers. It responds to member-identified needs by offering a taxonomy and framework for annotating GenAI use, supporting both regulatory alignment and shared trust across the education ecosystem.

This document is intended for use by publishers, edtech providers, educators, and institutions across K–12, primary and secondary, higher education, and workforce learning globally. It does not prescribe a single implementation pathway but instead provides recommendations, examples, and frameworks that organizations can adapt to their own contexts. There has been consideration that there are different levels of use and restrictions on AI use worldwide and some countries and regions can be more or less stringent with their policies. The focus is on labeling, annotating, and reviewing AI-generated content in ways that ensure educational integrity, accessibility, and fairness.

Key audiences include:

• Publishers and content providers who develop instructional and assessment materials
• Institutions and educators who use GenAI-assisted tools to adapt and deliver content
• EdTech companies integrating GenAI into platforms and services

This Best Practices document is non-normative—it does not establish certification, compliance, or conformance requirements. Instead, it offers recommendations, frameworks, and examples that institutions, providers, and publishers can adapt to their own contexts.

The guidance reflects input from member organizations, emerging state legislation, and market exemplars, but it is not prescriptive. 1EdTech encourages its members and the broader community to provide feedback and participate in the ongoing development of this work. To engage with the 1EdTech AI-Generated Content Best Practices Community, please visit the 1EdTech AI-Generated Content Best Practices Community (member login required) or learn more at 1EdTech.org.

For related guidance on issues such as privacy, bias, or hallucinations, members are encouraged to use companion resources such as the 1EdTech TrustEd Apps™ Generative AI Data Rubric , AI Preparedness Checklist, 1EdTech AI Data Rubric that is in development, or contact 1EdTech at support@1edtech.org.

Stakeholders benefit when it is clear where GenAI was used, what role it played, and what level of subject matter expert oversight was applied. Publisher-produced content may call for structured labeling and metadata; educator-produced and student-produced content may emphasize classroom-level disclosure; and hybrid content may combine both approaches.

Human experts play a central role in ensuring accuracy, rigor, and appropriateness. For publisher-produced and hybrid content, subject matter experts are encouraged to validate outputs; for educator-produced content, teachers are best positioned to provide oversight; and for student-produced work, guardrails can help keep the focus on learning objectives.

GenAI can support equity and inclusion when carefully reviewed. Publisher and hybrid content may benefit from formal bias audits, while educator- and student-produced content may focus more on cultural sensitivity and fairness in classroom interactions.

AI-generated or GenAI-assisted content should be designed with accessibility in mind, following internationally recognized standards such as WCAG, EN 301 549, and WAI-ARIA, where applicable. While not a mandatory requirement in this policy, adherence to these standards is strongly recommended to help ensure that digital content is perceivable, operable, understandable, and robust for users with diverse abilities.

The use of GenAI should strengthen, not diminish, confidence in educational resources. For publisher and hybrid content, practices such as provenance tracking and SME review help build trust. Transparency and consistent use of the guidance will help students and educators to build trust in the content.

The development of this Best Practices document is based on both member research and emerging legislation from around the world, such as US states and the European Union's Artificial Intelligence (AI) Act. 1EdTech members and staff have compiled resources from across the community and reviewed some of the recent state-level laws that address transparency, disclosure, and oversight in GenAI use.

The Implementation Framework translates guiding principles into recommended practices and options that institutions and providers can adopt. While approaches may vary by context, the following recommendations are consistently emphasized across this document:

1. Transparency and Labeling
   • Clearly label when and how GenAI contributed to content creation.
   • Ensure disclosure metadata follows content across platforms (via TCC, QTI, etc.). (Examples)
2. Provenance Tracking
   • Maintain records of GenAI involvement, model version, and human oversight.
   • Store provenance metadata with content to support auditability and accountability. (Examples)
3. Risk-Based Oversight
   • Apply different levels of review depending on whether content is high-, medium-, or low-risk in educational outcomes.
   • Prioritize subject matter expert validation for high-stakes use. (Examples)
4. Subject Matter Expert (SME) Review if the Organization Requires Review:
   • Require human expert validation to ensure accuracy, rigor, and alignment with standards and also provide transparency of the review (when and how it happened, potentially by whom)?
   • An organization’s staff, educators, or SMEs should fact-check content for accuracy, evaluate for potential bias, and review for accessibility compliance, with particular attention to disciplines such as science, mathematics, and social studies where errors or omissions may carry greater consequences. (Examples)
5. Bias and Accessibility Checks
   • Conduct bias audits to identify inequities. It is suggested to assess content to identify ableist terms, harmful stereotypes, or misleading information about disability or disability-related topics.
   • Validate accessibility conformance to accessibility standards and compatibility with assistive technology
   • Continuously test and monitor AI systems for bias using defined fairness metrics and ongoing post-deployment checks to ensure equitable outcomes.
   • See the 1EdTech TrustEd Apps Accessibility Rubric and the 1EdTech AI Data Rubric for more information (Examples)
6. Metadata & Storage Practices
   • Store GenAI-use disclosures and provenance together with content.
   • Adopt versioning and audit trails, especially in high-stakes contexts.
   • Comply with regional accessibility laws, based on country or region (e.g., GDPR, FERPA, ADA, EAA). (Examples)
7. Alignment with Open Standards
   • Use interoperability standards (QTI, Common Cartridge/Thin CC, LTI, Caliper) to ensure transparency and metadata portability across systems. Organizations who use interoperability standards, may want to add annotations in the metadata. (Examples)
8. Public GenAI-Use Statements
   • Publish organizational statements summarizing how AI is used, for what purposes, and what safeguards are in place.
   • Link these policies to content-level metadata for a layered approach to trust. (Examples)

This section breaks these best practices into specific focus areas—such as risk rating, subject-specific review, accessibility, metadata and storage, and standards alignment—that help institutions and providers put the principles into action. Each subsection offers a lens on implementation: how to recognize different levels of risk, how to address the nuances of content across disciplines, how to safeguard accessibility, and how to ensure transparency and portability through metadata and standards. Taken together, these elements form the building blocks of a practical framework for responsible use of AI-generated content.

AI-generated content in education can be grouped into several categories, each with distinct risks and oversight needs. Publisher-produced content requires the highest standards of transparency, provenance, and bias auditing, as it forms the foundation of curriculum and assessment. Educator-produced content, while more localized, still calls for clear disclosure of GenAI assistance and safeguards against bias.

Student-produced content raises unique concerns around fairness and academic integrity, especially when GenAI influences evaluation or feedback. Finally, hybrid or co-produced content should always follow the stricter requirements of the categories involved, with clear documentation of the human and GenAI contributions. By tailoring oversight to these categories, organizations can ensure compliance, accountability, and trust across the learning ecosystem.

AI-generated content can play very different roles in education, from influencing grading and placement decisions to supporting lesson planning or providing drafting assistance. Because these uses carry different levels of potential impact, this framework proposes a three-tier risk model. The categories that follow—high, medium, and low risk—are offered as recommendations to help institutions and providers reflect on how AI-generated content may affect educational outcomes and what level of oversight may be most appropriate. This model is not prescriptive, but rather one way to approach risk in a structured, transparent manner. Institutions and individuals are encouraged to adapt the categories to their own policies, risk tolerance, and regulatory environment.

AI-generated content presents different challenges depending on the subject area. In some disciplines, accuracy is largely objective, while in others, nuance and interpretation play a larger role. To maintain quality and trust, GenAI outputs should be reviewed with subject-specific considerations in mind:

• Math – Accuracy is critical, both in problem-solving and in the presentation of equations or symbols. Best practices include rigorous accuracy checking, verification of solutions, and ensuring styling aligns with established mathematical conventions.
• Language and Literacy – Whether in English or any other language, AI-generated text must be fact-checked for accuracy and reviewed for clarity, tone, and grade-level appropriateness. In addition, linguistic precision and cultural sensitivity are essential, particularly in world language instruction, where errors in vocabulary, grammar, or context can distort meaning. Best practices include ensuring fidelity to the target language, avoiding literal translations that lose nuance, and maintaining cultural authenticity.
• Science – GenAI outputs must be checked against authoritative sources to confirm factual accuracy. Additionally, scientific terminology, units of measure, and formatting should be consistent with domain standards.
• Social Studies – Fact-checking is especially important to avoid inaccuracies, oversimplification, or biased perspectives. Content should reflect multiple viewpoints where appropriate and align with curricular frameworks.
• Other Disciplines – Beyond core subjects, AI-generated content must also be reviewed with discipline-specific nuances in mind. In the Arts, reviewers should confirm that creative expression is authentic and appropriately contextualized. In technical and vocational subjects, accuracy of procedures, tools, and terminology is essential. For interdisciplinary or project-based learning, GenAI outputs should be checked for coherence across domains and for alignment with intended learning outcomes. While the details vary by field, the guiding principle remains the same: subject matter experts must validate GenAI contributions to ensure accuracy, appropriateness, and educational integrity.

By tailoring review processes to subject-specific needs, organizations can reduce the risks of error or bias while ensuring AI-generated content maintains pedagogical integrity across the curriculum.

Storage practices for AI-generated content should be designed to support transparency, accountability, and compliance across the full lifecycle of content. While the recommendations below reflect broadly applicable best practices, institutions and providers must recognize that storage policies may be shaped by additional factors such as sector requirements, institutional risk tolerance, and the stakes of the content involved.

Recommended practices include:

• Store Content and Metadata Together AI-generated content should always be stored with its associated metadata—provenance records, disclosure labels, timestamps, training reuse declarations, and review history—so that the context of creation is preserved. Separating metadata from content risks losing critical transparency and accountability information. Institutions and/or the tech provider will own the content, metadata, and prompts, depending on the use case and ownership agreement in the procurement process.
• Enable Interoperability Across Platforms Because content often originates in one platform and is delivered through another, storage systems should ensure metadata portability. Standards such as Thin Common Cartridge (TCC) and Learning Tools Interoperability (LTI) make it possible for disclosure and provenance information to remain intact across systems.
• Implement Versioning and Audit Trails AI-generated content often evolves through multiple cycles of human and GenAI contributions. Some organizations, like summative assessment providers, store Ai-generated content. Depending on the organization's policies around storing content, the best practice is to maintain version control and audit logs that record when GenAI was used, the nature of changes, the model version employed, and who validated the outputs. (See ETS’s example) This helps ensure compliance and supports institutional accountability.
• Record Training Reuse Declarations Metadata should indicate whether content, student data, or educator-generated materials will be used to train or refine GenAI models. This practice clarifies ownership, permissions, and downstream risks and ensures alignment with institutional and legal expectations.
• Ensure Security and Compliance Storage must align with regional data protection and privacy laws (e.g., FERPA, GDPR) and institutional requirements. This includes access controls, encryption, retention policies, and audit readiness.
• Sector-Specific Requirements In some contexts, additional storage considerations apply. For example, high-stakes assessment organizations such as ETS have emphasized stricter retention, audit, and evidentiary standards for assessment materials. Where AI is used in summative assessment, placement, or credentialing contexts, storage solutions should reflect the higher compliance and legal risks associated with these use cases.

These practices are consistent with emerging frameworks such as the 1EdTech TrustEd Apps Generative AI Data Rubric which recommends disclosure of when GenAI is in use, what data sources are involved, and whether stakeholders may opt in or out of training reuse. Aligning with these practices ensures storage solutions not only meet compliance needs but also reinforce trust in AI-generated educational content.

Accessibility is a critical dimension of AI-generated content, as automated tools can both introduce new opportunities for inclusion and create risks if outputs are not carefully reviewed. Unlike human-created materials, AI-generated text, images, and multimedia often lack built-in safeguards for accessibility. For example, GenAI may generate alt-text that is vague or inaccurate, produce transcripts that omit context or produce inaccurate content, or create diagrams that are visually polished but unusable with assistive technologies.

Institutions and providers are encouraged to treat accessibility as an integrated part of the AI content workflow. This includes:

• Checking AI-generated outputs against relevant accessibility standards, frameworks and laws to ensure all learners can engage equitably with content. An example of a framework is the Framework for Accessible and Equitable AI in Education. Key frameworks and laws include WCAG 2.1 AA and Section 508 (U.S.), the European Accessibility Act (EU), and the Accessible Canada Act. Institutions and providers should also be mindful of regional laws in other jurisdictions, such as the New Zealand Web Accessibility Laws, India’s Rights of Persons with Disabilities Act, Brazil’s eMAG (Electronic Government Accessibility Model), and the Dubai Universal Design Code. These frameworks, while varied, share a common expectation: AI-generated educational content must meet accessibility requirements to avoid exclusion and ensure compliance.
• Validating AI-generated metadata and alt-text to ensure it accurately conveys meaning, particularly for complex content such as math equations, scientific diagrams, or charts.
• Reviewing automatically generated captions or transcripts for accuracy, since GenAI systems often misinterpret technical or domain-specific vocabulary or fail to recognize the varied ways people with disabilities communicate.
• Applying universal design principles so that AI-generated materials support diverse learning needs, regardless of jurisdiction, and make sure it is also compatible with assistive tech. (See the 1EdTech TrustEd Apps Accessibility Rubric for more information)

Because accessibility laws vary by region, institutions and providers should follow the requirements applicable to their context (e.g., ADA in the U.S., EAA in Europe) while adopting a proactive approach that treats accessibility as a baseline expectation for all AI-generated content. Embedding accessibility early reduces remediation costs and ensures that machine-generated materials support equitable participation for all learners.

There are multiple ways institutions and publishers can disclose the role of GenAI in educational content, but what matters most is that a clear labeling system is in place. The GenAI Content Transparency Rating System presented here is one example of how content can be categorized to show the degree of GenAI involvement and the extent of subject matter expert review. By using a structured approach—whether this one or a comparable model—organizations can ensure transparency, build trust with educators and learners, and meet emerging regulatory expectations, and do so in adherence to professional citation standards. The table that follows illustrates one labeling framework, demonstrating how content can be classified from fully human-created to fully AI-generated.

Best practices for AI-generated content extend beyond technical considerations to include the responsible conduct of all content producers. Whether the creator is a publisher, educator, or student, transparency and disclosure are essential for maintaining trust, academic integrity, and alignment with institutional or professional standards.

Recommended practices include:

• Clear disclosure of GenAI contributions by any content producer (publisher, educator, or student), with labeling that makes the extent of AI involvement visible.
• Alignment with institutional or professional integrity codes, so GenAI use is governed by the same principles as citation, authorship, and originality.
• Role-appropriate oversight — publishers ensure compliance and SME review; educators validate appropriateness for learning; students differentiate their own work from GenAI support.
• Consistent policies and consequences, communicated institution-wide, so expectations for responsible GenAI use are clear regardless of role.

Strong metadata and storage practices are essential to ensuring transparency, accountability, and trust in AI-generated content. These practices help organizations track how content was created, communicate GenAI involvement clearly to end users, and maintain consistency when content is delivered across multiple platforms. The following best practice recommendations address provenance, disclosure, and public transparency.

A recommended best practice is to maintain provenance records that document the origin and history of AI-generated content. These records should capture information such as the GenAI system or model used, the version, the date of creation, and the role of human or subject matter expert oversight. By storing provenance metadata alongside the content itself, institutions and providers can address questions of authorship, accuracy, bias, or compliance long after the content is delivered. Provenance metadata also enables interoperability, ensuring that information about content creation persists when content moves between systems.

Disclosure statements are recommended as a practical way to inform users that GenAI contributed to the creation of content. These statements can take the form of visible labels, annotations, or notes embedded within platforms. In addition to surfacing at the point of use, disclosure information should be stored as metadata, linked directly to the labeling strategy (e.g., human-authored, GenAI-assisted, AI-generated). This approach ensures that disclosure remains consistent when content is shared across platforms. For example, metadata carried through 1EdTech’s Common Cartridge® (CC), Thin Common Cartridge (TCC) or Question & Test Interoperability® (QTI®) can allow a disclosure label applied by a publisher to be displayed in the institution’s LMS, even though the delivery platform differs from the origination platform. This consistency is a cornerstone of trust and transparency. This group has not worked on a technical solution around the metadata format or model within these standards specifications, but the suggestion is to look into a format in the future with the technical experts in those standards’ groups.

As a complement to provenance tracking and disclosure labels, organizations are encouraged to publish public GenAI-use statements or policies describing how GenAI is used across their content and services. These statements should outline the purposes of GenAI use, safeguards in place to protect fairness and accessibility, and the organization’s overall approach to responsible GenAI. A public statement builds credibility with educators, learners, and policymakers, while also signaling alignment with emerging regulatory expectations. Linking organizational statements with metadata practices and labeling strategies creates a layered approach: detailed provenance at the content level, consistent disclosure at the point of use, and organizational accountability through public communication.

For AI-generated or GenAI-assisted assessments, metadata should be embedded using QTI structures. This allows provenance and disclosure information—such as whether items were generated with GenAI or validated by a subject matter expert—to be preserved when assessments are exchanged between authoring systems and delivery platforms. Best practice is to use QTI’s extensible metadata fields to capture GenAI-related annotations without altering the underlying assessment logic.

AI-generated content used in course materials can be packaged with metadata labels in Common Cartridge or Thin CC formats. When disclosure statements or risk ratings are included at the content-object level, they will carry through when imported into a Learning Management System (LMS). Best practice is to align GenAI labeling with existing metadata fields so that LMSs can display consistent transparency information regardless of whether content originated with a publisher or educator.

When AI-generated content is accessed through third-party tools using LTI (Learning Tools Interoperability), labeling and provenance metadata should remain visible to the end user within the provider’s environment. While LTI does not itself transmit metadata, best practice is for the content provider’s platform to surface disclosure labels (e.g., “Human-Created,” “AI-Assisted,” “AI-Generated”) in a consistent and transparent way once launched. Institutions should ensure that their LTI-connected tools support label visibility and provenance consistency—so that whether content originates in a publisher system or a local LMS, users encounter the same transparency indicators. In other words, LTI serves as the access pathway, while the content provider maintains and presents the labeling metadata.

For tracking the use and impact of AI-generated content, Caliper provides a framework for capturing learner interactions. Institutions and providers may extend Caliper event models with fields indicating whether the content or activity was AI-generated or GenAI-assisted. This enables more granular analytics on how GenAI-labeled content is being used in practice and supports institutional oversight. Best practice is to incorporate GenAI labeling into Caliper events in a way that aligns with privacy and data governance requirements. Together, these standards provide the infrastructure to make AI labeling and metadata portable, visible, and trustworthy across the education ecosystem. Adopting them as part of best practice ensures that transparency does not stop at the point of creation but extends through delivery, use, and evaluation.

• Communicate transparently when AI has been used to generate content through clear labeling and disclosure.
  • Example from Unicon: Unicon defers to the client for their use cases. For one of their clients they included a statement at the beginning of the course that GenAI tools were used as co-collaborators in the creation of the content. Unicon recently released our AI Tenets on Linkedin which emphasize that “we are human first, AI-enabled.”
  • Example from D2L: For D2L AI internal tooling, they have identified AI status: list of status. AI content is gated until a human reviews it, only when users use their internal AI tool (see section on AI Utilization for list of statuses tracked in data sets)
  • Example from EU AI Act: The EU AI Act is directed towards providers of General Purpose AI (GPAI) systems.
  • Example from the state of California: CA requires the labeling of AI- generated documents.

• Software providers and educators should maintain provenance records and/or metadata that support trust, accountability, and portability across platforms, such as model, version, hosted themselves, number of tokens used, prompts, date, link to the LLM, metadata, etc.
  • Example from ETS: Tracking of AI-generated content is standardized across ETS programs. AI-generated content that is banked externally carries metadata information that includes Source (e.g., “Artificial Intelligence”; differentiating a human contribution from an AI-assisted contribution to the item pool), Label (identifying the AI-enabled platform from which the item was sourced), and SessionID (identifying the full AI-human conversation that led to the original draft item).

• Apply risk-based oversight that differentiates between high-, medium-, and low-stakes use cases. The AI Generated Content Best Practices is in some alignment with the EU AI Act, in that there are recommendations based on risk-assessment levels of the AI systems.
  • Example from Carnegie Learning: Any AI-generated content that directly influences educational outcomes (grades, placement, etc) requires human validation by a SME.
  • EU Artificial Intelligence Act (EU AI Act): The Annex III use cases list high risk AI systems. The EU AI Act has prohibited the use of some types of AI use and has risk levels of AI use including:
    • Acceptable Risk
    • Minimal Risk
    • Limited Risk
    • High Risk- The majority of the High-Risk level is placed on providers. There is also an EU AI Act Compliance Checker to inform providers, developers, and users if or how the EU AI Act may or may not apply to them.

• Incorporate review by the organization’s subject matter experts (SME) for accuracy, rigor, and alignment with disciplinary and institutional standards.
  • Example from D2L: The educator /SME is encouraged to review content created by an AI engine prior to posting in a class in their LMS.

• Verify accessibility needs and validate AI-generated supports such as alt-text, captions, or formatting.
  • Example from Teach Access: Teach Access discloses how AI tools are used in the development of their educational content and resources. Any AI-generated content, such as alt-text, captions, and plain language adaptations, is reviewed by staff or external collaborators to ensure accuracy, accessibility, and adherence to quality standards before publication.

• Establish storage practices that retain provenance, disclosures, and usage logs appropriate to context in alignment to your organization’s retention policy if relevant..
  • Example from SameGoal: AI-generated content can be stored for seven years or longer depending on the customer’s needs.
  • Example from ETS: AI-generated assessment content is retained indefinitely.

• Support responsible use and integrity across publishers, educators, and students, with clear expectations and disclosure policies.
  • Example from ETS: AI-generated content is noted in the test administration directions provided to a proctor or test taker.
  • Example from North Carolina Department of Public Instruction- In educational contexts, educators and students are responsible for reporting how they used Artificial Intelligence (AI) in the generation of materials, assignments, or instructional resources. By modeling transparent practices, educators help students understand how to responsibly integrate AI into their learning and creative processes. This expectation reinforces ethical technology use as a key component of digital citizenship and academic integrity.
  • Carnegie Learning AI Content Transparency Rating System for Educational Publishing
  • Digital Promise AI Literacy: A Framework to Understand, Evaluate, and Use Emerging Technology
  • Educational Testing Service (ETS)
    • Responsible Use of AI In Assessment: ETS Principles AI
    • Responsible AI for Measurement and Learning: Principles and Practices
    • Foundational Policy Ideas for AI in Education (published through teachAI.org)

1. Feldman-Maggor, Y., Cukurova, M., Kent, C. et al. The Impact of Explainable AI on Teachers’ Trust and Acceptance of AI EdTech Recommendations: The Power of Domain-specific Explanations. Int J Artif Intell Educ (2025). https://doi.org/10.1007/s40593-025-00486-6
2. Denny, P., Khosravi, H., Hellas, A., Leinonen, J., & Sarsa, S. (2023). Can we trust AI-generated educational content? Comparative analysis of human and AI-generated learning resources. arXiv preprint arXiv:2306.10509. https://ui.adsabs.harvard.edu/abs/2023arXiv230610509D
3. SCALE Initiative. (2024). Understanding students’ acceptance, trust, and attitudes towards AI-generated images for educational purposes. arXiv preprint. https://scale.stanford.edu/genai/repository/understanding-students-acceptance-trust-and-attitudes-towards-ai-generated-images
4. Córdova-Esparza, D.-M. AI-Powered Educational Agents: Opportunities, Innovations, and Ethical Challenges. Information 2025, 16, 469. https://doi.org/10.3390/info16060469
5. Fulsher, A., Pagkratidou, M. & Kendeou, P. GenAI and misinformation in education: a systematic scoping review of opportunities and challenges. AI & Soc (2025). https://doi.org/10.1007/s00146-025-02536-y
6. Xiaoyu, W., Zainuddin, Z., & Hai Leng, C. (2025). Generative artificial intelligence in pedagogical practices: a systematic review of empirical studies (2022–2024). Cogent Education, 12(1). https://doi.org/10.1080/2331186X.2025.2485499
7. Yusuf, A., Pervin, N. & Román-González, M. Generative AI and the future of higher education: a threat to academic integrity or reformation? Evidence from multicultural perspectives. Int J Educ Technol High Educ 21, 21 (2024). https://doi.org/10.1186/s41239-024-00453-6
8. Melo-López, V.-A.; Basantes-Andrade, A.; Gudiño-Mejía, C.-B.; Hernández-Martínez, E. The Impact of Artificial Intelligence on Inclusive Education: A Systematic Review. Educ. Sci. 2025, 15, 539. https://doi.org/10.3390/educsci15050539