Beyond operational skills: Teachers' AI knowledge and interactions with generative AI in lesson planning

Synthesis: Velander (2026) investigates how K-12 teachers conceptualize and enact AI-related professional knowledge during GenAI-supported lesson planning, combining three data sources from two in-person workshops (April–May 2025; 75 K-12 teachers total, 60 and 15 respectively): a pre-workshop questionnaire (N = 61), interaction logs of ~1,300 prompt–response pairs from 60 participants planning lessons with a GPT-4.o-based chatbot over ~1.5 hours, and group-based SWOT reflections from 17 groups. Using the Intelligent Technological Pedagogical Content Knowledge (TPACK) (I-TPACK) framework for questionnaire and SWOT coding and a collaborative Problem Solving (CPS) framework for the interaction logs, the study finds that teachers most often articulate AI competence in technological (TK, n = 22) and technological-pedagogical (TPK, n = 16) terms, with Ethics mentioned explicitly only rarely (EK, n = 3). Yet in enacted practice teachers predominantly delegated task responsibility to the GenAI system (task delegation 54.8%; low interactional Learner Agency 68.8%), positioning GenAI as the primary generator of instructional content rather than negotiating or co-constructing outputs — revealing a gap between the knowledge teachers claim and the knowledge they enact.

Key Findings

Articulated vs. enacted knowledge gap. Teachers frame AI-related professional knowledge mainly in operational terms (understanding how GenAI works, prompting) and pedagogical support (lesson planning, efficiency), but their actual interaction shows delegation of substantial epistemic responsibility to the system — a divergence the study frames as knowing about AI integration versus enacting professional agency in AI-mediated collaboration.

Questionnaire patterns. Of 61 respondents, 49% had >20 years' teaching experience and 21% had 0–10 years; only 4.9% had never tried GenAI, with a substantial proportion reporting weekly use. In professional contexts, 32% reported using AI in teaching, 28% for lesson planning, and 16% for Assessment. Knowledge was mostly acquired incidentally (news, social media) rather than through structured training, and 22 respondents explicitly expressed uncertainty.

I-TPACK coding of articulated knowledge. Deductive qualitative content analysis mapped responses onto I-TPACK domains: TK (n = 22), TPK (n = 16), TCK (n = 9), and EK (n = 3), with 9 multi-domain and 22 uncertainty responses. Residual segments were analyzed inductively into Critical GenAI Literacy categories covering professional identity, relational positioning toward AI, and epistemic authority. A second researcher reviewed ~15% of coded material for interpretive agreement (no formal Cohen's Kappa).

Interactional positioning in logs. Using the CPS framework adapted from prior teacher–AI research, task delegation dominated at 54.8% of coded teacher turns, with negotiation/coordination 17.0%, team maintenance 14.0%, and shared-understanding prompts 10.2% (4.0% unclear). Overall 68.8% of prompts reflected low interactional agency and 27.2% high agency; sessions averaged ~4 prompts, and AI responses were substantially longer than teacher contributions, indicating teachers often treated GenAI as a content generator (e.g., "write a text about the Bronze Age") rather than a co-construction partner.

I-TPACK domain definitions and benefits/challenges. Coding was deductive-first with predefined categories: TK (how GenAI works, prompting), TPK (integration into lesson design/assessment), TCK (subject-specific use), Ethical Knowledge (bias, Trust, transparency, misuse, authorship), and Intelligent TPACK (explicit integration). When teachers described benefits, TPK was the most frequent domain (simplified lesson planning, fast Feedback, efficiency), while ethical concerns surfaced mostly in relation to challenges — Academic Integrity, Over-Reliance, and reliability of AI-generated content (e.g., difficulties distinguishing student work from AI text, cheating on home exams).

SWOT reflections. Post-workshop group SWOT responses aligned most strongly with TPK and EK (fewer TK/TCK). Teachers valued GenAI as a time saver, "sounding board," and support for differentiation and personalized learning, but flagged weaknesses and threats including unreliability, bias, cheating and academic-integrity risks, unreflective Cognitive Offloading, and a feared loss of "human nuance," alongside epistemological questions ("What is important knowledge?").

Limitations. The study is exploratory and qualitative; interaction and SWOT frequencies are descriptive rather than inferential, no generalizable effect sizes are reported, and the enacted-practice patterns observed in a workshop setting may not fully transfer to everyday classroom planning.

Implication. Professional Development must move beyond operational AI skills to build pedagogically meaningful, ethical knowledge that teachers actually enact, connecting to Professional Development, Technological Pedagogical Content Knowledge (TPACK), and AI Literacy and to teachers' evolving Teaching.

What this means for practice

• Instructors. Move from delegating to the model toward negotiating with it: task delegation dominated 54.8% of coded teacher turns and 68.8% of prompts showed low interactional Learner Agency, so state your own pedagogical intent and constraints before prompting rather than asking GenAI to generate a whole lesson element.
• Instructors. Use GenAI as a sounding board for differentiation and planning while reviewing every output — teachers valued it as a time saver but flagged inaccuracies, hallucinations, and the review time they demand.
• Faculty developers. Design professional learning that closes the articulated–enacted gap: teachers described competence in technological (n = 22) and technological-pedagogical (n = 16) terms while enacting delegation, so use practice-based, log-based reflection rather than knowledge transmission.
• Faculty developers. Make ethical reasoning explicit, since it surfaced in only n = 3 questionnaire responses, covering Academic Integrity, unreliability, bias, and over-reliance — concerns teachers raised mainly as threats rather than as knowledge they claimed.
• Administrators. Fund structured AI training rather than relying on incidental learning: most teachers reported acquiring AI knowledge from news and social media, and 32% reported using AI in teaching.

Limitations

• Two in-person workshops produced 75 K-12 teachers (60 and 15), recruited by open invitation through one Swedish university's outreach network with voluntary participation, so respondents self-selected.
• The workshop was a situated elicitation context, not a controlled intervention — about 1.5 hours of lesson planning, with sessions averaging ~4 prompts — so the delegation patterns observed may not transfer to everyday classroom planning.
• Frequencies from the questionnaire (N = 61), approximately 1,300 prompt–response pairs from 60 participants, and 17 group SWOT reflections are descriptive; the study reports no inferential statistics or effect sizes.
• Coding reliability is limited: a second researcher reviewed only about 15% of the coded material, and no formal inter-coder coefficient such as Cohen's Kappa was calculated.

Citation

Velander, J. (2026). Beyond operational skills: Teachers' AI knowledge and interactions with generative AI in lesson planning. Computers and Education Open, 100371. https://doi.org/10.1016/j.caeo.2026.100371