The research behind PROODOS

In brief

PROODOS is an online professional development programme on the pedagogical use of artificial intelligence, for practising primary and secondary teachers. It is structured around the UNESCO AI Competency Framework for Teachers, in fifteen modules followed by a closing Epilogue. Each module follows the same path: an introduction, the main study material, practical activities, an assessment, and a written reflection. Parts of every module adapt to the teacher's subject and grade level. The programme is available in Greek and English, and the Greek edition is adapted to the Greek school context rather than translated word for word.

PROODOS is the research platform of a doctoral dissertation at the International Hellenic University. Its research protocol has been approved by the university's Research Ethics Committee. There are no results yet. Section 6 sets out plainly what is not yet known.

1. Why preparing teachers for AI matters now

How AI is used in learning matters as much as whether it is used. In a field experiment with nearly a thousand high-school mathematics students, unrestricted access to a general-purpose AI chatbot raised grades while students had it. Once access was removed, however, those students did 17% worse than students who never had access. A version designed to safeguard learning largely avoided this harm (Bastani et al., 2025). The same technology can support learning or undermine it, and in schools, how it is used depends largely on teachers.

Research on teacher professional development points the same way. A systematic review of 43 studies on what primary and secondary teachers need in order to integrate AI concludes that technical training alone is not sufficient. Pedagogical knowledge, positive attitudes, organisational support and continuous training are also needed (Aravantinos et al., 2026). The field is also young. A systematic review of AI professional development programmes found only 14 publications that met its criteria, and several of them were conference papers or project reports (Dogan et al., 2025).

Ethics tends to be the part that is left out. An analysis of 98 recorded AI lessons in schools in central China found that AI ethics featured in only 5.1% of the instruction observed (Wu et al., 2024). The study comes from one national context, but it describes a risk that the programme is designed against: ethics is addressed at each of the three levels of the programme, not in a single module.

Working with AI can change how teachers see their own role. In interviews with thirty teachers, many first feared that AI would erode their role. Over time they came to see it as something that could support their teaching rather than replace it, and most described a shift from traditional instruction toward facilitation (Isma'il & Ibrahim, 2025). This is why the programme treats professional identity, and not only skills, as part of its subject matter.

2. The reference framework, and how fully the programme covers it

The programme is structured around the UNESCO AI Competency Framework for Teachers (UNESCO, 2024). The framework describes what teachers should be able to do with AI across five aspects (a human-centred mindset, the ethics of AI, AI foundations and applications, AI pedagogy, and AI for professional development) and three progression levels (Acquire, Deepen, Create). It was published in September 2024, and its implementation is still at an early stage.

Coverage. The programme's internal content validation matrix breaks the framework down into 170 indicators. In the May 2026 assessment, 163 of them (95.9%) were rated as strongly covered. The remaining seven are documented one by one: some can only be demonstrated through real classroom use, and the others reflect deliberate design choices. This is the research team's own assessment. It has not yet been checked independently.

The European policy direction is consistent with this approach. In its 2026 conclusions on teachers in the era of artificial intelligence, the Council of the European Union states that AI tools must support teachers rather than replace or isolate them. It also states that professional judgement, pedagogical autonomy and responsibility remain essential, and it calls for AI literacy to be integrated into teachers' continuous professional development (Council of the European Union, 2026).

How the programme compares. A systematic literature search carried out in August 2026 (41 recorded searches in six languages, including grey literature) found no programme or platform by another team that covers the whole framework, all five aspects at all three levels. The closest external comparison is a 14-week graduate course for in-service science teachers that covers all five aspects, without documented progression levels (Alexandron et al., 2026). In Greece, an analysis of 15 university continuing-education programmes on AI for teachers found that they concentrate mainly on foundational skills (Τσιούκας & Κώστας, 2026). On this basis we describe PROODOS as one of the first documented, complete implementations of the framework at platform level, and never as "the first". The claim rests on that search alone. PROODOS grew out of an earlier programme by the same research team, which also covered the full framework (Kokkonis & Dourvas, 2025); that work is its predecessor, not evidence about what other teams have done.

3. Design principles and the research behind them

3.1 Reflection as professional practice

The programme's central position is that teachers need professional judgement about when, how and why to use AI more than they need techniques for particular tools, which will change. Every module ends with a written reflection on the teacher's own practice.

This position is developed in a published framework by this project's own research team, which reframes prompt writing as reflective professional practice rather than a technical skill (Dourvas et al., 2026). That paper is a conceptual framework. Its authors state that it offers hypotheses that still need empirical testing, and this programme is one place where they will be examined. Two further strands of research shape how reflection is used. The first is that reflective writing often stays descriptive, and that depth of reflection has to be explicitly supported (Moon, 2007). The second comes from an eight-month programme with 50 primary teachers in Taiwan: after repeated cycles of planning, classroom use and reflection, with community support, the teachers reported higher confidence in teaching with AI tools (Yang et al., 2024). That was a self-reported measure in a single context, but it is consistent with building the programme as a sustained sequence rather than a single workshop.

3.2 Keeping the teacher in charge

The programme teaches human oversight of AI and applies the same rule to itself.

• The use of AI is disclosed to each teacher, and must be acknowledged, before any AI feature is used. That acknowledgement can be withdrawn at any time from the privacy dashboard.
• The four AI analyses of a teacher's reflections (the feedback, the mapping of pedagogical tensions, the summary of the teacher's development across modules, and the synthesis of peer reflections) each carry a button for disputing the output.
• No AI is involved in scoring, in deciding whether a teacher can move on, or in issuing the certificate.

This follows the human-centred AI position that reliable, safe and trustworthy systems combine high levels of automation with high levels of human control (Shneiderman, 2020). It also follows research on trust in automation, which treats the aim as calibrated trust, trust that matches what a system can actually do, rather than as maximum trust (Lee & See, 2004). Experimental evidence shows that over-reliance on AI suggestions can be reduced by designs that make people think before accepting them, although people tend to like those designs less (Buçinca et al., 2021). The programme's reflection steps follow the same logic.

3.3 Transparency toward the teacher

Every AI-generated output on the platform, from the feedback on a reflection to the answers of the help assistant, comes with a short panel headed "How was this generated?". The panel states which model produced the output, what input it was given, the logic it followed and what the output is for. The AI Impact Assessment, linked at the foot of this page, describes each AI feature in full.

Transparency is also part of what the programme teaches. People tend to overestimate how accurate a large language model's answers are, and longer explanations increase their confidence even when the answer is no more accurate (Steyvers et al., 2025). In a study with 41 in-service chemistry teachers, explanations of AI recommendations framed in the teachers' own subject were easier to understand than explanations based on data, and understandability was linked to trust and acceptance (Feldman-Maggor et al., 2025). Both findings inform how the programme asks teachers to judge AI output: by whether its reasoning holds up in their own subject, not by how fluent or confident it sounds.

3.4 Personalisation by subject

The teacher's subject and grade level shape what they see. Each module has an introduction written for the subject. The main text contains subject-specific examples and boxes. Practical activities are chosen by subject and grade, and the reflection question differs by subject. The AI feedback on reflections is given the teacher's profile as context. The optional practice environment offers scenarios grouped into eight subject clusters.

Two limits are deliberate. First, personalisation changes the examples and the framing, not what the teacher is asked to do, so that the work of teachers with different profiles stays comparable. Second, the synthesis of peer reflections deliberately draws on teachers of other subjects, as a counterweight to an echo chamber.

The reason for this design is the review finding, already cited, that AI professional development has to be grounded in pedagogy and context rather than in technical training alone (Aravantinos et al., 2026). It also draws on work showing that teachers build a better understanding of AI, including its ethical and context-specific uses, when they construct it together rather than receive it (Dilek et al., 2025). The literature search described in section 2 found no programme that personalises AI professional development by teacher subject across several layers in this way.

3.5 Keeping content current

Content about AI goes out of date quickly. The research team reviews module content as new research, tools and policy appear, and every change is made by a person. Nothing on the platform is updated automatically. Teachers can also report content they believe is out of date, from any module page.

3.6 Cost as a design variable

Taking part in the programme is free of charge. Behind that, what the programme costs to run is treated as a design variable, not an afterthought. The project sets an explicit goal for the AI service calls behind its features: under one euro per teacher, for the whole fifteen-module programme. The goal covers only these AI calls. Hosting, email and the research team's own time are not part of it.

The AI calls are those behind the feedback on a reflection, the mapping of pedagogical tensions, the summary of a teacher's development across modules, the synthesis of peer reflections, the practice environment, the help assistant and the facilitator replies in the Module 10 workshop, together with the search that finds the relevant research for the feedback. The platform records the cost of each of these calls, by teacher, by module and by feature.

Our estimate for a teacher who completes the whole programme, built from the measured size of real calls, is below that goal. It is an estimate, not a measurement: the actual figure per teacher will come from the teachers who take the programme in full.

We treat cost as part of the design because we want a programme that others could realistically run after this research ends.

4. How the programme is evaluated

The research asks two things. First, it documents what was learned by building a complete implementation of the framework, so that others can reuse that design knowledge. Second, it gathers evidence on how teachers' judgement in using AI develops during the programme.

The evaluation follows a design-based research approach in three consecutive cycles.

Cycle 1: appraisal by experts. Design reviewers (academic researchers in education or AI, and educational technology professionals) take a guided tour of a trial account and complete a structured questionnaire based on established design-quality criteria: relevance, consistency, practicality and expected effectiveness (Nieveen, 1999). Education advisors use the programme for real, in modules of their own choosing. The minimum target is 20 people in each role.

Cycle 2: checking the research tools (exploratory). Reflections written by the advisors are coded by human coders, and the agreement between the human coding and the platform's automatic analysis is examined statistically. The main contribution of the research does not depend on this cycle.

Cycle 3: field pilot. Practising teachers take the programme as normal. They complete a validated questionnaire on teachers' AI literacy (Ning et al., 2025) three times: on registration, after module 5 and after module 15. The Greek version of the questionnaire was prepared through an established cross-cultural adaptation procedure with independent back-translation (Beaton et al., 2000), and the case for its validity is built in the argument-based way (Kane, 1992). There is no control group from which the programme is withheld. The analysis looks at change within the same teachers and at the relationship between depth of engagement and change. A later-start comparison group may be added if registration numbers allow, in which case those teachers simply start later and lose nothing. Each research question has a minimum analysis declared in advance for each possible sample size. No figure is published from a group of fewer than five participants.

5. Ethics, consent and data

This platform's research protocol has been approved by the IHU/ΔΙΠΑΕ Research Ethics Committee (ΕΗΔΕ), Αρ. Πρωτ. 229/2026 (Session 35, 14 September 2026).

Taking part in the research is voluntary, separate from using the programme, and can be withdrawn at any time from the privacy dashboard. Teachers who decline keep full access to the programme and can still receive the Certificate of Attendance, which is issued on completion of all fifteen modules. The certificate records participation in the programme. It does not show scores and does not depend on taking part in the research.

The Privacy Policy sets out what is collected, why, for how long, and how it is transformed at the end of the retention period, including what happens to the text teachers write when it is sent to an AI service to produce feedback. The AI Impact Assessment describes each AI feature. This page does not interpret the law and is not legal advice.

6. What we do not know yet

• At the time of writing (September 2026), no teacher has completed the programme as a research participant. The modules have been tested extensively with trial accounts, which is not the same thing.
• There are no results. Nothing on this page should be read as an outcome.
• The programme length of 75 hours is a design estimate, not a measurement. It was built from a time budget for each activity in a module and checked against the measured length of the module texts. For comparison, in a review of research on professional development and student achievement, the nine studies that met What Works Clearinghouse evidence standards involved an average of almost 49 contact hours; the review's authors note that the evidence does not allow conclusions about how much professional development is needed (Yoon et al., 2007). In European credit terms it corresponds to 2.5 to 3 ECTS credits (European Commission, 2015). How long the programme really takes a practising teacher is one of the things the pilot will show.
• The automatic analysis of reflections is a prototype, not a validated instrument. It serves its teacher-facing purpose. How well it agrees with human reading of real reflections has not yet been established.
• The Greek scoring in the practice environment is a first version. It has not yet been used with real Greek-language submissions.
• The automatic mechanism intended to help spot out-of-date content is being redesigned and is not relied on today. Whether its signal reflects genuinely out-of-date content has not been tested against human judgement.
• There is no control group, so change cannot be attributed to the programme with certainty.
• The certificate carries no official credit or recognition at present. It documents completion of a programme developed within this doctoral research.
• The AI cost per teacher has not yet been measured. Section 3.6 gives an estimate from the size of real calls; the pilot will give the measured figure.

7. Who we are

PROODOS is the empirical instrument of a doctoral dissertation at the International Hellenic University (IHU), Thessaloniki, under the supervision of Assoc. Prof. Georgios Kokkonis, Department of Information and Electronic Engineering. The research is carried out by John Dourvas, doctoral researcher.

Contact: idourvas@ihu.gr

Further reading on this site: Privacy Policy · AI Impact Assessment (EU AI Act transparency notice) · the key studies listed at the start of each module