AI in Higher Education: A Guide for Teachers and Researchers

What does the French framework provide for AI in higher education?

The main official reference is the report "IA et enseignement supérieur : formation, structuration et appropriation par la société" (AI and higher education: training, structuring and societal appropriation), presented on 10 July 2025 to Minister Élisabeth Borne [4][32]. In the file reviewed, there is no binding text specific to AI in higher education. This report comes from a mission and recommendation process.

The dated milestones

• Mission: Patrick Hetzel launched a mission on AI in public higher education institutions, in line with his roadmap [5].
• Leadership: it was entrusted to Frédéric Pascal, director of the DATAIA institute, and François Taddei, director of the Learning Planet Institute [4][29].
• Consultation: an online questionnaire was open from 17 March to 17 April 2025 [17].
• Delivery: the report was presented on 10 July 2025, alongside a second report devoted to school education [4].
• Recommendation: it proposes a research-action programme on the appropriation and uses of AI, both in training and in administration [1].

The report starts from the observation that AI is a technological revolution set to profoundly transform society [1]. For its part, the ministry states that France has chosen to invest massively in AI and digital technology [3].

Limits of these milestones. These elements date from 2025. The file does not provide all the detailed conclusions of the report, and more recent texts (circulars, usage frameworks, institutional charters) may have been published since. Before any institutional decision, the full report should therefore be read and the ministry's publications after July 2025 should be checked on its websites.

How can AI be used for academic research and bibliography?

Generative AI intervenes in academic research at three documented levels: document searching, analysis and literature review, and formatting of references.

Document searching. A training course from the IA Sup programme describes how generative tools are transforming information-seeking practices, from traditional search engines to conversational assistants [25]. According to EBSCO, a database publisher, students can now begin by querying an AI assistant before consulting a database, which raises the question of the quality of the sources it relies on [36].

Analysis and literature review. A publisher's blog indicates that AI can help with data analysis, text exploration and literature review, by facilitating the synthesis of knowledge [28].

References. A university library guide indicates that AI can normalise references, apply a style (APA, MLA, Chicago) and format according to the requirements of a university or journal [40].

A cautious workflow for a literature review

1. Identify articles using a search tool. One guide suggests Semantic Scholar for initial articles [38].
2. Check each selected reference in the database or on the publisher's website. This step is a precaution on our part, justified by the issue of source reliability [36].
3. Synthesise only the most relevant articles, for example with NotebookLM, as this same guide suggests [38].
4. Review the synthesis against the original texts, because AI can also be used to fabricate false data [40].

To be taken with a grain of salt

The tool comparisons cited here come from commercial websites [33][38][39]. One of them, Sonix, presents itself as the best tool [39], and their rankings are not independent assessments. The gain from "several hours to a few minutes" announced for a bibliographic overview comes from one of these comparisons [33]: it is not an independent measurement.

How can students who already use AI be supported?

Students are already using AI, and supporting them is less about giving them access to the tools than about teaching them to evaluate their outputs. According to EBSCO, a database publisher and therefore a commercial player, they can now begin their searches by querying an AI assistant, even before consulting a database. The quality of the sources the assistant relies on then becomes a real question [36].

AI literacy rather than equipment

The integration of generative AI into higher education cannot be reduced to a matter of equipment, tool access or performance [11]. The authors propose moving from literacy to "AI literacy" and to "co-intelligence", that is, learning differently with generative AI [11]. The IA Sup framework goes in the same direction: it is a guide for integrating AI into education systems, aiming to prepare pupils and students [34].

Personalisation and lecture-based teaching

On the pedagogical level, two developments are emerging:

• Tailored pathways. AI can analyse students' skill levels, learning preferences and individual needs in order to adapt to them [23]. It can thus propose personalised learning pathways [20].
• Lecture-based teaching called into question. The automation of the transmission of theoretical knowledge is pushing higher education to move away from lecture-based teaching, which is weakened by AI's capabilities [16].

One competence among others

Knowing how to use an AI tool may be considered a skill expected by employers, but only one among others [21]. This source is a community article (LinuxFr), not a study. It nonetheless points out that mastering a tool replaces neither critical thinking nor disciplinary skills, which AI literacy specifically aims to strengthen [11].

What risks exist for scientific integrity and source reliability?

AI poses two distinct risks for research: a direct threat to scientific integrity, and uncertainty about the reliability of the sources that assistants draw on.

Scientific integrity: fabricated data and images

A university library guide warns that AI can be used to generate false data or doctored images [40]. For reviewers, laboratory heads and institutions, verifying the provenance of data therefore becomes a question in its own right.

Reliability of the sources used by assistants

Students can now begin their research by querying an AI assistant, even before consulting a database [36]. The quality of the sources on which the assistant relies then becomes a central issue. This analysis comes from a publisher of documentary databases, therefore from a commercial player [36]. It calls for systematically checking each cited reference against the primary source.

Bias, disinformation and ethical frameworks

The IA Sup framework includes the analysis of disinformation and bias in generative AI among its areas of work [15]. Ethical frameworks are considered necessary to guide responsible use in higher education [19]. The proposed charters aim in particular to prevent plagiarism and errors [15]. Other works formulate recommendations for responsible generative AI [7] and recall ethical principles of use [10]. The ethical risks of AI in higher education are explicitly discussed in them [9].

Detecting generated texts: a research topic

The detection of AI-generated texts is the subject of research work. One of these works aims to optimise detection protocols through comparative analysis and recommendations [12]. The file provides only an extract of it, with no quantified results: it does not make it possible to establish the performance of detectors. It would therefore be imprudent to base a disciplinary decision on this type of tool alone.

How can teachers be trained and the institution prepared?

Training teachers in AI requires drawing on resources that already exist and formalising an institutional strategy. The need is real: according to a survey cited by the consulting firm Headway Advisory (a commercial source), nearly 60% of teachers believe their institution is not ready to effectively train students in AI [22].

Existing resources

• IA Sup provides resources for teachers, support services, programme managers and institutions [15]. These include a charter and training, including an online course on the impact of generative AI on documentary research [15][25]. A separate framework aims to prepare students for AI [34].
• The IA4sup network has brought together teachers and university staff since 2022 to share their resources [30].
• Dedicated courses and teaching. Sciences Po has offered a course on AI to all first-year master's students since spring 2024 [31]. In an interview given to info.gouv.fr, Isabelle Ryl mentions programmes for specialists and non-specialists, "offering dual-competence opportunities" [2].

Formalising an institutional strategy

Integrating AI is presented as a strategic turning point, which calls for formalising the institution's approach rather than multiplying isolated initiatives [18]. The resources above can form its basis: a charter, teacher training, student pathways.

Another avenue is that of local AI, anchored in scientific corpora, teaching resources and institutional databases. Dataloma offers this type of deployment, with GDPR and sovereignty guarantees [26]. This offer comes from a commercial player: these guarantees are the provider's assertion and should be verified contractually.

How can Oupi and OUPI Academy fit into this approach?

Oupi addresses two needs identified above: document searching and training in uses. It does not replace them; it equips them.

For document searching and synthesis

Oupi queries academic sources and supports synthesis. This corresponds to a use described in the sources: generative AI tools are transforming information-seeking practices [25], and they can help with exploring texts and literature review, according to a commercial player [28]. Oupi's academic solutions operate in this area: identifying sources, then drawing a first synthesis from them.

This synthesis remains a starting point. AI can be used to generate false data or doctored images [40], and the reliability of the sources on which assistants rely is a central issue [36]. The researcher or teacher must therefore verify each cited reference, consult it in its original version and personally validate the conclusions before reusing them. Oupi does not exempt users from this verification and does not guarantee its outcome.

For training in uses

OUPI Academy trains in uses. The need is documented: nearly 60% of teachers believe their institution is not ready to effectively train students in AI, according to a consulting firm [22]. Moreover, the integration of generative AI is not reduced to equipment or access to tools [11], which argues for building skills rather than simply deploying technology. OUPI Academy's training courses can be aligned with usage charters, which aim in particular to prevent plagiarism and errors [15].