While 98% of European students identify at least one risk associated with AI, and 79% fear a decline in their academic skills, 61% use it regularly due to a lack of alternatives that are perceived as effective and accessible. This “resigned” adoption reflects an ecosystem where practicality trumps reliability.
Launched at the end of November 2022, ChatGPT quickly became a very practical alternative for students around the world. For this population, whose academic workload has increased considerably, the “practical” nature of access to information is a determining factor in its use. How has AI changed the information landscape for students in just a few years? The answers to this question can be found in a survey conducted by the European university alliance CIVICA among students from its 10 member institutions, including Sciences Po, summarized by Cécile Touitou, head of the Prospective Mission at the Sciences Po Library. The survey was distributed to 72,000 students via a conversational interface on smartphones in November 2025 and received 2,294 responses from all Alliance universities (51% master's, 43% bachelor's, 4% doctoral/post-doctoral, 2% other).
In a landscape saturated with information, where information overload and the proliferation of alternative content blur the lines, CIVICA students in the humanities and social sciences surveyed in a major transnational survey conducted at the end of 2025 are developing often fragile strategies to conduct their research. Between the quest for efficiency and time pressure—eroded by the omnipresence of screens—they favor “beneficial” content over reliable sources. They tell us that they make little use of primary documents, which can be demanding, time-consuming, and difficult to read, and now turn to AI to obtain summaries.
Students are divided: they trust university libraries but find some services inadequate, follow their teachers' recommendations above all else, and lack the skills to use information and generative AI effectively. Aware of the risks of errors and bias, they feel helpless and are asking for more training and support.
Students' new information practices
Information search: time-consuming, stressful, and unreliable
While 41% of the students surveyed spend 2 to 5 hours per week searching for information, an equally large proportion (41%) spend 6 hours or more. Twenty-four percent of doctoral students even report spending more than 10 hours on it. This activity constitutes an important foundation for acquiring or validating knowledge outside of class.
Fifty-five percent of respondents cite stress generated by the abundance of sources or answers in a search, with the proportion rising to 62% among doctoral students and even 64% among law students.
On the thorny issue of fake news, students generally feel capable of spotting it, but their confidence is moderate rather than absolute. Confidence is lowest among master's students (only 6 out of 10 consider themselves confident), while confidence is higher among doctoral students, with 8 out of 10 expressing confidence. The lack of access to reliable information is widely perceived as a direct academic risk by 7 out of 10 students, regardless of their level.
A new hierarchy: Google, ChatGPT... then, far behind, libraries
The internet is the main point of entry (71%), particularly via general search engines such as Google: 67% of master's students use it, compared to 49% of doctoral students. On the other hand, academic search engines such as Google Scholar are used more by doctoral students: 76% of them prefer them.
Respondents reveal that generative AI has already become commonplace (61% reported use), ranking second behind reading lists provided by teachers (59%) and far ahead of library catalogs (38%). Similarly, 35% of doctoral students report using AI to find information, compared to 63% of master's students; 70% of economics students compared to 32% of history students. This wide dispersion also reveals a wide variety of uses for AI depending on the level and discipline.
The relationship with books (printed and digital)
The digital world is fragmenting students' time, invading their downtime, and forcing them to make choices based on academic profitability.
In this context, how can students find the time to “read” and do research? When asked, “My academic workload is so heavy and deadlines so tight that I don't have enough time for in-depth and critical reading,” 73% of bachelor's students agreed, along with 76% of master's students and 69% of doctoral students. Seven out of ten master's students even stated, “Most of the time, I read summaries of books/articles rather than entire books/articles” (4.5 out of 10 for doctoral students)!
The use of printed books is generally low: 73% of respondents say they rarely or never use them, compared to only 14% who use them often or all the time. Doctoral students are the most likely to read them (42%), with 26% using them regularly, but 58% say they never read them.
Printed books have become a niche resource, mainly used by doctoral students in work requiring in-depth literature or in fields where digital technology is less developed.
65% of students say they prefer digital media to paper when they have a choice between the two versions, especially doctoral students (75%), more than bachelor's students (63%). The reported use of ebooks is on average 5 points higher than that of print, yet only 33% of respondents report using them; 22% use them all the time or frequently, 12% use them sometimes, but 67% rarely or never!
When asked which resources they find most useful or relevant to their work, they cite web resources first (56%), followed by books recommended by teachers (36%) and AI (35%). The rest of the suggestions received less than 20% of responses.
Generative AI: an omnipresent but controversial assistant
Widespread adoption but mixed confidence
AI has become an essential study tool for students, but this widespread adoption is not accompanied by equivalent confidence in the content generated or clear institutional recommendations. Students thus find themselves alone in managing risks that are perceived as high, with uneven support depending on the institution and teacher.
Thus, while 61% of students use generative AI (GIA), only 35% consider it useful and relevant for their studies. Among AI users, the most common applications are understanding complex content (73%), summarization (72%), and idea development (61%), which positions AI primarily as a tool for comprehension and productivity rather than as a research or methodology tool.
AI: perceived reliability inversely proportional to academic stakes
Students consider generative AI to be relatively reliable for linguistic and comprehension tasks—translation (86%), summarization (71%), understanding complex concepts (69%)—but much less so for research activities: searching for scientific articles (38%) and suggesting sources (32%), precisely where errors can compromise academic quality.
Bachelor's students are the most confident and heavy users in almost all categories. Master's students fall somewhere in between, standing out for their high use of AI for formatting bibliographies. Doctoral students, who are more critical, use AI in a more targeted way for searching for articles and less for comprehension or idea generation.
Embracing AI while fearing its effects
Nearly all students identify at least one risk, with inaccurate information (70%) and unintentional plagiarism (65%) at the top of the list, followed by concerns about uniformity of ideas (61%) and loss of credibility (58%).
Concerns extend beyond integrity to the erosion of long-term skills, indicating anxiety about dependence on this tool.
An explicit request for support
Nearly half say that teachers neither encourage nor discourage the use of AI (47%), while 66% say their institution has provided a clear framework, leaving a significant minority to navigate the use of AI without stable reference points. Three-quarters of them disclose their use of AI to their teachers at least sometimes (75%), but only 31% do so frequently, and 52% would like to receive training on the responsible use of AI in an academic setting.
Uneven information skills
About 73% feel comfortable finding information, and 71% feel comfortable evaluating its reliability, but this figure drops to 58% for managing bibliographic references and 54% for using academic databases.
When it comes to verifying the reliability of sources, most students say they check information frequently. Bachelor's students are the least rigorous in this regard, followed by master's students and doctoral students, with 43% of the latter saying they systematically check their sources (and 35% saying they often do so).
Practices in case of difficulty
The most common reflexes reported by students are turning to teachers (33% overall, but 45% for doctoral students) and generative AI (29% overall, but 15% for doctoral students and 30% for bachelor's students), followed by Internet searches (13%).
In just three years, traditional internet searches have been overtaken by the arrival of GAI, which students say they use “most often” much more (29% vs. 13%).
Appreciated but underused: the paradox of university libraries
The greatest difficulty identified concerns the availability and accessibility of sources (73%), due to the difficulty of accessing relevant sources (58%). This difficulty of access is compounded by information overload: 66% report research difficulties, particularly an excessive number of sources (55%).
Only 46% report having received training, while 37% have not received training but would like to, indicating a strong demand for more practical, skills-based support.
Libraries are widely used (81% of students), with usage increasing according to level of study, but this usage remains polarized between intensive users (51%) and persistent non-users (25%). Non-use is mainly due to practical obstacles: complexity of access (38%), lack of awareness of needs (37%), and lack of awareness of services (34%).
The library remains primarily valued as a quiet workspace (75%), far ahead of access to digital resources (48%) or borrowing books (39%).
Thus, while 98% of students identify at least one risk associated with AI and 79% fear a decline in their academic skills, 61% use it regularly due to the lack of alternatives perceived as effective and accessible. This “resigned” adoption reflects an ecosystem where practicality outweighs reliability.
Libraries have a major strategic lever at their disposal: improving resource reporting, making access paths more intuitive, and developing targeted training on the responsible use of AI and academic research methods.