Can AI Help Students Think for Themselves?

Findings from a Year 12 Psychology classroom

Read the original research paper in the International Coalition of Girls' Schools Research Library.

Does generative AI simply encourage students to outsource their thinking for quick, easy answers? Or is it possible to design AI that fosters independent, self-regulated study habits? This core tension was on our minds when we teamed up to bring Bloom into Jayne's classroom at St Hilda's School, with the support of the International Coalition of Girls' Schools.

The research project was conducted with a class of thirteen Year 12 girls (aged 17-18) enrolled in Psychology. We trained an AI tutor, Bloom, on specific syllabus content, key terms, and cognitive verbs aligned with the external examination requirements. Bloom was already designed around pedagogical principles intended to encourage independent thinking and metacognition, as well as avoiding doing the work for students. One aim of the study was to explore whether these design choices would translate into more self-regulated learning behaviours over time. Students were given access to Bloom for a semester (June–November) and encouraged to integrate it into their study routines in a non-prescriptive way. Throughout the semester, data was collected through surveys, focus groups, and user analytics through the Bloom dashboard.

We discovered four interconnected themes showing how students experienced and engaged with an AI tutor. The headline change was that students learned to use AI differently over time: shifting their use from seeking information and generating study materials towards actively directing the AI to test, challenge and deepen their thinking.

1. Receiving immediate, explanatory feedback strengthens confidence and reduces procrastination.

Students consistently reported that Bloom’s instant, syllabus-specific feedback boosted their confidence and reduced anxiety across all phases of study. What began as frustration with not receiving direct answers immediately, evolved into appreciation for what students called the “Socratic push.” They recognised that this questioning approach encouraged deeper thinking, reinforced learning and reduced procrastination. Student 1 stated, “It doesn’t give me the answer right away. At the beginning it was annoying, but now I realise I’m learning so much more.” By the end of the study, all 13 participants agreed that Bloom motivated them to engage in study more readily.

2. The availability of a Socratic tutor builds independence while complementing teacher presence.

Students emphasised that while their teacher remained irreplaceable, Bloom effectively filled critical gaps when the teacher support was not available. Prior to Bloom's introduction, 8 of the 13 students reported studying fewer than two hours per week and identified confusion, distraction and lack of immediate help as barriers. By the end of the project, 10 of 13 described themselves as more motivated to study, while 9 of 13 reported studying at least four hours per week. While the approach of the external examinations is an important factor to consider, students consistently described Bloom's immediate availability as helping them persist when they would previously have stopped. As Student 8 reflected, “I used to stop revising when I got confused, but now I can just ask Bloom.”

3. Using Bloom AI to structure study materials and prompt thinking encourages self-regulation strategies.

Students’ engagement with Bloom evolved significantly over the course of the project. Bloom platform analytics showed that initially, most girls used the AI tutor for basic study tasks such as creating palm cards, summarising content, and clarifying definitions. Over time, and with some direction, their interactions became more purposeful and complex. They began refining their prompts to request harder questions, made deeper connections between theory and mandatory readings and explored real world applications of concepts. By the post-mock exam phase, 9 of the 13 students were intentionally using Bloom to plan their revision and test their understanding. Student 3 said, “I started giving Bloom more detail in my prompts to get deeper answers” while Student 2 reflected, “I enjoyed figuring it out myself, it made me question and understand better.” Importantly, greater use did not necessarily mean greater dependence. One student deliberately limited her use of Bloom because she did not want to become over-reliant on it or lose the ability to work through content herself.

4. Developing prompt literacy through Socratic dialogue expands students’ sense of agency in learning.

In the lead up to their external exam, Bloom analytics showed how students began deliberately experimenting with how they interacted with Bloom, using increasingly targeted prompts, requesting specific forms of feedback, and asking the AI to challenge their thinking. Rather than simply asking AI for information, they were learning how to direct it. Bloom evolved from being viewed as a study tool to being a cognitive partner, one that enabled agency, autonomy and intellectual risk taking.

---

A further lesson from the project was the value of co-design. Rather than receiving a finished AI tool and implementing it with students, Jayne and her students were able to shape how Bloom developed throughout the project. The ability to provide this feedback, identify areas for improvement and suggest changes meant that the agent could be refined in response to what was actually happening in the classroom. Knowing there was a team listening, supporting and responding to all messages gave the teacher and students confidence to experiment, and a sense of ownership over the development of the agent and its functionality.

This was a small action-research project involving thirteen students, without a control group, and it was not designed to establish that Bloom caused improvements in examination performance. What changed most clearly was how students described and demonstrated their approach to studying: their confidence, help-seeking, questioning, use of feedback, metacognitive awareness and ability to direct their own learning.

These findings are now informing further trials of Bloom beyond Psychology, including with Year 11 and 12 students in French Extension, Japanese, Accounting and Literature. Early feedback from teachers and students has been positive, providing an opportunity to explore whether the same principles can support self-regulated learning, metacognition and student agency across different subjects and learning contexts.

So, does generative AI inevitably encourage students to outsource their thinking? Our experience suggests that how the AI is designed matters. Bloom was deliberately built to resist simply doing the work for students, instead using questioning, feedback and scaffolding to keep them actively involved in the learning process.

Acknowledgements

Our thanks go to the thirteen Year 12 Psychology students at St Hilda's School, whose curiosity, openness and thoughtful feedback made this project possible. We also thank the International Coalition of Girls' Schools (ICGS), Jayne's GARC Research Advisor Núria Tapias Nadales, school-based mentor Majda Benzenati, and St Hilda's School Principal Virginia Warner for their support throughout the project.