AI tutoring: Bridging the educational disadvantage gap

MARK LESWELL, RESEARCH LEAD, SWALE ACADEMIES TRUST, UK

This article examines how AI tutoring could address the educational attainment gap for socio-economically disadvantaged students. Recent studies suggest that AI tutoring can deliver significant learning gains, while overcoming the scaling limitations of traditional one-to-one tutoring. While implementation challenges exist, the technology shows promise as a cost-effective tool for reducing educational inequalities.

Context

The educational attainment gap between disadvantaged students and their more affluent peers remains one of education’s most persistent challenges. In England, disadvantaged students at Key Stage 4 have a Progress 8 score 0.75 of a grade lower than their wealthier counterparts (DfEDepartment for Education - a ministerial department responsible for children’s services and education in England, 2023a), with the gap widening to around one grade for high-achieving learners (Holt-White and Cullinane, 2023). This disparity has far-reaching implications, as academic research consistently shows that socio-economically disadvantaged students not only perform less well in educational settings but also face barriers to future educational opportunities (Hanushek et al., 2019; Lessof et al., 2018).

The factors contributing to this gap are complex and interconnected. Students from disadvantaged backgrounds often face multiple challenges, including health issues, learning difficulties and a lack of resources at home (Rutkowski et al., 2018). These individual challenges are compounded by systemic issues, including high teacher turnover in deprived areas (Allen et al., 2018) and disparities in OfstedThe Office for Standards in Education, Children’s Services and Skills – a non-ministerial department responsible for inspecting and regulating services that care for children and young people, and services providing education and skills gradings based on deprivation demographics (Thompson, 2022). The cumulative effect of these barriers creates a gap that is harmful to society as a whole (Gorard et al., 2023).

Evidence base for tutoring

Traditional one-to-one tutoring has proven to be a highly effective intervention for improving student outcomes. Bloom’s seminal ‘2 sigma problem’ research demonstrated that students receiving one-to-one tutoring performed two standard deviations better (Bloom, 1984). More recent evidence from the Education Endowment Foundation reinforces this finding, demonstrating that students who receive one-to-one or small-group tutoring make five months’ additional progress in their studies (EEF, 2021).

However, traditional tutoring faces significant scaling challenges. The resource-intensive nature of one-to-one tutoring, combined with limited availability of qualified tutors, particularly in disadvantaged areas, makes widespread implementation challenging. These limitations have led to growing interest in AI tutoring as a potential solution.

How AI tutoring works

Generative AI, particularly large language models (LLMs), offers a compelling solution to traditional tutoring constraints. These AI systems, trained on extensive text data, can generate human-like contextual responses to diverse queries, enabling personalised, scalable educational support (Mollick and Mollick, 2023). 

The integration of evidence-based pedagogical approaches within AI tutoring systems presents significant advantages. Through active learning techniques, retrieval practice and sequential knowledge-building, these systems can foster deeper comprehension and improved retention. For instance, AI tutors can be prompted to facilitate active engagement through strategic questioning and can encourage learners to explain their reasoning processes. The adaptive nature of AI tutoring systems allows for bespoke learning experiences, with customised explanations and exercises tailored to individual performance (Mollick and Mollick, 2023). This dynamic approach enables continuous assessment and adjustment of support levels, ensuring that learners receive precise guidance aligned with their educational needs (see an example prompt in Figure 1).

Act as an expert, encouraging tutor designed to:

– Assess current knowledge level and ongoing understanding

– Deliver tailored explanations with diverse examples

– Guide learning through open-ended questioning

– Adjust support based on student performance

– Encourage active knowledge construction

– Prompt students to articulate concepts in their own words

– Ask metacognitive questions throughout the discussion

Rule: Avoid providing the answer; your role is to guide students in their learning.

Begin by asking for the learning intentions, the current course and then assess the student’s current knowledge. 

Figure 1: Example AI tutor prompt for a frontier model (e.g. ChatGPT o1, Claude Sonnet 3.5)

This structured approach allows AI tutors to facilitate deeper understanding through:

• active learning strategies that engage students in constructing knowledge
• retrieval practice through regular prompting to recall and apply concepts
• continuous assessment and adaptation to individual student needs
• personalised feedback based on student performance.

Emerging evidence for AI tutoring 

Emerging research highlights the significant potential of AI-powered tutoring to enhance student learning. A pilot study in Nigeria examined the effects of an AI-driven after-school programme, with preliminary findings revealing substantial gains in English language proficiency, AI literacy and digital skills. Students who participated in the programme outperformed their peers on end-of-year curricular exams, with learning gains equivalent to 0.3 standard deviations in just six weeks – a pace comparable to nearly two years of traditional learning and greater in impact than 80 per cent of education interventions in developing countries. Notably, the study, soon to be published, also found that the programme had a particularly strong impact on girls, who were initially struggling academically (De Simone et al., 2025), and diminishing returns were not observed within the data, meaning that with further attendance, student performance continued to increase.

Complementing these preliminary findings, Wang et al.’s (2024) Tutor CoPilot study, involving 900 tutors and 1,800 K-12 students from historically under-served communities, found that students working with AI-supported tutors were four percentage points more likely to master topics. Notably, students of lower-rated tutors experienced the greatest benefit, improving mastery by nine percentage points relative to the control group. The study estimated an annual cost of $20 per tutor based on usage patterns, offering a cost-effective, scalable approach to tutoring.

Challenges 

While AI tutoring shows promise, several significant challenges require careful consideration. The Department for Education’s recent call for evidence identified risks including potential over-reliance on AI and concerns about output quality (DfE, 2023b). UNESCO (2023) has highlighted the need for proper guidance and regulations, particularly regarding data privacy and pedagogical quality.

Implementation challenges extend beyond technical considerations to include:

• ensuring equitable access to technology
• ensuring that evidence-informed pedagogy is used
• maintaining appropriate oversight of AI–student interactions
• supporting teachers in effective integration
• monitoring and evaluating impact.

Conclusion

AI tutoring presents a promising avenue for addressing the educational disadvantage gap, offering the benefits of personalised support at scale. While early evidence from studies suggests significant positive impacts on student outcomes, further studies are needed to assess the long-term impact of these interventions, and successful implementation requires careful consideration of both opportunities and challenges. If thoughtfully integrated into practices, AI tutoring could help to narrow the persistent achievement gap between disadvantaged students and their peers, making high-quality educational support more accessible to those who need it most.

The examples of AI use and specific tools in this article are for context only. They do not imply endorsement or recommendation of any particular tool or approach by the Department for EducationThe ministerial department responsible for children’s services and education in England or the Chartered College of Teaching and any views stated are those of the individual. Any use of AI also needs to be carefully planned, and what is appropriate in one setting may not be elsewhere. You should always follow the DfE’s Generative AI In Education policy position and product safety expectations in addition to aligning any AI use with the DfE’s latest Keeping Children Safe in Education guidance. You can also find teacher and leader toolkits on gov.uk .