Technology-Supported Formative Assessment in Higher Education: AI, Learning Analytics and the Future of Assessment for Learning

Synthesis: Bulut (2026) maps twenty-six years of research on technology-supported formative assessment (TSFA) in higher education, drawing on 407 studies published between 2000 and August 2026. The review's central claim is a shift in what technology is for: early work treated digital tools as delivery mechanisms for existing assessment practice, whereas the recent literature increasingly positions analytics and AI as constitutive parts of a formative cycle of evidence, feedback, reflection, and pedagogical action. Bulut names this configuration a technology-mediated formative ecosystem, joining technological capacity to pedagogical regulation, learner agency, and human judgment. The findings are distributional rather than experimental: bibliometric mapping shows acceleration after 2020 and convergence around formative assessment, Feedback, students, and AI, while qualitative synthesis and a SWOT reading identify theoretical fragmentation, thin longitudinal evidence, ethical exposure, and a pronounced geographic imbalance as the field's constraints. The practical argument is that institutions should pursue human-centered AI integration, transparent governance, and equitable access if technology is to support rather than displace formative pedagogical judgment.

Key Findings

• The review's corpus is 407 studies drawn from Web of Science and Scopus, published between 2000 and August 2026, screened from 1,045 initial records.
• Annual output accelerated markedly after 2020, which Bulut reads as consolidation of TSFA as a recognized field in higher education.
• Thematic clustering concentrates influence in human-centered and continuous assessment, with formative assessment, students, and AI as central nodes, while feedback, education, and performance appear comparatively underdeveloped as themes.
• Research collaboration clusters around the United States, United Kingdom, and China, with Canada and parts of Europe and the Asia-Pacific also connected, and limited connectivity across much of Africa and Latin America.
• The SWOT reading treats strengths (integration of feedback, analytics, AI, and self-regulation) against weaknesses (theoretical fragmentation, methodological imbalance, scarce longitudinal evidence) and threats (algorithmic bias, privacy, unequal access, dependence on automated judgment).
• Bulut frames the SWOT as an interpretive device rather than an independent analytical technique, so its entries synthesize the bibliometric and thematic results rather than adding a separate analysis.

What the bibliometric record shows

The corpus was built from two databases, Web of Science and Scopus, which together returned 1,045 records, 336 from Web of Science and 709 from Scopus. Screening removed 81 duplicates, then 496 records on document-type criteria (101 from Web of Science and 395 from Scopus) and a further 61 on language criteria (18 and 43 respectively), leaving 407 studies for mapping and synthesis. The steepest growth in annual production comes after 2020, and Bulut interprets that acceleration as more than volume: it marks a conceptual transition in which formative assessment moves "beyond the digital replication of conventional practices" toward an adaptive, data-informed, learner-centered paradigm. The argument matters because it changes the unit of analysis. If the interesting question were whether a digital quiz works, the literature would be a sequence of intervention studies. Because the review frames technology as a mediating mechanism for continuous feedback and pedagogical decision-making, it reads the field as accumulating a new assessment infrastructure, and then asks what that infrastructure lacks.

Where the literature is strong, and where it is thin

The thematic clustering is the part of the review with the most consequence for practice. Human-centered and continuous assessment occupies a highly developed and influential position, and the strong association of AI with performance, learning systems, deep learning, and higher education signals a research orientation toward adaptive and data-informed environments. Bulut reads this as a genuine strength, since technological mediation has moved from a peripheral support function to a constitutive element of assessment practice, a positioning he argues is consistent with sociocultural, Constructivism, and assessment-for-learning traditions that treat Assessment as iterative rather than certifying. The counterweight is that feedback, education, and performance are comparatively underdeveloped as themes despite their pedagogical centrality, and algorithm-related themes sit at the periphery of the knowledge structure. The imbalance is worth stating plainly: the field is developing its instruments faster than its account of what good feedback is, which is why the review calls for an integrative framework tying technical advance to feedback processes and learner agency. The same configuration supports the paper's emphasis on preserving human judgment, since self-regulated learning depends on learners interpreting formative information rather than receiving verdicts.

The equity problem buried in the map

The collaboration network is where the review turns from description to criticism. Connectivity concentrates around the United States, the United Kingdom, and China, with Canada, European, and Asia-Pacific systems also well linked, while large parts of Africa and Latin America remain thinly connected. Bulut treats this as an epistemic problem rather than a bibliometric curiosity, and he is right to. If the evidence base is generated overwhelmingly in well-resourced digital environments, then the models and practices it validates carry untested assumptions about infrastructure, learner access, and assessment tradition. He extends equity past the familiar question of who has devices to the representativeness and contextual transferability of the knowledge itself, which is a sharper claim than most access arguments make. The threat analysis follows the same line: algorithmic bias, privacy deficits, dependence on automated systems, and the marginalisation of professional judgment are named as risks, and Bulut adds that overreliance on automated feedback can diminish the dialogic, interpretive, and co-regulatory processes that assessment-for-learning research has spent decades establishing. If technical outputs are treated as authoritative rather than provisional evidence, learner agency and teacher judgment are displaced rather than supported. The Digital Divide literature and the Equity literature both bear on this, but the review's contribution is to show the divide operating inside the research base as well as inside classrooms.

Why the framing matters for assessment design

The most useful move in the paper is definitional. Bulut defines a technology-mediated formative ecosystem as the joint product of technological capacity, pedagogical regulation, learner agency, and human judgment, and then measures the field against that definition rather than against tool sophistication. On this basis the review's practical recommendations are coherent rather than generic: prioritize human-centered AI integration, establish transparent governance, pursue equitable access, and invest in professional development so that technology supports formative judgment instead of substituting for it. The recommendation to treat AI output as provisional evidence echoes work on trust calibration and on feedback quality, and the insistence that the value of TSFA "does not primarily lie in automation" but in supporting cycles of evidence, feedback, reflection, and action is the claim an institution can actually act on. That claim also explains why the review is skeptical of the growth story it otherwise documents: a field expanding faster than its theory can justify is accumulating tools without a settled account of what they are for.

What this means for practice

• Judge a formative tool by which part of the evidence-feedback-reflection-action cycle it strengthens, not by whether it automates the assessment step.
• Keep AI-generated and analytic outputs framed as provisional evidence for interpretation, so that students and staff treat them as inputs rather than verdicts.
• Audit which student populations a system's underlying model was validated on before adopting it, since research conducted in well-resourced settings may not transfer to your infrastructure or assessment culture.
• Make governance and data-handling expectations explicit before procurement, because privacy, bias, and transparency problems are cheaper to refuse than to remediate.
• Treat staff development in interpreting analytics and AI feedback as part of the assessment system itself, not as an optional add-on to implementation.

Limitations

• This is a review of published research, so every figure describing the field is a fact about the literature rather than about learners; the review measures what has been studied, not what works.
• The corpus is restricted to Web of Science and Scopus, and the screening excluded 61 records on language criteria, which likely reinforces the geographic imbalance the review itself criticizes: scholarship published outside English-language indexed venues is structurally absent from the map that indicts it.
• Bulut states that the SWOT analysis was used as an interpretive device rather than an independent analytical technique, so its entries restate the bibliometric and thematic findings rather than triangulating them.
• The thematic synthesis is qualitative, and the paper does not report inter-rater reliability for the coding that grouped records into themes.
• The review's own limits on longitudinal evidence constrain its conclusions about trajectory: growth in publication volume is documented, but whether TSFA improves learning outcomes over time is not something a bibliometric map can establish.
• Bibliometric co-word and co-authorship structures reflect indexing and publishing practices as well as intellectual affinity, so clusters and centrality scores should be read as features of the recorded literature rather than as a neutral measurement of the field.

Citation

Bulut, A. (2026). Technology-Supported Formative Assessment in Higher Education: AI, Learning Analytics and the Future of Assessment for Learning. Journal of Computer Assisted Learning.