Generative AI Enters Early Childhood Assessment Faster Than Research Can Follow

Generative artificial intelligence is being woven into the digital platforms that early childhood educators already use every day, yet almost no empirical research exists on how educators working with children from birth to five actually deploy these tools in assessment and evaluation. That is the central warning of a new open-access paper published on 8 May 2026 in the journal AI, Brain and Child, authored by Kelly Bittner, Kate Highfield and Nicole Thompson. The study combines a scoping review of eighteen empirical studies on digital technology in early childhood assessment with a position piece projecting what happens when generative AI, or GenAI, becomes embedded in those same workflows. Its conclusion is stark: the rapid integration of GenAI is outpacing the availability of evidence to guide it, and a database search conducted as late as March 2026 found no research papers addressing AI use by early childhood educators for documentation, assessment or evaluation.

The timing matters because the sector is under real pressure. The authors note that early childhood education and care systems globally struggle to attract and retain educators, with reports from Australia, the United Kingdom and the United States citing heavy workloads, administrative burdens and reduced time for the assessment and planning cycle. Within this environment, GenAI is increasingly marketed as a remedy, whether through standalone tools such as ChatGPT or Claude or as an embedded feature in commercial documentation platforms. The paper argues that while AI presents a wide array of potential opportunities, it also carries an extensive range of risks, including bias, misinformation, hallucinations, intellectual property disputes and opaque training datasets, all of which are especially pronounced in contexts involving young children.

To build its argument, the team followed the five-stage scoping review framework of Arksey and O’Malley, searching the Education Research Complete and ProQuest databases using a PICO-derived string combining early childhood terms, assessment-for/of/as-learning terms and digital technology terms. From 400 initial records, 124 duplicates were removed, abstract screening eliminated 232 articles, and full-text review excluded a further 26, leaving 18 empirical studies published between 2012 and July 2025. Only studies focused on early childhood settings, excluding school contexts, and explicitly linking digital technology to assessment processes were eligible. The review deliberately excluded any studies of educators using AI, because none appeared in the search, making the archive a portrait of the pre-GenAI digital landscape against which the position piece reasons forward.

What those eighteen studies reveal is a profession already deeply reliant on digital documentation. Almost two thirds of the articles focused specifically on digital images in documentation, with photographs anchoring pedagogical documentation and learning stories. Digital images were found to prompt conversation and reflection between children, families and educators, and multimodal approaches made documentation more accessible and engaging. Apps and tools such as email, WhatsApp and social media were valued for sharing information with families, although often in non-interactive ways. Critically, one study found that documentation apps designed for adult use came at the cost of children’s voice, agency and active involvement, directly contradicting another study showing that successful digital documentation depended on children accessing and engaging with it themselves.

The review also mapped how digital tools intersect with the three dominant assessment framings. Assessment for learning, the formative gathering of information about what children know and can do, featured in all eighteen articles, with portfolios and learning stories the most commonly cited approaches at roughly half of the studies. Assessment as learning, in which children actively participate in observation, documentation and analysis, appeared in thirteen articles, with children adding photographs, choosing portfolio content and co-authoring documentation with educators. Assessment of learning, the summative snapshot, was substantively absent, mentioned briefly in only four articles. The authors attribute this imbalance to the less interactive nature of summative assessment.

Beneath these patterns lie persistent problems that predate any AI. Documentation was repeatedly shown to be subjective, shaped by educator preferences, pedagogical beliefs and contextual factors, with what the reviewed literature calls space for ambiguity. Educators’ domain-specific content knowledge mattered enormously: one study found educators could only document children’s mathematical thinking to the extent of their own understanding of learning trajectories, and gaps in science and art knowledge similarly narrowed what was captured and translated into play-based learning. Family participation improved assessment credibility, yet families were sometimes treated as consumers rather than participants, and educators often lacked confidence in engaging them. Children’s capacity to contribute was frequently underestimated by educators. Lack of time emerged as a key barrier throughout.

It is against this evidence base that the position piece evaluates GenAI, framing it not as a rupture but as the latest step on a continuum running from paper portfolios through tablets to cloud documentation platforms. Many platforms already used by educators, including commercial services named in the paper, now incorporate GenAI features that can generate summaries, analyse text or suggest learning outcomes, shifting tools from recording information to creating or interpreting it. The authors identify particular pressure points: pedagogical documentation, children’s participation and voice, and educator skills and knowledge. Their core finding is an alignment between the potential benefits and challenges of GenAI and those already associated with other digital tools, meaning the sector’s existing difficulties will not simply be automated away.

The risks they enumerate are technical as much as pedagogical. Large language models are pre-trained on extensive and often opaque datasets, making outputs prone to bias and inaccuracy, and responses are less likely to accurately represent minority groups because of gaps and inequities in training data. Educators lacking confidence in a content area may struggle to detect hallucinations or subtly flawed content that appears correct. Research on AI in other educational settings cited in the paper reports that external tools can reduce opportunities for active recall, problem-solving and critical thinking, and can paradoxically increase time burdens as humans compensate with hidden labour for undesirable algorithmic decisions. Contemporary AI systems, the authors add, are not designed for children and present risks to their safety, wellbeing and development.

Yet the paper resists a purely cautionary stance. GenAI’s capacity to deliver targeted information quickly could help bridge the content-knowledge gaps the review documented, and because it draws on large and diverse datasets it may, in principle, extend perspectives beyond an individual educator’s personal beliefs. If educators use GenAI to produce more learning stories, children could gain additional content to reflect on, potentially increasing their opportunities for active involvement. The authors stress, however, that bias is a longstanding feature of assessment predating GenAI, since educator practices were already subjective, so the remedy they propose is not the tool but the practice: critical reflection and critical digital literacy, with the human remaining the expert in the loop and professional expertise, ethical judgement and contextual understanding staying central.

The paper’s limitations are openly acknowledged. Combining a scoping review with a position piece is an unconventional methodology, and a position piece by design offers a defendable perspective rather than solutions. Because no empirical studies of early childhood educators’ GenAI use exist, the review findings were considered against research from other educational contexts and literature on GenAI in children’s lives, an inferential step the authors justify by the urgency of the gap. Their bottom line is a call rather than a verdict: empirical research on early childhood educator use of GenAI is urgently needed to understand current practices, alongside professional learning and policy support, so that whichever direction this increasingly GenAI-assisted world takes, children’s voice and authentic participation in assessment are not the price of efficiency.

Subject of Research: A scoping review and position paper examining digital technology and generative AI in early childhood assessment and evaluation

Article Title: Digital technologies for early childhood assessment and evaluation: emerging implications in a GenAI world

Article References: Bittner, K., Highfield, K., & Thompson, N. (2026). Digital technologies for early childhood assessment and evaluation: emerging implications in a GenAI world. AI, Brain and Child, 2(1), Article 8. https://doi.org/10.1007/s44436-026-00032-8

Image Credits: AI Generated

DOI: 10.1007/s44436-026-00032-8

Keywords: generative AI, early childhood education, assessment and evaluation, digital documentation, scoping review, pedagogical documentation, educator digital literacy, learning stories, child voice, large language models, critical reflection, education technology