Artificial intelligence is beginning to change how assessments are designed, delivered and marked. But a new report from the e-Assessment Association suggests that the assessment sector is not rushing blindly into an AI future. The picture is one of experimentation, cautious optimism and growing awareness that trust cannot be automated.

The AI in Assessment: eAA Research Report 2026 is based on 145 responses from assessment providers, educators, employers and technology vendors. Nearly half of respondents were based in Europe. Half worked in colleges, universities or vocational education, while others came from professional certification, schools and corporate or workforce development. (The e-Assessment Association, 2026)

The report is therefore particularly relevant to vocational education and training, where assessment often has to demonstrate more than the recall of information. It may need to show that a learner can perform a task safely, solve a problem in an unfamiliar setting, communicate with others or meet occupational standards.

Adoption is still at an early stage

The first important finding is that most organisations are still learning how to use AI in assessment. Almost 69 per cent of respondents described their organisation as “just starting out”. Around 44 per cent said that their organisation already used AI in assessment, while a further 43 per cent had no current use but planned to adopt it. Most respondents said they personally used general-purpose AI tools such as ChatGPT, Copilot or Gemini in assessment-related work.

This is not a picture of fully automated assessment. It is a picture of people experimenting with tools while organisational practice and regulation are still catching up. The report also notes that 78 per cent of organisations already had an AI policy, suggesting that many institutions are beginning to build formal structures around this experimentation.

Looking ahead, 68.8 per cent of respondents expected AI to support specific aspects of assessment without dominating overall practice. A smaller group, 28.1 per cent, thought that AI would become central to their assessment strategy within the next three to five years.

That distinction is significant. The dominant expectation is not that AI will replace assessment professionals. It is that it will be used selectively for particular tasks.

Where AI is being used

The current uses reported include question and item generation, marking or grading, reporting and analytics, personalised learning, candidate identification and examination invigilation. Respondents also mentioned administrative support, accessibility and adaptive testing.

Those planning to adopt AI expected to use it mainly for marking and reporting. Organisations already using AI were more likely to report experience with item generation, marking, analytics and candidate identification. This may indicate that organisations become more cautious and selective once they move from general interest to actual implementation.

For teachers and trainers, the most realistic immediate use is probably as an assistant. AI can help produce a first draft of a question, suggest possible distractors, structure a rubric, generate feedback prompts or identify patterns in assessment data. But the report repeatedly stresses that these outputs require professional review.

This is particularly important in vocational education. A technically plausible question is not necessarily a good occupational assessment. A generated rubric may overlook safety, tacit knowledge, teamwork or the difference between an acceptable novice performance and an expert one. A marking system may reward the language used in an answer without recognising whether the learner can actually perform the task in practice.

The benefits are real, but mostly organisational

Respondents identified time savings and efficiency as the main benefits of AI in assessment. AI can help with repetitive drafting, item writing, routine feedback and some aspects of marking. It may also improve consistency and standardisation, support more adaptive assessment and help provide feedback more quickly.

These benefits should not be dismissed. Teachers and assessors spend a considerable amount of time on administrative and repetitive work. If carefully designed AI tools reduce this burden, they could create more time for the activities that machines cannot provide: explaining difficult ideas, observing performance, encouraging learners and making professional judgements.

The report also highlights accessibility and inclusion as potential benefits. AI may support alternative forms of interaction, personalised feedback or adaptive testing. In VET, this could help learners practise skills, receive additional explanations or prepare for assessment at their own pace.

But potential is not the same as demonstrated benefit. A system that generates more feedback is not necessarily improving learning. A system that marks more quickly is not necessarily more valid. The central question should be whether AI improves the educational purpose of assessment rather than simply reducing its cost.

The concerns become clearer with experience

The main concerns reported were reliability, academic integrity, the validity of existing assessment methods, loss of nuance, over-automation, privacy, intellectual property, bias and environmental impact.

The report makes an especially interesting distinction between organisations that already use AI and those planning to use it. Planned adopters tend to emphasise efficiency, speed and cost control. They often imagine AI as getting “most of the way there” before a human review. Organisations with practical experience place greater emphasis on quality, consistency, human oversight and the need to manage limitations.

Experience does not make the concerns disappear. It makes them more concrete.

This finding should be taken seriously by institutions considering AI assessment systems. Early enthusiasm may be based on demonstrations or general claims about what AI can do. Practical use reveals the less visible work involved in checking outputs, correcting errors, managing data, explaining decisions and maintaining trust.

The report also notes that some respondents questioned whether existing assessment and psychometric methods remain fully valid in an AI-enabled environment. This is a larger issue than detecting cheating. If learners can use AI to generate written responses, then the assessment may no longer be measuring the knowledge or ability that it was designed to measure. The answer is not simply to purchase a detection tool. It may require a reconsideration of the assessment itself.

What does this mean for vocational education?

VET should not respond to AI by abandoning assessment standards or by assuming that all assessment must become traditional, supervised examination. Instead, AI creates an opportunity to ask what occupational competence really involves and how it can best be demonstrated.

For some learning outcomes, an AI-supported written task may be appropriate. For others, it may be a poor measure. If the outcome is the ability to diagnose a fault, care for a patient, prepare food safely, install equipment, work with a customer or respond to an unexpected workplace problem, assessment may need to include observation, demonstration, dialogue and practical evidence.

This does not make AI irrelevant. AI could help create realistic scenarios, simulate customer or workplace interactions, support practice, generate varied cases or help assessors organise evidence. But the final judgement should remain connected to the occupational context and to the learner’s ability to act.

VET providers should also be careful when using AI for progression, certification or placement decisions. These are not merely administrative uses. They can affect access to employment and further education. If an AI system contributes to such decisions, learners need to know its role, and staff need the authority and information required to review and challenge its outputs.

The report’s findings point towards a simple principle for VET: use AI to strengthen the evidence of competence, not to replace the professional judgement that makes competence meaningful.

What should teachers and trainers do?

For teachers and trainers, the report suggests that AI literacy must be closely connected to assessment literacy. It is not enough to know how to prompt a system. Educators need to ask what an assessment is intended to measure, whether AI changes that construct, how reliable the output is and what kinds of learners may be disadvantaged.

Any AI-generated question, rubric, mark or feedback should be treated as a draft or recommendation until it has been professionally reviewed. Teachers should check factual accuracy, level, relevance to the curriculum, accessibility and possible bias. Trainers should also consider whether the language and assumptions of a system reflect the occupational and cultural context of their learners.

When AI contributes to marking or feedback, students should understand what role it played. They should have a route to question an inaccurate or inappropriate result. A human reviewer needs more than a formal title. They need time, training, access to relevant evidence and genuine authority to override the system.

Assessment design may also need to change. This could mean using more process evidence, oral questioning, practical demonstration, collaborative work, reflective explanation and authentic workplace tasks. The purpose should not be to make assessment more difficult for its own sake. It should be to ensure that the evidence still represents the knowledge, skills and judgement that learners are expected to develop.

The future is likely to be selective, not fully automated

The e-Assessment Association report does not predict the disappearance of teachers, assessors or trainers. Its conclusion is more measured. AI may bring benefits in efficiency, consistency and scalability, but successful adoption will depend on assessment design, governance, organisational capability and sustained human oversight.

That conclusion is especially relevant to VET. The value of vocational assessment lies in its connection to real practice, professional standards and human responsibility. AI can help generate, organise and analyse evidence, but it cannot by itself decide what counts as competent performance in every workplace and every situation.

The most credible future is therefore not one in which AI takes over assessment. It is one in which educators use it selectively, understand its limitations and redesign assessment where necessary. The question is not whether AI can produce a mark. It is whether the assessment remains valid, fair and meaningful for the learner, the teacher, the training provider and the workplace.

References

The e-Assessment Association. (2026). AI in assessment: eAA research report 2026. https://www.e-assessment.com/files/resources/2026/ai-in-assessment-report-2026.pdf

About the Image

Algorithmic memory references the way AI systems increasingly interact with memory. It is inspired by the work of digital memory scholars. It suggests the ways algorithms and automation now shape how we experience memories – and impact how we feel about them – through processes of platformisation, quantification and retrieval. The image is compiled in Microsoft PowerPoint. It includes one public domain image from Cosmos.