AI and vocational assessment: does the evidence still prove competence?

Artificial intelligence is becoming part of everyday work and learning. It can explain unfamiliar subjects, help learners organise their ideas, improve written communication and identify gaps in their knowledge.

For vocational assessors, however, the central question is not simply whether AI has been used. It is: Does the available evidence demonstrate that the learner is competent?

Vocational competence is established through a combination of three connected elements:

1. knowledge and understanding;
2. practical skills and workplace performance; and
3. occupational behaviours.

AI presents a different level of risk—and opportunity—to each element.

Knowledge evidence is currently the area most obviously affected by generative AI.

A learner can use AI to research a subject, explain a technical concept, suggest the structure of an answer or improve the clarity of written work. Used transparently, this can supplement learning and help the learner develop their understanding.

The problem arises when an AI-generated response is submitted as evidence of the learner’s own knowledge. A detailed and professionally written answer may be accurate, but it does not necessarily prove that the learner understands it.

This means written evidence should not always be accepted in isolation. Assessors can confirm understanding through:

• supplementary questioning;
• professional discussion;
• examples drawn from the learner’s workplace;
• asking the learner to explain technical language used in an answer;
• applying knowledge to a different situation; and
• exploring why a particular decision or approach was selected.

AI can support the production of knowledge evidence, but the assessor must still establish that the knowledge belongs to the learner.

Practical performance evidence may initially appear less vulnerable to AI because it shows someone completing a task. However, technology is increasingly capable of creating or manipulating photographs, video, audio and workplace records.

Evidence could also be carefully selected, edited or presented without sufficient context. A recording may show a successful outcome without demonstrating who completed every stage, whether the correct process was followed or whether the performance was consistent over time.

Assessors should therefore consider:

• who produced the evidence;
• when and where the activity occurred;
• whether the learner can be clearly identified;
• whether the evidence shows the complete activity;
• whether supporting records are authentic;
• whether a credible witness can confirm the performance; and
• whether the learner’s explanation is consistent with what is shown.

Where practical competence carries significant safety, regulatory or operational consequences, direct observation remains particularly valuable. Live remote observation, questioning, witness testimony and workplace products can also be combined to strengthen the assessment decision.

The solution is not to distrust all digital evidence. It is to use different sources of evidence to build a reliable overall picture of performance.

Behaviours are generally less vulnerable to direct AI manipulation because they are demonstrated through the way a person performs and interacts over time.

Behaviours may include:

• reliability;
• professional conduct;
• communication;
• teamwork;
• respect for others;
• attention to safety;
• willingness to accept responsibility; and
• responding appropriately to changing circumstances.

AI can produce a polished reflective account describing these behaviours, but it cannot consistently demonstrate them on the learner’s behalf in a real workplace.

Evidence of behaviour is strongest when it comes from:

• repeated workplace observation;
• supervisor or employer verification;
• feedback from colleagues or customers;
• records of real workplace activities;
• professional discussions based on actual situations; and
• examples of how the learner responded to problems or unexpected events.

A single statement or checklist should not automatically be treated as proof. Behaviours are most convincingly established through consistent performance over time.

The three elements of competence should not be assessed as entirely separate matters.

A competent employee must know what to do, be able to do it and behave appropriately while doing it. Strength in one area should not conceal a significant weakness in another.

For example, a learner might use AI to produce an excellent explanation of a safety procedure. The assessor must still establish that the learner understands the procedure, can apply it correctly and consistently behaves in a way that protects themselves and others.

This is where vocational assessment has an advantage. It can draw on real work, direct observation, questioning, products, witnesses and performance over time. Together, these make it considerably more difficult for AI-generated content to create a false impression of competence.

Assessors should not attempt to identify AI use solely by looking for unusually polished language or changes in writing style. These may justify further discussion, but they do not prove inappropriate use.

AI-detection tools should not replace professional judgement. Where concerns arise, the assessor can:

• discuss the evidence with the learner;
• ask supplementary questions;
• request an explanation of how the work was produced;
• seek additional workplace evidence; or
• use another assessment method to confirm competence.

Any suspected misuse should be handled fairly and through the organisation’s established malpractice procedure.

Implications for internal quality assurance

Internal quality assurance should consider whether assessment systems remain reliable as AI use increases. This includes checking:

• whether learners receive clear guidance on acceptable AI use;
• whether AI assistance is acknowledged;
• whether assessment methods establish individual competence;
• whether assessors appropriately combine different evidence sources;
• how the authenticity of digital performance evidence is checked;
• whether behaviours are verified over time;
• whether assessors respond consistently to concerns; and
• whether AI-related risks are reflected in IQA sampling.

Standardisation will be particularly important. Different assessors should not take fundamentally different approaches to similar uses of AI.

AI should not automatically be treated as cheating. It is becoming a normal workplace tool and learners need to understand how to use it responsibly.

A proportionate policy should allow AI to support learning while making clear that:

• learners remain responsible for their submitted evidence;
• AI use must be acknowledged where relevant;
• confidential or personal information must not be entered into inappropriate systems;
• assessment must still demonstrate the learner’s own competence; and
• final assessment decisions must be made by competent people.

The aim is not to prove that a learner completed every stage without technological assistance. It is to establish that the learner possesses the required knowledge, can perform to the expected standard and consistently demonstrates the appropriate occupational behaviours.

AI may change the evidence presented to assessors, but it does not change the fundamental purpose of vocational assessment: to make a valid and reliable judgement about competence.

RI Training supports assessment centres and training providers with assessor and IQA development, assessment design, quality-assurance systems and practical policies for the responsible use of AI.

If your organisation is reviewing how AI affects vocational assessment, contact RI Training to discuss the support available.