Online and AI-generated LLND assessments: keeping human review in the process
By Carolyn Willis, Coast Wide Training Solutions
Online language, literacy, numeracy and digital literacy (LLND) tools can make a Registered Training Organisation's (RTO's) pre-enrolment process faster. A score from a system is one input to the suitability decision, and the decision itself stays with the RTO. The Standards for RTOs 2025 set this out, and ASQA's Principles for the Responsible Use of Artificial Intelligence (AI) in VET reinforce it.
Since the 2025 Standards took effect on 1 July 2025, RTOs have been reviewing their LLND processes. The market has responded. There are now online LLND tools that auto-assess in minutes, student management and learning management systems with built-in LLND modules, and AI tools that will write an LLND assessment in seconds.
These tools can genuinely help. In our work with RTOs, though, we keep seeing the same pattern. A link goes out, a number comes back and the enrolment proceeds. When that happens the tool has taken over the whole process, and that is where the compliance exposure sits.
What the 2025 Standards require
Under Outcome Standard 2.2, an RTO must have procedures to review each prospective student's skills and competencies before they enrol. That includes language, literacy and numeracy (LLN) proficiency and digital literacy. Based on the outcome of the review, the RTO advises the student whether the training product is suitable for them.
ASQA's 2025 Standards FAQs clarify three points:
The review happens in context. You review LLN proficiency and digital literacy against the training product the student wants to enrol in. A generic score says little on its own.
You choose the method. You can purchase or develop a tool, or gather other evidence, depending on the training product, delivery mode and cohort.
No framework is prescribed. You analyse each training product to work out the LLN and digital literacy levels it demands. The Australian Core Skills Framework (ACSF) is optional, and it remains the most widely used and defensible option for LLN. For digital literacy, you could use the Australian Digital Literacy Skills Framework or the European DigComp 3.0 (Digital Competence Framework for Citizens).
Automation or AI? Know which one you are using
ASQA draws a clear line between automation and AI in its guidance on the responsible use of AI in VET. ASQA describes automation as rules-based: the same task done the same way every time, such as auto-marking multiple-choice questions. AI interprets and analyses complex inputs, such as giving feedback on a student's written answer.
Most online LLND tools use both. Numeracy items and multiple-choice reading questions are usually automated. Many tools now also use AI to score writing samples, assign an ACSF level from open responses, or recommend whether a student is suitable.
Do you know which components of your LLND tools are automated and which use AI? If not, ask your vendor. You cannot manage a risk you have not identified.
Where online LLND tools need human oversight
Online tools fail in predictable ways when they run without human oversight. The Principles to Guide Foundation Skills Assessment describe good practice as learner-centred, valid, fair and reliable, and conducted by skilled assessors. The Principles are not mandatory, but measured against them, five risks stand out.
The score is not the decision. A tool reports how someone performed on a set of items. It cannot tell you whether that level is enough for your specific training product, delivered in your chosen mode, to this cohort. Making that comparison is the review.
Confirming who completed it. Unsupervised online assessments are easy to complete with help, or by someone else entirely. The Principles ask how online assessment will verify identity and confirm the work is the learner's own.
Digital literacy. A learner with limited digital skills may score low across the whole online assessment. The tool then reports a reading or numeracy gap when the gap is digital. Assess digital literacy separately and offer a non-digital option for the rest.
Writing and speaking. Closed questions mark well automatically. Open writing, speaking and listening need human judgement, and any AI scoring needs human oversight.
Using the benchmark as a pass mark. A result below the course benchmark is a starting point for a conversation. The Foundation Skills Principles frame LLND as diagnostic, so a low result should prompt a discussion about support or other pathways before anyone concludes the student is not suitable.
AI-generated LLND assessments: fast to make, easy to get wrong
AI can create an LLND assessment in seconds. Using AI to write your LLND questions is not prohibited, and the risk sits in using the output without review. Common problems we see in AI-generated LLND tools include:
Levels that do not match the claim. An AI-generated task labelled ACSF Level 3 is not necessarily a Level 3 task. ACSF levels depend on text complexity, task demand, context and support, so someone with LLND knowledge should check any AI-generated task against the framework before it is used.
Generic content. The Foundation Skills Principles stress contextualisation. Tasks should reflect the course, the industry and the learner's world, and generic AI scenarios often do not.
Hidden bias. ASQA names bias as one of the key risks of using AI in VET. In LLND it can appear in two ways. An AI-written task may rely on idioms, cultural references or local knowledge that some learners will not have. AI marking may also score down writing that departs from standard English patterns, even when the meaning is clear. Either way, the result reflects the tool and says little about the learner's skills.
Marking guides that have not been checked. AI can produce a marking guide and sample answers along with the task, and they are only as good as the task behind them. If the benchmark answers do not reflect the level being assessed, every result will be measured against the wrong standard. Check them as carefully as the questions.
If you build a tool with AI, treat it like any other assessment tool. Have someone with LLND knowledge check its levels and contextualisation, trial it, document the review and keep version control.
Our free AI Self-Assessment Checklist is built around ASQA's five Principles for the Responsible Use of AI in VET, including human oversight, and gives you a way to see where AI is already used in your RTO and what you can evidence.
Who makes the suitability decision
The suitability decision is made by a suitably skilled person within your RTO. A system does not make it, and neither does whoever happens to process the enrolment.
ASQA's AI principles are direct on this. Principle 2 states that decisions affecting students remain the responsibility of qualified trainers, assessors and staff, and that AI does not replace human decision-making or accountability. A pre-enrolment suitability decision is exactly that kind of decision.
The Foundation Skills Principles support a practical split. Administration staff can issue the assessment and help with technical issues. The review and decision sit with a trainer and assessor who knows the course and can interpret the results against its LLND demands.
A defensible model
You can keep the efficiency of an online tool and still meet your compliance and student support obligations by placing a skilled person between the score and the enrolment. A process might look like this:
Document each course's LLND requirements in your training and assessment strategy, using the ACSF and a digital literacy framework such as the Australian Digital Literacy Skills Framework or DigComp 3.0, or another framework you can justify.
Offer a non-digital alternative, such as a paper-based, oral or assessor-led option, then administer the tool. Administration staff can help with access and technology, and the assessment content stays with the assessor.
Let the system score what it can, and flag results against the course benchmark.
Apply human oversight. A skilled staff member reviews every result, including writing samples and any AI-generated outputs, and applies them in context.
Interpret results in context. Talk to the learner where results are borderline, below the benchmark, or inconsistent with what you know about them. Results may not reflect a learner's full capabilities.
Record the decision before enrolment is confirmed. Store it against the learner's record in your student management system (SMS), capturing who reviewed the result, when, and what was decided. Where gaps are identified, develop a learner support plan and record that too. The support services, reasonable adjustment and escalation content in your Student Support Guide gives that plan something to point to.
Give the learner feedback on the outcome in plain English.
Evidence to have ready for audit
Documented LLND requirements for each training product
The tool's framework mapping and your record of reviewing it before use
A documented and followed LLND procedure
Completed review records showing human review and a decision for each learner
Support plans or referrals linked to identified gaps
Records of alternative methods offered to learners
Technology can make LLND reviews faster. Responsibility for them stays with your RTO.
Where to from here
If your RTO uses online or AI-generated LLND tools, start by working out which parts are automated and which use AI. The free AI Self-Assessment Checklist walks through governance and accountability, human oversight, privacy and data protection, equity and accessibility, and alignment with training product requirements, with targeted questions and evidence checks for each.
Once a review identifies a gap, students need clear support pathways. The Student Support Guide Pack are editable and cover support services, reasonable adjustment, privacy and confidentiality, escalation pathways, and progress monitoring and intervention, aligned to the Standards for RTOs 2025.
Comments