In a recent assessment, I asked students to do something that sounded fairly simple: find references to support a set of assumptions used in a tissue engineering simulation model.
The task was part of a tissue engineering assignment where students were given several assumed values (such as oxygen diffusivity and cell density) within a scaffold. Students were asked to find credible references supporting each value and to provide relevant quotes from those references. The aim was not to find a perfect numerical match – for any biological researcher, biological values naturally vary – but to demonstrate that the assumptions were grounded in credible literature and within a reasonable range.
On the surface, this seemed like a straightforward literature verification exercise. In reality, it revealed a growing challenge for assessment design: how do we ensure that cited evidence is actually accurate, supported, and grounded in real sources?
When a citation isn’t backed by the source
I do want to first assure readers: the vast majority of students completed the task appropriately and genuinely, as is typically the case. They located legitimate sources, extracted relevant evidence and quotes, and used the exercise to demonstrate an important professional skill: justifying assumptions.
However, a small number of submissions raised a more concerning issue. Some students provided references that did not appear to exist. Others provided real references, but with quotes that were not present in the cited paper. In other words, the citation looked legitimate, but the evidence was not grounded in the cited source.
This second category was particularly difficult to identify. A non-existent paper can often be spotted relatively quickly. A real paper paired with an unsupported quote is much harder to catch – it requires the marker to open the source, search for the quoted wording, check whether the claim appears elsewhere in the paper, and sometimes manually read sections when search functions don’t work reliably (for example, if the source isn’t accessible through the university library, or if it’s a scanned image of an older text).
This creates a marking challenge, but more importantly, it points to a deeper issue around assessment and academic integrity (Sydney login required).
Has technology changed what ‘citation’ can look like?
In pre-generative AI contexts, student referencing and source-use problems tended to centre on poor paraphrasing, patchwriting, incomplete attribution, or difficulty integrating sources into a coherent argument.
Generative AI has made a different kind of issue more visible. Tools can produce references that look plausible, are formatted correctly, and appear academically convincing. They can also generate sentences that resemble direct quotes from papers, even when those sentences don’t appear in the source.
This means the mere presence of a citation is no longer sufficient to show that a claim has been verified.
A reference list may look scholarly while still lacking a genuine evidence base.
This matters across disciplines. Whether in engineering, the sciences, the humanities, or professional fields, students are regularly expected to work with and integrate literature through literature reviews, theses, research proposals, design reports, policy analyses, and more. The academic skill being assessed is not simply “can the student produce references?” It is whether the student can evaluate, verify, and use evidence responsibly.
When assumptions shape the outcome
In engineering, assumptions are not decorative. They shape the model, the outputs, and the conclusions that follow.
In the oxygen diffusion simulation above, students were not simply being asked to “add references”. They were being asked to justify values used within a model. If a student assumes a particular oxygen diffusivity or consumption rate, that assumption affects the simulated oxygen distribution – and therefore the interpretation of whether a scaffold design is viable.
This is why justified and credible assumptions are a key engineering and scientific capability. Engineers often work with incomplete information, but they must be able to show where their assumptions came from, why they are reasonable, and what limitations they introduce.
When evidence is not genuinely supported, this process is weakened. The issue is not the tool used to generate the reference, but the absence of a real evidence base.
What happens in longer reports and theses?
This issue also raises a broader concern about scale.
In my tissue engineering assessment above, students were asked to verify a relatively small number of references and quotes. Even then, checking whether a reference existed, whether the quote was accurate, and whether the source supported the claim added complexity to the marking process.
The concern becomes greater in assessments requiring extensive literature integration. If unverified or unsupported evidence appears in a short task with fewer than 20 references, what happens in a literature review, thesis, or dissertation with 50, 100, or more references, where checking all 100+ references may not be practical?
This does not mean that literature-based assessment is no longer valuable. If anything, the ability to evaluate and verify evidence is becoming more important, not less. But it does suggest that traditional formats – where the final product is assessed mainly as a polished written synthesis – may need to be complemented by approaches that better capture how well students can actually work with evidence.
Rather than focusing only on the final narrative, assessments may need to look at how students select sources, interpret evidence, and demonstrate that their claims are supported.
Practical strategies for educators
These challenges are best met by helping students develop strong evidence habits, rather than by catching them out. The strategies below focus on making expectations clear and building students’ capability to work with evidence well.
- Make it clear that working with evidence includes a responsibility to check it – that every reference, quote, and claim genuinely says what students think it says, regardless of how the source was found. Framing this as a professional skill, rather than a rule, helps students see why it matters.
- Encourage students to include hyperlinks (to a DOI or library record) for web-based sources. This makes it easier for students to double-check their own work, and for markers to follow the trail of evidence.
- Assess the quality of the evidence-to-claim connection, not just citation formatting. A citation works best when it functions as evidence rather than decoration – and showing students this distinction is itself a valuable teaching moment.
- Use selective or sample-based verification when marking, where feasible. This helps you spot patterns and signals to students that genuine engagement with sources is what’s valued.
- Help students understand that finding a real source isn’t the finish line – the skill lies in showing that the source actually supports the claim being made.
A citation works best when it functions as evidence rather than decoration
The broader challenge
Students need to understand that plausible-looking references and claims are not necessarily reliable. The risk is not only deliberate misconduct, but also misplaced trust. A student may rely on a tool to produce references, check that the papers exist, and still submit work that is academically fragile because the evidence doesn’t support the claims.
This issue is likely to become more visible in assessments that rely heavily on literature integration, which only reinforces how important these assessments are. Students will need to learn how to navigate evidence in an environment where convincing but unsupported claims can be produced easily.
The response is not to move away from evidence-based assessment, but to rethink how we teach and assess it. Working with evidence is a capability: evaluating sources, verifying claims, and showing that conclusions are grounded in credible support. That’s a skill worth teaching deliberately, and one students will carry well beyond university.
The shift is from asking “Has the student cited a source?” to “Has the student shown that the source supports the claim?”
