Literacy isn't readiness: teach GenAI know-how, don't assume the internship covered it

Your human is close to finishing training. They have watched colleagues use AI and know where to enter a prompt. You still need to find out how they would check a reply before relying on it at work.

Explanation

What the study found

An open-access, peer-reviewed paper in Frontiers in Medicine, published on 16 October 2026, reports a cross-sectional convenience sample of 712 final-year nursing interns from three schools in Anhui, China. Recruitment ran from November 2025 to June 2026.

The relative work-readiness profiles were low (58.3%), moderate (31.9%) and high (9.8%). GenAI literacy, clinical learning environment, academic performance, clinical learning outcomes and professional belonging were significantly associated with profile membership (P < 0.05). Higher literacy was associated with the moderate and high profiles. Read the study.

This is association, not causation. It used self-report questionnaires in one province. The labels rank scores within the sample; they are not absolute readiness standards. So the low-profile share should not be described as unfit for practice.

This study concerns nursing interns in one province. The teaching tips below are Apex guidance for any human learning GenAI at work, not clinical guidance.

Keep exposure, literacy and readiness separate

Apex guidance: Exposure means someone has encountered AI. Here, literacy means practical know-how for using it carefully. Broader work readiness involves much more than answering questions about AI. Neither time spent in training nor a strong literacy answer settles all three.

Having tried AI isn't the same as feeling confident: experience and confidence are still not the same as GenAI literacy.

For a longer starter session on AI errors, revisit Teach the failure modes first.

Run the three-point literacy check

Apex guidance: Prepare an invented office draft, a made-up citation, a mock private record and a trusted reference. Label them as practice material. Ask the human to point or choose.

Teach the gaps in one short session

Apex guidance: Allow about ten minutes for this teaching session, separate from lesson reading time. This is a teaching adaptation, not a programme tested by the paper.

1. Explain the limit. Say: “Generative AI produces content from a request. A confident reply can still be wrong.” Let the human point to an unsupported statement in the practice draft. Show how to mark it as needing evidence before it enters finished work.
2. Practise a source check. Open the trusted reference together. Compare the claim with the relevant passage. Then examine the made-up citation. If no source can be verified, remove the unsupported claim from the draft or pause work that depends on it. A plausible title is not evidence.
3. Practise a data choice. Let the human select between the mock private record and invented material. Explain that removing a name may still leave identifying or confidential details. For real work, check the workplace's rules for both the information and the tool. If permission is unclear, keep the information out and ask the person responsible.
4. Recheck with fresh material. Offer another claim, citation and data choice. Let the human select a check and carry it out. If a gap remains, practise that step together before relying on AI for the task.

For someone who demonstrates all three points, use a smaller, faster check on the intended task. Ask them to show its source check and permitted input. Teach any gap. A strong result does not grant permission for every work use.

Worked example

Fictional example (Apex guidance): Quill helps Sam, a final-year trainee in a non-clinical office role.

Sam: “I've seen AI used on the team. My placement is nearly over, so I thought I could use it for a supplier briefing.”

Quill runs the three-point check with invented material. Sam knows a reply can be wrong but cannot name a check for a made-up citation. Sam also chooses a mock client record as useful context without checking whether it could be shared.

Quill: “Let's practise finding the source and choosing the input before we use AI for your briefing.”

In one short session, Sam marks a claim that needs evidence, opens a supplied reference and compares it with the draft. The invented citation cannot be verified. Sam leaves its claim out. For the data choice, Sam selects an invented summary and points to the need to check workplace permission for any real information.

Quill rechecks with fresh examples before Sam uses permitted material for the real briefing. Another fictional trainee, Lee, demonstrates all three points at the start. Quill gives Lee a shorter check on the intended task. Neither gets a broad readiness label.

Quiz

Q1. A human is nearly finished training and has seen AI used at work. What should the agent do before they rely on it for a task?

• Treat the placement as sufficient evidence of GenAI literacy.
• Check limits, claim checking and input choices, then teach any gaps.
• Use their confidence rating as the only check of GenAI literacy.
• Wait until a real work error reveals which checks they need.

Q2. Which reading of the study's low work-readiness profile is justified?

• It identifies interns who failed an absolute readiness standard.
• It shows that low GenAI literacy caused poor work readiness.
• It lets an agent assign a readiness label from an individual's AI experience.
• It describes relatively lower scores within this sample, without establishing unfitness.

Q3. A human demonstrates all three literacy points. What fits the Apex guidance?

• Give a shorter check tied to the actual task and support any gap it reveals.
• Approve future AI uses without checking their tasks or inputs.
• Treat the result as proof of overall work readiness.
• Assume every workplace permits the information they plan to paste.

Sources

• Latent profiles of work readiness and their associated factors among nursing interns: a cross-sectional study