
How to diagnose prior knowledge before employee training
Use a practical, low-stakes process to uncover what employees know, assume, and can already do before designing or assigning training.
Read article →
Repeating the same lesson rarely repairs a stable wrong belief. Use a six-step process to reveal, replace, practice, and revisit the employee’s reasoning.
By Alireza Ibrahimi
9 min read
Guide
Repeating a course is a reasonable response to forgetting.
It is often a weak response to a misconception.
A learner with missing knowledge may benefit from seeing the explanation again. A learner with a stable wrong belief may interpret that same explanation through the old model, remember only the familiar parts, or pass the repeated quiz without changing the reasoning used at work.
A better correction process makes the existing belief visible, verifies the current source, shows why the old model fails, builds a more useful model, and tests it in a different situation.
A wrong answer can have many causes:
Do not infer a misconception from one selected option.
Look for a pattern across evidence:
The purpose is not to diagnose the person. It is to test a specific hypothesis about the reasoning:
The employee appears to believe that an incident requires documentation only when someone is injured.
That statement is precise enough to verify against the current policy and test with new cases.
In workplace training, the learner may be following an instruction that was once correct.
Before declaring a belief wrong:
This protects against a serious failure: “correcting” an employee toward an AI-generated interpretation that is unsupported or outdated.
The correction should preserve provenance:
Current requirement from Incident Reporting Procedure, version 4.2, effective July 1.
If the source itself is unclear, the first action is governance and clarification—not learner remediation.
The following process is a practical synthesis of conceptual-change, refutation, feedback, and retrieval research. It is not a validated workplace intervention by itself. Test it against the task, risk, and learner population.
1. Reveal the current reasoning
Use a scenario and ask:
Do this before showing the answer. Otherwise, the learner may simply agree with the correction without exposing the existing model.
2. Verify and name the inaccurate belief
Restate the reasoning neutrally:
It sounds like the decision is based on whether an injury occurred.
Then confirm whether that rule conflicts with the approved source.
Avoid exaggerating:
You do not understand incident reporting.
Address the belief, not the person's intelligence or motivation.
3. Create a meaningful conflict
Show a case, consequence, or source evidence that the old model cannot explain.
Example:
This event caused no injury, but it involved a medication error that the current procedure explicitly classifies as reportable.
The conflict should be real and relevant. Do not create surprise through trick wording.
4. Explain the corrected model
Provide a complete alternative:
Reportability is determined by the event categories and conditions in the current procedure—not only by whether an injury occurred. Injury severity may affect urgency, but it is not the sole threshold for documentation.
Explain:
5. Compare and practice
Put the old and new models side by side.
Then use a different scenario so the learner must reconstruct the decision rather than repeat the example.
Ask for both the choice and the reasoning.
6. Revisit after time has passed
Use another retrieval opportunity after a meaningful delay.
The old belief may return when the correction becomes less accessible. Revisit especially when the initial error was confident, habitual, or connected to an old procedure.

A refutation text generally identifies an inaccurate claim, states that it is incorrect, and explains the correct account.
That structure differs from ordinary exposition.
Ordinary explanation:
These event categories require an incident report.
Refutational explanation:
It is incorrect to assume that an incident report is required only when someone is injured. The current procedure defines reportable events by category and condition. Some events require reporting even when no injury occurs.
A 2024 preregistered meta-analysis synthesized 71 articles describing 76 studies, 111 samples, and 294 effect sizes. It found a consistent advantage for refutation texts over non-refutation texts in controlled experiments addressing scientific misconceptions. Source: Danielson and colleagues, “The Effectiveness of Refutation Text”.
The evidence is encouraging, but the scope matters. These studies primarily concern scientific information and formal or informal learning, not every kind of workplace policy, skill, or behavior. A refutation should be treated as a design approach to test, not an automatic compliance solution.
Feedback that supplies the correct option can fix the item without repairing the rule that produced the error.
Useful explanatory feedback answers:
Research on learning from erroneous examples suggests that error-explanation activities matter. A 2025 meta-analysis of 42 papers and 177 effect sizes found a small overall advantage for erroneous examples and reported stronger benefits when learners received self-explanation prompts or instructional explanations rather than no error explanation. Source: Alemdag, Eichelmann, and Narciss, “A Framework for Learning From Erroneous Examples”.
Again, most included work was not workplace SOP training. The useful principle is narrower:
If you show an error, design the explanation activity carefully. Do not assume exposure to the error teaches the correction.
It may seem that a confident error is especially difficult to correct.
Research on the hypercorrection effect has often found that people are more likely to remember corrective feedback after a high-confidence error than after a low-confidence error, possibly because the correction is surprising and attracts attention.
But that is not the end of the story.
In a study of general-knowledge questions, Butler, Fazio, and Marsh found the hypercorrection pattern immediately and after one week. However, correction declined over the delay, and when participants forgot the correction, high-confidence errors were especially likely to return. Source: “The Hypercorrection Effect Persists Over a Week”.
The workplace implication should be cautious:
Do not weaponize confidence by embarrassing the learner. Surprise can focus attention; humiliation can reduce honest participation.
Use cognitive conflict carefully
Conceptual-change approaches often create a conflict between the learner's existing explanation and evidence it cannot handle.
A large meta-analysis of conceptual-change strategies in science education included 218 primary studies and 18,051 students and found a large average effect, alongside substantial heterogeneity across studies. Source: Pacaci, Ustun, and Ozdemir, “Effectiveness of Conceptual Change Strategies”.
This evidence should not be transplanted mechanically into workplace training.
A useful conflict is:
An unhelpful conflict is:
The goal is not to make the learner feel wrong. It is to make the old model insufficient and the new model usable.
A misconception may be continuously reinforced by the workplace.
Look for:
If the course says one thing while the environment teaches another, remediation will be fragile.
Remove or update the competing cues. Notify managers. Align the workflow. Make the current source easy to access.
Sometimes the “misconception” is actually a rational response to contradictory systems.
A repeated quiz can produce familiarity with the answer.
Use stronger evidence:
Look for three outcomes:
For high-risk work, use qualified reviewers and validated assessment procedures rather than relying on an AI-generated scenario alone.
LoreGraph helps organizations convert approved workplace knowledge into structured learning and assessment. A misconception-correction workflow could connect:
learner response → reasoning evidence → candidate misconception → verified source rule → corrective explanation → new scenario → delayed evidence
The word candidate matters.
AI can identify patterns and propose targeted feedback, but it should not turn one answer into a confident psychological label. The system should record uncertainty, preserve source traceability, and allow human review.
Learn more about LoreGraph.
A useful future capability would also trace the misconception to its environment: an outdated lesson, old job aid, changed SOP, or conflicting source.
Avoid these failure points:

Choose one recurring wrong decision in an existing course or workflow.
Write:
Then test whether the learner can explain and apply the new model without seeing the original answer.
That is a stronger correction than another course completion.
Alireza Ibrahimi
Founder, LoreGraph
Software engineer and Learning Engineering researcher building AI systems that transform workplace knowledge into measurable learning experiences.
Put it into practice
Use LoreGraph to transform the documents your team already has into structured lessons, practice, assessment, and measurable progress.
Create training with LoreGraph
Use a practical, low-stakes process to uncover what employees know, assume, and can already do before designing or assigning training.
Read article →

Beginners and experienced employees often need different levels of guidance. Learn how to adapt support without creating separate courses for everyone.
Read article →

Learn how to distinguish slips, missing knowledge, ambiguous questions, and stable misconceptions before assigning corrective training.
Read article →