Academic Integrity, Research Skills and Referencing

Artificial intelligence (AI), including generative artificial intelligence (GenAI), is increasingly incorporated into study, research, professional communication, information retrieval and everyday digital tools. Occupational therapists completing the AAOT credential are therefore expected to develop sufficient AI literacy to use these technologies appropriately, critically and transparently.

This unit introduces participants to the responsible use of AI in academic study and professional learning. It considers the opportunities AI may provide while addressing academic integrity, authorship, critical thinking, evidence verification, bias, hallucination, privacy, confidentiality, intellectual property, accessibility, referencing and professional accountability.

The purpose of the unit is not to prohibit AI. Participants are instead expected to demonstrate that they can use AI as an assistive learning tool while remaining the author of their work and retaining responsibility for their academic and professional decisions.

AI must not substitute for the participant’s knowledge, critical analysis, clinical reasoning, ethical reasoning, professional judgement or demonstration of competency.

The principles contained in this unit are informed by Australian higher-education guidance, including TEQSA guidance on artificial intelligence and academic integrity, together with university guidance concerning AI literacy and responsible learning.

What Is Artificial Intelligence?

Artificial intelligence is a broad term for computer systems capable of performing tasks that would ordinarily involve aspects of human intelligence.

Generative AI refers to systems capable of generating new content based on patterns learned from large amounts of existing information. Generated material may include:

• written text;
• summaries;
• images;
• audio;
• video;
• computer code;
• tables and structured information;
• study questions;
• explanations;
• simulated conversations; and
• suggestions or recommendations.

Examples of contemporary AI-enabled tools include ChatGPT, Microsoft Copilot, Google Gemini and AI functions embedded within software such as word processors, search systems and writing tools.

Participants should be aware that the distinction between an “AI tool” and an ordinary application is becoming increasingly blurred. AI functionality may be incorporated into applications without the user necessarily initiating a conventional chatbot.

AI is an assistant, not an authority

A fluent AI response is not necessarily a correct response.

GenAI may produce material that:

• sounds authoritative but is incorrect;
• contains fabricated references;
• misstates legislation;
• invents quotations;
• confuses similar concepts;
• uses outdated information;
• omits important qualifications;
• reproduces bias within its underlying data;
• oversimplifies complex clinical issues; or
• gives a confident answer where the available evidence is uncertain.

Participants must therefore distinguish between linguistic confidence and evidentiary reliability.

AI Literacy

AI literacy involves more than knowing how to enter a question into a chatbot.

For the purposes of AAOT, AI literacy includes the ability to:

• understand broadly what an AI system is doing;
• identify situations where AI may be useful;
• recognise when AI should not be used;
• formulate useful prompts;
• interrogate AI outputs;
• identify limitations;
• detect potential bias;
• verify information against authoritative sources;
• protect confidential information;
• acknowledge AI assistance;
• preserve personal authorship;
• recognise ethical issues; and
• remain accountable for decisions made with AI assistance.

An AI-literate practitioner does not ask only:

“Can AI do this?”

They also ask:

“Should AI be used for this purpose, what are the risks, and how will I verify the result?”

Academic Integrity and Authorship

The purpose of assessment within AAOT is to determine whether the participant has demonstrated the required knowledge, reasoning and competency.

Assessment is therefore not intended to determine what an AI system can produce.

Using AI does not automatically constitute academic misconduct. Whether a particular use is appropriate depends on:

• the assessment instructions;
• the purpose of the task;
• the extent of AI involvement;
• whether its use has been disclosed where required;
• whether the resulting work genuinely demonstrates the participant’s competence; and
• whether the participant remains the intellectual author of the work submitted.

The participant remains responsible

Submitting material generated or assisted by AI does not transfer responsibility to the AI provider.

Participants are accountable for:

• factual claims;
• interpretation;
• references;
• legislation;
• clinical statements;
• professional recommendations;
• ethical conclusions;
• analysis;
• wording submitted in their name; and
• compliance with the assessment requirements.

“I got it from ChatGPT” is not an acceptable explanation for an error.

Appropriate Uses of AI in AAOT Study

Where an assessment permits AI use, appropriate applications may include using AI to:

• explain an unfamiliar concept in simpler language;
• generate revision questions;
• create flashcards;
• develop a study timetable;
• assist with retrieval practice;
• brainstorm possible perspectives on a topic;
• identify keywords that may assist a literature search;
• suggest possible structures for a paper;
• explain the difference between concepts;
• generate hypothetical practice scenarios;
• role-play a professional conversation;
• suggest questions for reflection;
• provide feedback on clarity or organisation;
• identify grammatical or spelling errors;
• assist with formatting;
• reorganise the participant’s own notes;
• convert information into another learning format;
• assist with accessibility;
• help explain software or research processes; or
• challenge the participant’s reasoning by presenting alternative perspectives.

AI can therefore function as a study coach, brainstorming partner, simulated participant, questioning tutor or editing assistant.

For example, AI may be asked to generate retrieval-practice questions, analogies, concrete examples or simulated scenarios. These applications can encourage active learning rather than merely producing final answers.

Inappropriate Uses of AI

Unless an assessment specifically requires otherwise, participants must not use AI to replace the competency being assessed.

Examples may include:

• asking AI to complete an entire assessment and submitting it as one’s own work;
• submitting AI-generated clinical reasoning that the participant cannot independently explain;
• fabricating references using AI;
• citing sources the participant has not checked;
• generating fictional case evidence and presenting it as genuine practice;
• generating reflections about experiences the participant did not have;
• asking AI to invent supervision, placement or professional-development activities;
• creating fictional client quotations;
• fabricating interview responses;
• representing AI-generated observations as direct observations of an animal;
• generating an animal behavioural assessment without adequately assessing the animal;
• asking AI to determine whether a participant has satisfied professional competency requirements;
• using AI to impersonate another practitioner or participant;
• submitting AI-generated work where AI has expressly been prohibited; or
• using AI in a way that prevents the assessor from determining what the participant actually knows.

The fundamental question is:

Does the submitted work demonstrate the participant’s competence, or primarily demonstrate the capabilities of the AI system?

The Human-in-the-Loop Principle

AAOT adopts a human-in-the-loop approach to AI.

AI may contribute information, ideas or assistance, but qualified human judgement must remain central.

The participant must:

ASK → REVIEW → VERIFY → THINK → DECIDE → TAKE RESPONSIBILITY

AI output should therefore normally be treated as a draft, suggestion or starting point, not as a finished authoritative product.

This is especially important in occupational therapy because professional decisions may affect:

• a disabled person;
• an assistance animal;
• family members;
• other health professionals;
• workplaces;
• education providers;
• accommodation providers;
• public-access environments; and
• community safety.

Effective Prompting

A prompt is the information or instruction given to an AI system.

Prompt quality can substantially affect the usefulness of the resulting output.

One useful model is the RTRI framework:

Role

Tell the AI what perspective or function you want it to adopt.

Example:

“Act as a study tutor familiar with occupational therapy.”

Task

Clearly explain what you want assistance with.

Example:

“Help me understand the difference between occupation-based intervention and activity-based intervention.”

Requirements

Set boundaries and specify what should be included.

Example:

“Explain the distinction at postgraduate level, include three examples relevant to assistance-animal practice, and identify common misunderstandings.”

Instructions

Explain how the AI should approach the task.

Example:

“Do not write an assessment response for me. Ask me three questions at the end to test whether I understand the distinction.”

Effective prompting can therefore be represented as:

ROLE + TASK + REQUIREMENTS + INSTRUCTIONS

The RTRI structure is one model rather than a mandatory formula. What matters is providing sufficient context and iteratively refining the request.

Prompting for Learning Rather Than Answer Production

Good academic AI use frequently involves asking AI to make the learner do more thinking, rather than less.

Compare the following.

Poor learning prompt

“Write my 1,500-word paper about assistance-animal welfare.”

This transfers much of the intellectual work to AI.

Better learning prompt

“I am preparing a postgraduate paper examining assistance-animal welfare. Ask me one question at a time to help me identify the major ethical tensions. Do not write the paper for me.”

Better critical-thinking prompt

“I believe occupational therapists should have a formal role in assistance-animal assessment. Challenge my position. Give me five plausible counterarguments and ask me to respond to each before providing feedback.”

Retrieval-practice prompt

“Using the notes I provide, ask me ten short-answer questions one at a time. Do not reveal the answer until I attempt each question.”

Evidence-evaluation prompt

“Identify the claims in this paragraph that require evidence. Do not invent references. Tell me what type of source I should search for to substantiate each claim.”

These approaches preserve the participant’s active role in learning.

Critical Evaluation of AI Output

Participants must critically interrogate AI-generated material.

A useful AAOT evaluation framework is:

RABBIT Check

R – Reliability

Does the information appear dependable?

A – Accuracy

Can the factual claims be independently verified?

B – Bias

What perspectives, assumptions or stereotypes may influence the response?

B – Basis

What evidence or authoritative source supports the claim?

I – Information currency

Is the information current enough for the question?

T – Traceability

Can the participant trace important claims back to genuine primary or authoritative sources?

The memorable RABBIT Check is particularly suitable within an animal-related credential while reinforcing systematic source evaluation.

For important information, participants should not merely ask AI whether its own answer is correct.

Verification should occur outside the original AI output through authoritative evidence.

Hallucination and Fabricated Information

Generative AI may “hallucinate”.

A hallucination occurs when AI generates plausible-looking but incorrect or fabricated information.

Examples include:

• nonexistent journal articles;
• incorrect authors;
• fabricated DOI numbers;
• incorrect case names;
• invented sections of legislation;
• false quotations;
• nonexistent organisations;
• incorrect professional standards;
• fabricated statistics; and
• fictional policy provisions.

A particularly dangerous feature of hallucination is that fabricated information may be delivered confidently and in polished academic language.

Therefore:

Never assume that a reference exists because an AI system supplied it.

Every reference used in an AAOT assessment must be independently verified.

Participants should ideally locate and read the original source rather than relying on an AI-generated summary of it.

Source Hierarchy and Verification

When checking an AI-generated claim, participants should seek the strongest source reasonably available.

Depending on the issue, this may include:

1. legislation or regulations;
2. reported cases and authoritative legal sources;
3. government agencies;
4. professional regulators;
5. professional standards and codes;
6. peer-reviewed research;
7. recognised clinical guidelines;
8. systematic reviews;
9. authoritative textbooks;
10. reputable professional organisations; and
11. other credible secondary sources.

AI itself is generally not evidence that the underlying claim is true.

For example, if an AI system states:

“Australian law requires all assistance dogs to pass an annual public access test.”

the participant should locate the relevant legislation or authoritative government source rather than citing the AI statement.

This is particularly important in assistance-animal work, where Commonwealth, state and territory laws may differ.

Bias, Representation and Missing Perspectives

AI systems learn patterns from data created within society. Those data may reproduce existing inequalities and biases.

Potential bias may relate to:

• disability;
• neurodivergence;
• race;
• culture;
• gender;
• sexuality;
• socioeconomic status;
• geography;
• language;
• age;
• body size;
• occupation;
• animal species;
• animal breed;
• mental illness; or
• assumptions about what constitutes a “normal” life.

Participants should consider both what the AI says and what it fails to say.

Useful questions include:

• Whose perspective is represented?
• Whose perspective is missing?
• What assumptions does the answer make?
• Does it pathologise disability?
• Does it assume independence is inherently preferable to interdependence?
• Does it privilege professionally delivered intervention over lived experience?
• Does it stereotype particular assistance-animal handlers?
• Does it assume dogs are the only possible assistance animal?
• Does it make unsupported assumptions about a particular breed?
• Does it overlook animal welfare?
• Does it confuse common practice with legal requirement?

Critical AI literacy therefore overlaps strongly with reflective occupational therapy practice.

AI and Disability

AI may provide significant accessibility benefits.

Participants may appropriately use AI, where permitted, to assist with:

• executive functioning;
• organisation;
• planning;
• language processing;
• dyslexia;
• summarisation;
• converting complex language into plain English;
• generating alternative explanations;
• transcription;
• speech-to-text or text-to-speech workflows;
• formatting;
• translation;
• breaking large tasks into manageable components; and
• creating different modes of study material.

Use of AI for accessibility should not automatically be regarded as academically suspect.

However, an accessibility tool must not inadvertently remove the competency that an assessment is designed to assess.

Where an assessment format creates disability-related barriers, reasonable adjustment should be considered rather than simply prohibiting assistive technology.

Privacy, Confidentiality and Data Security

AI systems may process, retain or otherwise use information entered into them.

Participants must therefore think carefully before placing information into a generative AI platform.

As a general rule:

Do not enter information into a public AI system that you would not be comfortable placing into a third-party external system without appropriate authorisation.

Particular caution is required for:

• names;
• addresses;
• dates of birth;
• photographs;
• medical information;
• NDIS information;
• occupational therapy reports;
• psychological information;
• veterinary records;
• assistance-animal documentation;
• organisational information;
• proprietary assessment tools;
• unpublished research;
• confidential emails;
• supervision notes; and
• client case material.

Client Information and AAOT Practice

This requirement is especially important for AAOT participants.

Identifiable client information must not be uploaded into publicly available AI systems merely for convenience.

For example, a participant should not upload an identifiable occupational therapy report and ask:

“Write an assistance-animal recommendation for this client.”

Nor should they upload identifiable photographs, assessments, clinical notes or veterinary documents without determining whether doing so is lawful, ethically appropriate and consistent with applicable consent, privacy, organisational and platform requirements.

Where AI is legitimately used to assist with learning from a case, information should be sufficiently de-identified and the participant must still consider whether the material could reasonably allow re-identification.

Removing a person’s name alone does not necessarily make a case anonymous.

Animals Also Generate Sensitive Information

Privacy discussions frequently focus exclusively on human clients, but assistance-animal records may indirectly identify their handlers.

Information such as:

• animal name;
• breed;
• photograph;
• geographic location;
• training organisation;
• unusual medical condition;
• social-media history; or
• distinctive assistance tasks

may make a handler identifiable even when the handler’s name has been removed.

AAOT participants should therefore consider the whole client–animal context when de-identifying information.

AI and Clinical Reasoning

Generative AI may assist learning about clinical reasoning, but it must not replace professional reasoning.

AI does not personally:

• observe the client;
• observe the animal;
• conduct an occupational analysis;
• examine the environment;
• establish therapeutic rapport;
• perform a veterinary examination;
• undertake behavioural assessment;
• experience the consequences of its recommendations; or
• assume professional responsibility.

It may also lack important contextual information.

Participants must therefore be particularly cautious when using AI for:

• suitability decisions;
• risk assessments;
• functional assessments;
• animal-selection recommendations;
• task-training recommendations;
• public-access questions;
• welfare decisions;
• clinical documentation;
• legal interpretation; and
• recommendations affecting client or animal safety.

AI may assist the practitioner to think about a decision.

It does not become the decision-maker.

Professional Scope and AI

AI cannot expand a practitioner’s scope of practice.

For example, asking AI to interpret veterinary findings does not make an occupational therapist a veterinarian.

Similarly, asking AI to provide:

• legal advice;
• veterinary diagnosis;
• psychological diagnosis;
• specialist behaviour assessment; or
• medical treatment recommendations

does not remove professional boundaries.

Participants should use AI in ways consistent with:

• their education;
• competence;
• registration;
• role;
• applicable codes and standards;
• legislation;
• organisational policies; and
• professional scope.

Copyright and Intellectual Property

Participants must consider copyright and ownership when uploading or generating material using AI.

Particular caution should be exercised before uploading:

• textbooks;
• paid journal articles;
• proprietary assessment instruments;
• course materials;
• unpublished manuscripts;
• copyrighted images;
• commercial training resources;
• client-created material; or
• another person’s work.

Possessing lawful access to a document does not necessarily mean the participant has permission to upload it to a third-party AI provider.

Participants should consider:

• copyright;
• licence conditions;
• intellectual property;
• confidentiality;
• contractual obligations; and
• the AI provider’s terms of use.

Referencing and Acknowledging AI

AI use must be acknowledged when required by the relevant assessment instructions or WAFA policy.

The method of acknowledgement may vary depending upon:

• the nature of the AI use;
• the referencing system;
• whether AI-generated text has been reproduced;
• whether an AI interaction is recoverable;
• institutional requirements; and
• the nature of the assessment.

Participants should not assume that placing “ChatGPT” in a reference list automatically makes otherwise inappropriate use acceptable.

Disclosure does not convert prohibited conduct into permitted conduct.

Similarly, failure to acknowledge material AI involvement may constitute an academic-integrity concern even where AI itself was permitted.

Recommended AAOT AI Declaration

For AAOT assessments where AI use is permitted, participants may be asked to include an AI-use declaration.

Example – limited learning/editing assistance

AI Use Declaration:
I used generative AI during preparation of this assessment to assist with brainstorming, organisation and review of written clarity. I independently located and reviewed the sources relied upon, developed the analysis and conclusions, and take responsibility for the final submitted work.

Example – more substantial permitted use

AI Use Declaration:
I used [tool] for the following purposes: [briefly describe]. AI-generated outputs were critically reviewed and were not treated as authoritative sources. References and factual claims were independently verified. The analysis, professional reasoning and final conclusions presented in this assessment are my own.

WAFA may prescribe a particular declaration for individual assessments.

Maintaining Evidence of Authorship

Participants should retain appropriate evidence demonstrating development of significant assessment work.

This may include:

• planning notes;
• drafts;
• tracked changes;
• research notes;
• reference-library records;
• handwritten notes;
• earlier versions;
• outlines;
• supervisor discussions; and
• AI interaction records where relevant.

An assessor may ask a participant to discuss their submitted work.

The participant should be capable of:

• explaining the argument;
• defining terminology;
• discussing sources;
• explaining why particular sources were selected;
• defending professional reasoning;
• identifying limitations; and
• applying their conclusions to a different scenario.

A participant who cannot meaningfully explain material submitted in their own name may not have demonstrated competency, regardless of whether the text itself appears sophisticated.

AI Detection Software

AI-detection tools will not be treated as infallible evidence of authorship or misconduct.

The central concern within AAOT should be whether the participant can demonstrate the relevant learning outcomes and authorship through the complete body of evidence.

Evidence may include:

• the submitted work;
• drafts;
• oral discussion;
• practical demonstration;
• reflective explanation;
• source notes; and
• other assessment evidence.

This is particularly important because language style alone is not reliable evidence that AI has or has not been used.

AI as a Tool for Reflective Practice

AI may be useful for facilitating reflection when used appropriately.

For example:

“Ask me questions to help me reflect on an assistance-animal assessment I conducted. Do not write the reflection for me. Ask one question at a time and challenge me to consider the client, animal, environment, ethics and my own assumptions.”

This retains the participant as the reflective practitioner.

By contrast:

“Write a reflection about an assistance-animal assessment for me.”

would not demonstrate the participant’s reflective capability.

AI as a Simulated Learning Partner

AI can also assist participants to practise communication.

Examples could include simulations involving:

• a client requesting an unsuitable animal;
• an employer questioning workplace access;
• a veterinarian raising welfare concerns;
• a dog trainer disagreeing with an OT recommendation;
• an assistance-animal handler reporting public-access difficulties;
• an interdisciplinary case conference;
• a difficult informed-consent discussion; or
• explaining the difference between assistance animals, therapy animals and companion animals.

The participant can respond to the scenario and ask AI to challenge their reasoning.

The educational value comes primarily from the participant responding, not from reading what AI would have done.

AI for Literature Searching

AI may help participants identify:

• search concepts;
• synonyms;
• keywords;
• Boolean search terms;
• possible databases;
• research questions; and
• broad areas of literature.

For example:

“I am researching occupational therapists’ roles in assistance-animal assessment. Suggest synonyms and Boolean search concepts I could use in CINAHL, Scopus and PsycINFO. Do not invent articles.”

The participant must subsequently conduct and document the actual search using appropriate academic databases.

AI-generated lists of references should never simply be imported into an assessment without verification.

AI and Evidence Synthesis

AI may be used, where permitted, to assist participants to organise information they have already reviewed.

For example, a participant may provide their own notes from several articles and ask AI to:

• identify common themes;
• suggest a comparison table;
• identify potential contradictions;
• generate questions about methodological quality; or
• assist with organisation.

However, participants remain responsible for returning to the original research and ensuring the resulting interpretation accurately represents the studies.

AI should not become an invisible substitute for actually reading the evidence on which professional conclusions are based.

AI and Legal Information

AAOT participants will encounter legislation concerning disability, discrimination, animal management, public access, privacy and professional practice.

AI-generated legal information carries particular risks because:

• legislation changes;
• jurisdiction matters;
• similarly named Acts may exist;
• Commonwealth and state laws may interact;
• case law may alter interpretation;
• AI may fabricate cases or statutory provisions; and
• legal questions often depend heavily upon the particular facts.

Participants must verify legal claims through authoritative legal sources.

AI may assist in explaining legal concepts, but it should not be assumed to provide authoritative legal advice.

AI and Animal Welfare

AI-generated animal-training or welfare advice must also be critically assessed.

AI cannot observe:

• gait;
• body condition;
• pain behaviour;
• fear signals;
• fatigue;
• stress;
• handler–animal interaction;
• environmental triggers; or
• subtle changes in behaviour.

Recommendations regarding an assistance animal therefore require appropriately qualified human assessment.

Participants should be alert to AI outputs that prioritise human functional benefit while overlooking animal welfare.

AAOT practice requires consideration of both members of the assistance-animal partnership.

A Practical Decision Tool: PAWS Before AI

Before using AI for AAOT-related academic or professional activity, participants should apply the PAWS check.

P – Permission

Am I permitted to use AI for this task?

A – Anonymity and information security

Does my prompt contain confidential, sensitive, identifying, proprietary or copyrighted material?

W – Work remains mine

Am I still doing the thinking, reasoning and competency that the task is intended to assess?

S – Sources and statements verified

Have I independently checked important claims, references and recommendations?

If any answer creates concern, the participant should reconsider their intended AI use.

AI Use Traffic-Light Guide

GREEN – Generally appropriate where permitted

Examples:

• flashcards;
• revision questions;
• study planning;
• explanation of concepts;
• brainstorming;
• generating keywords;
• practising communication;
• Socratic questioning;
• accessibility support;
• grammar suggestions;
• asking for feedback on structure;
• converting one’s own notes into study formats.

AMBER – Use cautiously

Examples:

• summarising research;
• interpreting legislation;
• analysing case information;
• suggesting interventions;
• drafting professional correspondence;
• editing assessment prose;
• generating literature-review themes;
• synthesising multiple sources;
• discussing clinical decisions.

These uses require considerable human oversight and verification.

RED – Generally inappropriate

Examples:

• uploading identifiable confidential client material to an unapproved public AI service;
• fabricating evidence;
• fabricating citations;
• fabricating case experiences;
• submitting AI-generated assessment work as one’s own;
• replacing required clinical reasoning;
• replacing observation or assessment;
• using AI to make autonomous clinical decisions;
• representing AI-generated content as professional observations;
• using AI contrary to assessment instructions.