Guideline on the Use of Generative AI | Department of Sociology

Click here to download the Guideline in PDF Form

Policy Guideline on the Use of Generative Artificial Intelligence for Education

METU Department of Sociology

NB: Everything claimed and argued here is subject to periodical and continuous revision by the department.

NB: This document sets shared departmental expectations. It does not override the academic autonomy of individual instructors, who remain responsible for the design, conduct and assessment of their own courses. Instructors may be more permissive or more restrictive than the provisions below, and where a course policy differs from this document, the course policy governs that course; instructors are asked to state their position in the syllabus and to give the reasoning behind it.

This discretion covers teaching and assessment. It does not extend to obligations arising from University regulations, research ethics approval, or the protection of research participants' data, which apply irrespective of course design and are set out in Annex D.

Updated: September 16, 2026.

0. Structure

Purpose and Scope
Definitions
Core Ethical Principles
Allowances by Assessment Type
Acknowledgement Requirements
Referencing AI Outputs
Verification, Integrity and Evidence
Consequences for Non-Compliance
AI use in Qualitative and Interpretive Research
AI in Thesis and Dissertation Research: Ethics Screening
Informed Consent
Equitable Access
Student Well-being
Roles and Responsibilities
Review and Versioning

Annex A: Course policy templates
Annex B: Assessment design guidance for instructors
Annex C: AI and Data Ethics Screening Form
Annex D: Regulatory framework

1. Purpose and Scope

This policy sets common expectations for ethical, transparent, and pedagogically sound use of Generative Artificial Intelligence (GenAI) in all undergraduate and graduate Sociology courses as well as M.S. and Ph.D. dissertations. It applies to all assessed and non-assessed coursework (essays, problem sets, data analysis notebooks, presentations, posters, portfolios, code, and creative pieces), thesis research, and dissertation writing unless a module outline explicitly states otherwise.

This policy applies to students, instructors, and thesis supervisors. Expectations of transparency are reciprocal: staff who use GenAI in the preparation of teaching materials, in assessment, or in feedback are also subject to the disclosure expectations set out in Section 14.

Instructors retain discretion over the rules that apply in their own courses. The Department nevertheless aims for coherence across a programme. Students experience a degree as a progression rather than as a series of unrelated modules, and inconsistent or unexplained rules from course to course are a predictable source of confusion and of inadvertent breach. Instructors are therefore asked to select from the templates in Annex A rather than drafting policy from scratch, and to coordinate within a programme where the same cohort takes several courses in parallel.

The Council of Higher Education's ethical guidance identifies the principal ways in which GenAI use can give rise to research misconduct, including non-disclosure, unattributed use of others' content, reliance on fabricated or unverifiable outputs, methods that cannot be repeated or explained, the amplification of bias against vulnerable groups, and the unlawful handling of personal data. The full list, together with the relevant national legislation, is reproduced in Annex D.

Why this policy? Four aims:

Learning first

Students must cultivate their own analytical, methodological and writing capacities. GenAI should be a tool to think with, not a machine to think for them.

Transparency and fairness

Students must openly acknowledge how, why, and when GenAI was used, so that staff can evaluate learning fairly and maintain academic integrity.

Scholarly originality and research responsibility

Research and publication must be grounded in the student's own intellectual contribution. GenAI may not be used to generate or substitute core scholarly work requiring disciplinary expertise, including the development of research questions or hypotheses, the interpretation of findings, argumentation, or the formulation of conclusions and implications. The student remains fully responsible for the design, analysis, interpretation and originality of the research.

Protection of the human relationships through which sociology is taught and done

Supervision, seminar discussion, fieldwork and collaborative analysis are not inefficiencies to be optimised away. They are the settings in which judgement is formed. Policy in this Department is designed to protect them.

2. Definitions

AI: “AI refers to a diverse set of automated and semiautomated technologies including (but not limited to) predictive machine learning systems, automated transcription, generative models that produce images and text, chatbots built on large language models (LLMs), and coding agents that automate programming tasks. Some forms of this technology offer substantial benefits for our research, while other forms threaten our core values.” (Crockett, 2026)
GenAI tools: Systems that generate text, images, code, or data transformations (e.g., ChatGPT, Claude, Gemini, Copilot, Perplexity, DALL·E, Midjourney, DeepSeek) and AI features embedded in software (e.g., “rewrite,” “summarise,” or “suggest” in Word, Google Docs, Grammarly, Zotero, MAXQDA, Nvivo, IBM SPSS/R/Python add-ins).
AI acknowledgement: A brief statement explaining which tools were used, for what purpose, when in the workflow, and why; include representative prompts and how outputs were verified/edited.
AI citation: A formal reference to AI outputs when content from GenAI is included in submitted work (see below on section 6).

3. Core Ethical Principles

Pedagogical primacy

AI use must not substitute for the student’s own analysis, argumentation, data work, or writing development.

A further consideration applies at the level of the cohort rather than the individual. Independent intellectual work is socially sustained: it is easier to maintain when it is common, and harder when it is not. Recent modelling of population-level AI adoption suggests that the resulting transition may be abrupt and path-dependent, so that restoring earlier practices requires a considerably larger correction than would have been needed to preserve them. Prevention is cheaper than reversal. This is the principal reason the Department protects unaided work by design (Section 4, Category D) rather than relying on disclosure alone.

Transparency

Any AI involvement must be plainly acknowledged; staff may also require a detailed appendix of prompts/outputs for selected assessments.

Academic Honesty

AI use should not risk the production and publication of scientific knowledge as being free from bias and deliberate errors and integrity in the context of planning, execution, evaluation, and reporting of research.

GenAI systems are not costless infrastructure. Their training and operation consume substantial energy and water; their expansion drives data centre construction, land clearing and electronic waste; their training corpora have largely been assembled from published and unpublished work without the consent of authors; and the filtering of harmful content depends on precarious and psychologically damaging labour, disproportionately in different parts of the world. Students and staff are expected to be aware of these costs, to weigh them when deciding whether a given use is warranted, and to be able to discuss them as part of the sociological analysis of the technology rather than treating them as external to it.

Students and staff may decline to use GenAI, whether on methodological, ethical, environmental or personal grounds. Non-use is never penalised, is never treated as evidence of incapacity, and is never grounds for suspicion. Where the subject matter allows, instructors should ensure that a route through the course exists that does not require the use of these tools, and should say so in the syllabus.

Principle of diligence

AI use does not reduce the student’s responsibility for careful and ethical academic work. Students must critically evaluate AI outputs, verify data and references, and correct inaccuracies or biases. Diligent use of AI includes avoiding plagiarism, fabrication, data manipulation, inappropriate authorship claims, and duplicate or fragmented publication practices, and ensuring that AI use does not harm research participants or collaborators.

Justice and respect

AI use must be fair and respectful to others. Students must properly credit sources, avoid misappropriating others’ work, and remain attentive to bias, harm, and unequal impacts on people, communities, and knowledge.

Critical use

Students must question AI outputs for accuracy, bias, appropriateness, and fit to task; unsupported AI claims should never be passed off as fact.

Attribution

When AI-generated content is incorporated, it must be cited as per the module’s referencing style; unreferenced external/AI content constitutes poor practice or misconduct.

Data protection and confidentiality.

Do not upload personal data, unpublished research data, or confidential materials into public AI tools. Use institutionally approved tools where required. Personal data should be anonymized or masked outside GenAI systems. It should be noted that data privacy protection has both ethical and legal dimensions (Personal Data Protection Law). Students must protect the privacy and confidentiality of all data used in their academic work. This includes respecting informed consent, safeguarding personal or sensitive information, and using AI in ways that do not expose, share, or misuse data about individuals or communities. Students are responsible for storing, transferring, and processing data securely and for seeking ethics approval when required. AI tools must not be used to bypass ethical or legal data-protection obligations.

Accountability and Responsibility.

Students remain fully responsible for all work they submit, including any content produced with the assistance of AI. The use of AI does not transfer responsibility to the tool: students are accountable for the accuracy, integrity, ethical soundness, and consequences of AI-assisted outputs. The fact that AI systems operate as “black boxes” does not excuse errors, bias, or misconduct. All AI-generated material must be critically evaluated, corrected where necessary, and ethically justified by the student.

Contributing to an ethical climate

Students share responsibility for fostering an ethical academic environment when using AI. This includes being attentive to potential ethical risks, raising concerns when problems arise, and using AI in ways that respect human dignity, equality, and fairness. AI should support learning and knowledge production for the common good, not reinforce bias, exclusion, or existing inequalities. Students are expected to adopt a human-centered approach to AI use and to contribute to a culture of ethical awareness and responsibility.

4. Allowances by Assessment Type (to be defined by the instructor according to the assessment - for further details, see Annex A and B)

A. Open use with acknowledgement (allowed):

Idea generation/brainstorming; planning/outlining; seeking clarifications of concepts; generating reading lists to search independently; language support (grammar/flow) with tracked edits and original meaning preserved.
Conditions: Provide the Short AI Acknowledgement (see section 6). Cross-check all facts with scholarly sources.

B. Limited use with acknowledgement (allowed if explicitly permitted by the brief):

Draft critique/revision suggestions; code snippet hints; pseudocode; unit-test scaffolds; boilerplate for data-cleaning scripts; figure-caption suggestions; surface paraphrase suggestions. Conditions: Provide the Detailed AI Log (see 6). Include your final code/analysis; note where you accepted or rejected AI suggestions.

C. Prohibited use (unless the brief explicitly instructs otherwise):

Generating substantive portions of literature reviews; writing whole essay sections; fabricating references or padding citation lists; creating data or simulating interviews, focus groups or survey respondents; auto-producing code or analysis wholesale; using GenAI to generate codes, themes or interpretations in interpretive qualitative work (see Section 9); using GenAI in place of consultation with a supervisor, a TA or peers.

Rationale: these practices undermine the learning outcomes the assessment exists to measure, and several of them constitute research misconduct irrespective of any pedagogical consideration.

C.1 Automated transcription and translation of research material. Uploading audio or video recordings of interviews or focus groups to an external transcription or translation service is a distinct act with distinct risks, and is treated separately from the analysis of text. Voice and image are identifiable biometric data and cannot be de-identified prior to processing. Automatic transcription and machine translation of research material therefore require, in every case: prior approval recorded in the research protocol; participant consent covering the transfer (Section 11); verification of the full transcript against the recording by the researcher; and documentation of the service, model and version used. Where a recording concerns a sensitive topic, or where participants are members of a group for whom exposure carries elevated risk, external processing should not be used.

D. AI-free by design.

Some assessed and non-assessed work in this Department is deliberately conducted without GenAI. This includes, as instructors specify: in-class writing, oral examination and viva-style discussion of submitted work, supervised coding and data work, closed-book synthesis, and first-pass reading and note-taking on set texts.

This category is not a sanction and does not imply distrust. It exists because the capacities this Department certifies, namely the ability to read a difficult text closely, to build an argument, to sit with ambiguous data and to defend an interpretation, are developed through unaided practice and decay without it. Protected unaided work is the mechanism by which the Department keeps that practice available and routine. Instructors using Category D should say in the syllabus which capacity is being protected and why.

5. Acknowledgement Requirements

Short AI Acknowledgement (minimum for any AI use)

At the end of the reference list, add three to six lines stating: (i) which tools were used, with model name and version; (ii) at what point in the workflow and for what purpose; (iii) how outputs were verified or edited; and (iv) what you did not use them for, where a reader might reasonably wonder.

Keep this short. Full prompts and raw outputs belong in the Detailed AI Log (5.2), not in the acknowledgement.

Example:

AI Acknowledgement. I used ChatGPT (GPT-5, OpenAI, accessed 14–16 March 2026) at the outset of this essay to generate candidate search terms for the structure and agency debate, and later to check the grammar of two paragraphs in Section X. All conceptual definitions were subsequently traced to and cited from the scholarly literature; two suggested references did not exist and were discarded. The argument, the selection and reading of sources, and the analysis are my own. No AI tool was used in the analysis of the interview material.

Detailed AI Log (when required by the brief and must for dissertations)

Submit as an appendix: (a) tool name, provider, model and version, with dates of use; (b) full prompts; (c) raw outputs, as an excerpt or file link; (d) what you accepted, modified or rejected, and on what grounds; (e) verification steps, including what proportion of output was checked and against what.

Recording the model version is not a formality. Providers update models without notice and outputs are probabilistic, so the same data and the same prompt will not return the same result twice. A log that does not identify the model and date cannot support any claim about how a result was produced.

Template headings: Tool and version; Date; Prompt; Output (excerpt or link); Your edits and decisions; Verification.

Theses and dissertations

Every M.S. and Ph.D. thesis submitted in this Department must contain an AI statement, placed with the other front matter declarations, describing how GenAI was used in the production of the thesis, including in transcription, translation, analysis, drafting and language editing, or stating that it was not used.

AI systems are not authors and must never be listed as co-authors or credited as contributors. Authorship entails responsibility, which a tool cannot bear. Where students prepare work for publication, they should check the policy of the journal or publisher in advance, as some prohibit categories of use that this policy permits.

6. Referencing AI Outputs

This Department uses APA (8th edition). Where AI-generated content is quoted, paraphrased, or has substantively shaped what you have written, provide an in-text citation and a reference entry.

Reference list format:

Company. (Year). Model name (Version) [Large language model]. URL

Examples:

Anthropic. (2026). Claude (Opus 5) [Large language model]. https://claude.ai

Google. (2026). Gemini (3 Pro) [Large language model]. https://gemini.google.com

OpenAI. (2026). ChatGPT (GPT-5, 2 March 2026 version) [Large language model]. https://chat.openai.com

In-text: parenthetical (OpenAI, 2026); narrative OpenAI (2026).

Give the version as precisely as the provider allows. Where only a dated version is published rather than a version number, use the date, as in the third example above.

Where the prompt and output go

If the exchange matters to your argument, reproduce it. A short exchange may be quoted in the body of the text with the citation attached; anything longer goes in an appendix, to which you refer in text. Where the brief requires a Detailed AI Log (5.2), the log serves this function and need not be duplicated.

Example, in text:

When prompted to summarise the structure and agency debate, the model produced an account organised entirely around Giddens and Bourdieu, with no reference to the critical realist intervention (OpenAI, 2026; see Appendix B for the full exchange).

What not to cite

Do not cite an AI tool as the source of a fact, a definition, a statistic or an attribution. If a claim matters to your argument, locate and cite the scholarly source that establishes it. If no such source can be found, the claim should be removed rather than attributed to the model. Citing a model for a factual claim does not transfer responsibility for that claim; it only records where the error entered.

Where the AI output is your object of study

If you are analysing model outputs as data, for example in a study of algorithmic bias, the material is data rather than a source, and is reported as such: document the model and version, the prompts, the dates, and the sampling procedure in the methods section, and cite the model in the reference list as above.

7. Verification, Integrity, and Detection

Student responsibility:

Fact-check all AI outputs. Trace every claim to a primary or peer-reviewed source. Verify that cited works exist and say what they are reported to say. Document the changes you make. You are accountable for everything you submit, including any inaccurate, biased, or fabricated content produced with AI assistance.

On AI detection software: The Department does not rely on AI-writing detectors, and detector output alone is never a sufficient basis for an allegation of misconduct. This position reflects three considerations.

First, these tools are unreliable in precisely the circumstances that matter. They perform tolerably on wholly machine-generated text and poorly on the mixed cases that make up nearly all real disputes, such as a student-written argument with an AI-edited paragraph.

Second, they misclassify systematically rather than randomly. Text written by people working in a second or third language, and by neurodivergent writers, is flagged at elevated rates. In a department where the overwhelming majority of assessed work is written by students in a language other than their first, a tool with this failure mode would not distribute its errors evenly across the cohort. It would concentrate them.

Third, reliance on detection produces an arms race that consumes the time of both parties and settles nothing, and it converts the relationship between instructor and student into one of surveillance and suspicion. That relationship is the Department's central pedagogical asset and it is not worth trading for a probabilistic score.

What staff may do instead

Where an instructor has concerns, the appropriate response is to seek process evidence and to talk to the student:

request drafts, notes, reading records, or the Detailed AI Log;
set staged deadlines, so that work is visible in progress rather than only at submission;
require that written work be produced on a platform that retains version history, and ask for that history alongside the submission;
ask the student to discuss the submitted work orally, explaining choices of source, method and argument.

An oral discussion is both fairer and evidentially stronger than any detector. A student who has done the work can discuss it; a student who has not, generally cannot. This is also a formative exercise and may be used routinely, not only where there is suspicion.

Standard of evidence

Acceptable evidence of misconduct includes an inconsistency between the submitted work and the student's demonstrated understanding of it, an absence of process evidence where process evidence was required by the brief, a documented discrepancy between the submission and its version history, or the student's own account. Detector output may be recorded as one item of background information but must not be cited as the basis of a case and must not appear as the sole evidence in any file forwarded under University procedures.

8. Consequences for Non-Compliance

Undisclosed but otherwise permitted use

Treated as poor academic practice. The normal response is a learning remedy: the student completes the acknowledgement or log and discusses the episode with the instructor. A grade penalty may apply where the brief specified one in advance.

Incorporating AI-generated content without citation, as one's own work.

Potential academic misconduct; penalties follow University regulations.

Use of AI in “Prohibited” ways

Assessed under misconduct procedures. Where the breach concerns research participants' data, the matter is also reported to the relevant ethics committee, irrespective of the academic outcome.

In all cases, the evidentiary standard in Section 7 applies. Where a case is uncertain, the Department's preference is a conversation with the student and a formative remedy; not a disciplinary referral. Repetition after such a conversation is a different matter and is treated as such.

9. AI use in Qualitative and Interpretive Research

In reflexive and interpretive qualitative traditions, including reflexive thematic analysis, phenomenological, ethnographic, narrative and discourse analytic approaches, the analysis is not a procedure applied to data but an act of meaning-making performed by a positioned, reflexive researcher. Coding, theme development and interpretation are where that work happens. GenAI systems generate text by statistical prediction and do not understand the material they process; they are also disposed by construction to reproduce dominant patterns of language and to flatten what is marginal, idiosyncratic or unexpected in a dataset, which in this tradition is frequently the most analytically valuable material.

A substantial body of expert opinion in the field holds on these grounds that GenAI should be rejected outright for such approaches, at every phase including initial coding. The Department does not adopt a blanket prohibition, but it does adopt a strong presumption against the use of GenAI for coding, theme generation and interpretation in interpretive qualitative work. Supervisors may set a stricter rule for their own students, including complete non-use, and should state it at the outset of supervision.

10. Ethics Screening for Thesis and Dissertation Research

Where a thesis or dissertation involves GenAI at any stage of data handling or analysis, the student completes the AI and Data Ethics Screening Form (Annex C) with the supervisor before fieldwork begins, and submits it alongside the application to the University's Human Subjects Ethics Committee. The form is revisited if tools, providers or institutional arrangements change during the project.

The screening covers ten questions. They are reproduced in full in Annex C; in summary:

Data sensitivity. How private is this material, and what would a breach cost the participants? Judgements of sensitivity are context-dependent: a topic discussed openly in one setting may be private or dangerous in another.
De-identification. What is the protocol, who applies it, and what cannot be de-identified? Names and places are easy; a person's occupational role, biography or the accumulated detail of a long interview frequently are not.
Transmission. Do the data leave the University's network? Under what encryption? Is a third party, such as a QDA software vendor, interposed between the researcher and the model provider?
Storage and retention. Where are the data held after processing, for how long, and under whose terms? Where a QDA assistant routes data onward, the material may be retained by both the software vendor and the model provider.
Training use. Are the data and prompts used to train or improve the model? What are the retention terms? Consumer tiers commonly default to training use.
QDA software assistants. If using MAXQDA, NVivo, Atlas.ti or comparable software, which underlying model is invoked, is it clear to the user when AI is being invoked, and can it be disabled? This must be checked against current documentation, since vendors change models and terms without notice.
Institutional alternatives. Is a local or closed system available, and is it appropriate for this data type?
Bias, reproducibility and integrity. What is the plan to mitigate bias, and what audit trail will be kept? Note that identical inputs will not reliably return identical outputs.
Human validation. Who verifies AI-assisted output, what proportion is verified, against what standard, and how are discrepancies reconciled? This plan must exist before analysis begins, not after.
Societal and environmental considerations. What is the cost of this use, and is it proportionate to what it achieves?

Awareness of a risk is not the same as managing it. For each applicable question, the form requires the student to state the risk, estimate its likelihood and severity, and name the mitigating action.

11. Informed Consent

Where research data from human participants will be processed by AI systems, participants must be told, in plain and accessible language, before they consent.

Tiers of risk.

The description in the consent form should be proportionate to the actual exposure. Three broad tiers apply:

Lowest. Processing that runs locally, without data leaving the researcher's machine or the University network.

Intermediate. Processing on a closed or institutional system with contractual limits on retention and no training use.

Highest. Processing by an external, publicly available service, where data leave the institution, may be stored for an indeterminate period, and may be used to develop the model.

Consent forms already routinely disclose the risk of a data breach. The point of this tiering is that the risk is not constant across tools and should not be described as though it were.

Proportionate description

Disclosure should match the level of detail used for other technologies in the same form. Most consent forms state that data are stored securely on institutional servers rather than describing server architecture. The equivalent applies here: state where data go, whether they leave the University, whether they may be retained or used for training, and what the participant's options are. Use of an internal system will require less description than use of an external one.

Reconsent

Tools, terms and institutional arrangements change quickly, and a change may invalidate the basis on which consent was given. Where the processing arrangement changes materially during a project, the researcher consults the supervisor and, where appropriate, the ethics committee about whether reconsent is required.

Researcher competence

A researcher who cannot explain to a participant what will happen to their recording is not in a position to obtain informed consent for it. Students should seek training where needed, and supervisors should satisfy themselves on this point before fieldwork begins.

The right to decline

Participants may consent to the research while declining AI processing of their data. Where this option is offered, the research design must be able to accommodate it.

12. Equitable Access

Commercial GenAI services are tiered by price, and the gap between free and paid tiers is substantial and widening. Subscription plans at the upper end costs an amount that is prohibitive for a large proportion of students in this Department. Where a course permits AI use without regard to this, students who can pay gain an advantage in assessed work over students who cannot, under a single nominal rule. This is a fairness problem produced by course design, not by student conduct.

Accordingly: Any assignment in which AI use is permitted must be completable, to the full range of available marks, using freely accessible tools. Instructors should test this assumption rather than presume it.

Language

Current systems are trained overwhelmingly on English-language material and perform unevenly in Turkish and in other languages used in this Department's research. Two consequences follow. First, permissive AI policies do not distribute their benefits equally across a multilingual cohort. Second, output in less well-resourced languages carries a higher error rate and requires correspondingly closer verification, particularly in translation of research material. However, as the language of education at METU is English, students are expected and encouraged to write their assignments directly in English at an operational and reasonable level (it is never a question of error-free perfect English), translation should not be necessary anyway.

13. Student Well-being

There are early indications, at this University and elsewhere, that some students are developing a dependent relationship with these systems, using them not only for coursework but for decision-making and emotional support, and reporting a diminished sense of agency, confidence and purpose as a result.

Instructors and advisors are asked to recognise this pattern where it appears and to respond as they would to any other indication of academic and personal difficulty, by speaking with the student and directing them to the University's health and counselling services. This is a matter of student welfare and is handled separately from questions of academic integrity. A student who raises a difficulty of this kind should not find that it is treated as a disclosure of misconduct.

14. Roles and Responsibilities

Students

Use AI critically and transparently. Keep drafts and process records. Comply with data protection, ethics approval and consent requirements. Verify everything.

Instructors

State the AI policy for each course prominently in the syllabus and on the course page, using a template from Annex A, and include the rationale: what the rule is protecting and why. Provide exemplars of acceptable acknowledgement. Specify whether the Short Acknowledgement or the Detailed Log is required for each assessment. Explain how AI use will be evaluated.

Instructors are also advised to observe the following expectations regarding their own use:

Disclose substantial use in teaching materials

Where lecture content, slides, problem sets or examples have been substantially generated by AI, say so. Students notice, and undisclosed use alongside restrictive student rules is read, reasonably, as a double standard.

Disclose use in assessment and feedback, and do not automate them entirely

AI may be used to check work against a rubric or to support the mechanical aspects of marking. It should not be the sole source of feedback on work into which a student has invested substantial effort. A preferable arrangement, where an AI rubric checker is useful, is to give it to students alongside the assignment as a formative tool, rather than using it to determine marks.

Model the practice.

Instructors who explain their own choices, including their decisions not to use these tools, do more to establish norms than any rule in this document.

Supervisors: Agree the terms of AI use with each student at the outset of supervision, including any stricter rule under Section 9.5. Countersign the Annex C screening form where applicable. Satisfy yourself that the student can explain the data handling to a participant before fieldwork begins.

Programme leads and the Department: Ensure consistency across modules within a programme. Maintain Annex A. Hold a periodic review under Section 15.

15. Review and Versioning

This policy is reviewed at least annually, and sooner if institutional regulations, available tools or the evidence base change materially.

Review is the responsibility of a small standing group appointed by the Department. The group is asked to:

gather feedback from staff and students, including through an item in end-of-year forums and, where feasible, in course evaluations;
monitor developments in institutional policy, in the tools themselves, and in the research on their effects on learning;
propose revisions to this text and to Annex A.

Given the rate at which the technology and the surrounding evidence are changing, this document should be read as a current position rather than a settled one. It will require revision, possibly substantial revision, and is written in that expectation.

Annex A: Course Policy Templates

Instructors select one of the following for each course, adapt the bracketed text, and include it in the syllabus. Where different assessments in a course fall under different tiers, state this per assessment. In every case, complete the rationale: a rule without a stated reason is read as arbitrary and is complied with accordingly.

Template 1: AI-free

Generative AI tools may not be used in the preparation of work for this course.

Why: this course is designed to develop [capacity, e.g. the ability to read a difficult theoretical text closely and reconstruct its argument in your own words]. That capacity is built by doing the work, and tools that perform it on your behalf remove the practice through which it develops. You will need this capacity in [subsequent context, e.g. your thesis, your comprehensive examinations, fieldwork]. Assessment in this course is designed accordingly: [e.g. in-class writing, oral discussion of submitted work].

If you have questions about whether a particular tool falls within this rule, ask before using it.

Template 2: AI-limited

Generative AI tools may be used in this course for [permitted uses, e.g. brainstorming, generating search terms, clarifying concepts, language support on your own draft]. They may not be used for [prohibited uses, e.g. drafting any portion of the submitted text, generating or summarising sources, analysis].

Why: [the capacity being protected, and why the permitted uses do not threaten it].

All use must be declared in a Short AI Acknowledgement (Policy §5.1). [For assessment X, a Detailed AI Log is also required.] Undeclared use of a permitted tool is poor academic practice; use of a prohibited kind is misconduct.

Template 3: AI-integrated

Generative AI tools are an expected part of the work in this course. [Describe the role, e.g. you will use them for X, and you will be assessed partly on the quality of your critical engagement with what they produce.]

Why: [why this matters for the discipline, and what you are expected to learn from it].

A Detailed AI Log (Policy §5.2) is required for [assessments]. You remain fully responsible for everything you submit, including any inaccurate, fabricated, biased, or offensive content produced with these tools. All assessments can be completed using freely available tools; you are not expected to purchase a subscription.

[Optional, where the subject allows: a route through this course that minimises AI use is available. Speak to me in the first two weeks.]

Annex B: Assessment Design Guidance for Instructors

Annex B: Assessment design guidance

This annex is advisory. Its premise is that disclosure rules alone will not hold.

B.1 Start from the outcome, not from the tool.

Do not begin by asking whether AI should be permitted. Begin by asking what a student should be able to do, unaided, at the end of this course. Then design assessment that measures and develops that, permitting AI where it helps and excluding it where it does not. The question "should I ban AI in this course?" is unanswerable in the abstract; "what must my students be able to do without help, and how will I know?" is answerable.

B.2 Forms of assessment less vulnerable to substitution.

Oral examination and viva-style discussion. A ten-minute conversation about a submitted essay reveals more about ownership than any detector, and is itself a formative exercise.
Portfolios across a semester. Assessed on development and reflection as well as on the final examination.
Staged submission. Proposal, annotated bibliography, outline, draft, final. Each stage is short and gives the instructor a view of the work in progress, which also prevents the end-of-term pile-up that is the strongest predictor of opportunistic AI use.
Out-of-class work paired with in-class discussion. If the essay is written at home, the seminar following it requires the student to defend and extend it.
Locally grounded tasks. Analysis of material the student has gathered themselves, of a seminar discussion, of fieldwork in a specific setting. Harder to produce from a prompt because the material does not exist in the training data.
Reflective components. Where AI use is permitted, ask students to report where it helped, where it misled them, and what they did about it. This is assessable, and it develops the metacognitive habit the policy is trying to instill.

B.3 A caution about the obvious reflex

The immediate response of many instructors is to shift weight onto timed, in-class examinations. This is understandable and partly right, and Category D of the policy depends on it. But it should be taken with caution. Timed examinations reward speed and recall; they cannot assess the extended, deliberative, research-based work that a sociology degree is supposed to certify. A programme that responds to AI by assessing only what can be done in ninety minutes under invigilation will have protected the integrity of its marks by narrowing what they signify.

B.4 Structured in-person work

Study groups, office hours and seminar participation are declining in departments where AI use is high, because students who can obtain an answer privately at two in the morning have less occasion to seek one from a person. Since these settings are where a good deal of disciplinary judgement is actually transmitted, they need to be built into course structure deliberately rather than left to emerge: graded seminar participation, group work with staff check-ins and individually assessed contributions, structured peer feedback against a rubric.

B.5 Say why, on day one

Explain to students at the start of the course what the assessment design is protecting. The productive difficulty of the work is the point, not an obstacle to it, and students who understand this are considerably more likely to cooperate with rules that cost them time.

Annex C: AI and Data Ethics Screening Form

To be completed by the student with the supervisor before fieldwork, and submitted with the ethics application. Revisit if tools or terms change.

Student: ____________________ Supervisor: ____________________ Project title:____________________ Date: __________ Version: __________

Intended AI use. State, in two or three sentences, what you intend to use and at what stage.

For each question: describe the risk, state its likelihood and severity, and name the mitigating action. Write "not applicable" where it does not apply, with a one-line reason.

#	Question	Risk	Likelihood / severity	Mitigation
1	How sensitive are these data, and is the data management plan proportionate to that sensitivity?
2	How will data be de-identified, by whom, and what cannot be de-identified?
3	Do data leave the University's network? Via what route, and is a third party interposed?
4	Where are data stored after processing, for how long, and under whose terms?
5	Are data and prompts used to train or improve the model? What are the retention terms?
6	If using QDA software with an AI assistant: which model does it invoke, is invocation visible, and can it be disabled?
7	Is a local or institutional closed system available, and is it suitable for this data type?
8	What is the plan for bias mitigation, and what audit trail will be kept (model, version, prompts, outputs, dates)?
9	Who verifies AI-assisted output, what proportion, against what standard, and how are discrepancies reconciled?
10	What are the environmental and social costs of this use, and are they proportionate to what it achieves?

Consent. Which risk tier applies (Policy 11.1)? Attach the relevant consent language.

Interpretive methods. If the project uses a reflexive or interpretive qualitative approach and AI is proposed for coding, theme generation or interpretation, attach the justification required under Policy 9.2.

Declarations

Student: I have completed this form accurately and will notify my supervisor if the tools, providers or terms change. Signature: ____________________ Date: __________

Supervisor: I have reviewed this screening and consider the proposed use, and the mitigations, appropriate to the project. Signature: ____________________ Date:

Annex D: Regulatory Framework

The provisions of this policy operate within, and do not displace, the following national and institutional instruments.

According to YÖK Ethical Guideline (2023) the main situations that may give rise to potential ethical issues (plagiarism, data fabrication, data falsification, data tampering, violation of privacy and confidentiality, etc.) in the use of AI in scientific research and publications are as follows:

-Failure to disclose the use of AI in content production in the work

-Unauthorized use of content produced by others

-Inappropriate quotation of information existing in the literature without attribution

-AI generating incorrect or misleading data, and the researcher using this generated data

-Data and results produced using unrepeatable, unexplained research methods entering the academic literature

-Deepening discrimination against vulnerable groups due to biased and limited data

-Collection, storage, transfer, use, and reuse of personal data in violation of regulations.

Yükseköğretim Kurumları Bilimsel Araştırma ve Yayın Etiği Yönergesi

https://www.yok.gov.tr/documents/documents/68fb63b8b2511.pdf

Yükseköğretim Kurumları Bilimsel Araştırma ve Yayın Faaliyetlerinde Üretken Yapay Zekâ Kullanımına Dair Etik Rehber published by CoHE (2024).

https://proje.yok.gov.tr/documentFiles/17539645334.Yükseköğretimde%20üretken%20yapay%20zeka%20kullanımı-tr.pdf

METU Academic Integrity Guide for Students

https://oidb.metu.edu.tr/sites/oidb.metu.edu.tr/files/php/academic_integrity_guide_for_students.pdf

6698 sayılı Kişisel Verilerin Korunması Kanunu (KVKK)

Personal Data Protection Law

https://www.kvkk.gov.tr/Icerik/6649/Personal-Data-Protection-Law

5846 sayılı Fikir ve Sanat Eserleri Kanunu

https://www.mevzuat.gov.tr/mevzuatmetin/1.3.5846.pdf

YÖK Lisansüstü Eğitim-Öğretim ve Sınav Yönetmeliği

https://www.mevzuat.gov.tr/mevzuat?MevzuatNo=21510&MevzuatTur=7&MevzuatTertip=5

TÜBİTAK Destek Süreçlerinde Üretken Yapay Zekânın (ÜYZ) Sorumlu ve Güvenilir Kullanımı Rehberi. https://tubitak.gov.tr/sites/default/files/2026-01/UYZ_Rehberi_v04_TR.p…

References

Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training (2026) Report. Massachusetts Institute of Technology, 13 August. Available at: https://aiandeducation.mit.edu/report/
Crockett, M. J. (2026). Crockett Lab Statement on AI and Scientific Research. Zenodo. https://doi.org/10.5281/zenodo.22144397
Griffith, J. (n.d.) On taking chances and, and with, predictability: A policy on the use of generative artificial intelligence in my courses, with an explanation of that policy [Unpublished manuscript]. Department of Philosophy, Middle East Technical University.
Guetterman, T.C., James, T.G. and DeJonckheere, M. (2026) 'Ethical considerations for the use of artificial intelligence in qualitative research: Questions for researchers, ethics boards, and reviewers', Methods in Psychology, 15, 100266.
Jowsey, T., Braun, V., Clarke, V., Lupton, D. and Fine, M. (2025) 'We reject the use of generative artificial intelligence for reflexive qualitative research', Qualitative Inquiry.
Solé, R., Ruffini, G., Castaldo, F., Tuccio, M., Seoane, L.F., de Domenico, M., Elena, S.F., Krakauer, D.C. and Levin, M. (2026) 'Large-language models as a cognitive virus', arXiv:2609.03344 [physics.soc-ph]