AI and your research
How to use generative AI well and honestly when you do research: the rules that apply, where AI helps, where it does harm, and how to disclose it.
A chatbot can explain a concept, suggest search terms or criticise your draft in seconds. It can also invent sources, leak interview data and give you the feeling that you understand a region you have not studied. This page helps you tell the difference. Above all, it asks you not to hand over the two things this quarter exists to build: your skills and your own thinking. It does not replace the rules of your course: where your course says something else, your course applies.
Keep for yourself ↓The rules ↓Where AI helps ↓Test yourself ↓Disclose it ↓Two things to keep for yourself
Most of this page is about using AI honestly. This part is about using it wisely. Even where the rules allow it, do not hand over to AI the two things this quarter exists to build: the skills of your discipline, and your own thinking.
This quarter trains skills you will use for the rest of your working life. A skill is learned by doing it, badly at first and then better. If AI does the doing, you hand in a product and keep none of the skill.
- Drawing and sketching a spatial strategy, by hand and on screen.
- Writing an academic argument: a paragraph with a claim, evidence and a reason.
- Reading a difficult text closely and summarising it in your own words.
- Interviewing, mapping, presenting, coding an analysis you can explain.
Framing a problem, deciding what counts as evidence, judging between options, taking a position and defending it: this is the work of an urbanist, and the work your report is graded on. AI can write an answer that sounds like thinking. Accepting it leaves the thinking undone.
- Choosing your problem and your research questions.
- Deciding which stakeholders matter and why.
- Interpreting your interviews, maps and data.
- Judging which strategy is fair, and for whom.
Sketch the strategy yourself. Then ask what you missed.
Write the paragraph yourself. Then ask for feedback on its structure.
Take your position. Then ask AI to argue against it.
Read the text. Then test your understanding, never the other way round.
AI used this way is a sparring partner. Used the other way round, it becomes a substitute, and the quarter teaches you nothing you can use when it is not in the room.
The rules, layer by layer
There is no single rulebook for AI. Rules come in layers, from your assignment up to European guidance. The closest layer is the most specific, so start there. Open each layer.
1Your assignment and your courseclosest+
This is the closest layer, and it decides most cases. The course guide, the assignment brief or Brightspace says which uses of AI are allowed for each piece of work. TU Delft advises teachers to state this with the AI Assessment Scale, which runs from no AI through AI planning, AI collaboration and full AI to AI exploration (Perkins et al., 2024, as cited in TU Delft Teaching Support, n.d.-b).
The rules of AR2U088, as set out with the AI use statement on the Report page:
- Allowed, if declared: language editing and translation, summarising your own notes, brainstorming, help with GIS or code.
- Not allowed: letting AI produce your research questions, analysis, interpretation, conceptual framework or individual reflection; citing sources you have not read.
- Never enter personal data from participants (transcripts, recordings, names, survey answers) into AI tools.
- Mark AI-generated or AI-altered images in their captions, and cite quoted AI output in APA 7th edition.
2Your programme and faculty+
The Board of Examiners of the Faculty of Architecture and the Built Environment decides whether fraud has taken place. TU Delft defines fraud as any act or omission by a student that makes a proper assessment of their knowledge, insight and skills impossible, in whole or in part (TU Delft Teaching Support, 2025). Using AI where it is not allowed, or without disclosing it, can fall under that definition.
For an example of how the faculty reasons, read the AI policy of the MSc Geomatics, another programme in our faculty. It allows AI as a learning aid, holds students fully responsible for everything they submit, asks for a disclosure statement, and judges fraud by whether AI output replaces the knowledge or skills being assessed. A suspicion is enough for an oral check of authenticity (MSc Geomatics for the Built Environment, 2026). It does not govern this course, but it shows what the faculty expects.
TU Delft advises teachers not to rely on AI detectors, which are often inaccurate (TU Delft Teaching Support, n.d.-c). The real test of your work is whether you can explain and defend every part of it.
3TU Delft+
TU Delft has no university-wide AI policy for education yet. In September 2026 the policymakers said a framework was expected in the autumn, if it passes the formal decision-making process, and that programmes will decide how to apply it (Lont, 2026; TU Delft Teaching Support, n.d.-a).
TU Delft’s teaching support pages name Microsoft Copilot Chat, used after logging in with your NetID, as the only generative AI tool permitted for use at TU Delft (TU Delft Teaching Support, n.d.-a). Earlier guidance called it the preferred chatbot because it runs in a secure environment (TU Delft Teaching Support, 2025). Whatever tool you use, keep personal and confidential data out of it.
If your research involves people, you must follow TU Delft’s Personal Research Data Workflow, write a data management plan and obtain approval from the Human Research Ethics Committee before you start (TU Delft Library, n.d.). That includes deciding where recordings and transcripts may go, which rules out most external AI tools.
4The Netherlands and Europewidest+
The Netherlands Code of Conduct for Research Integrity rests on five principles: honesty, scrupulousness, transparency, independence and responsibility (KNAW et al., 2018). TU Delft fully supports it. They apply to how you use AI as much as to any other method.
The European Commission’s Living guidelines on the responsible use of generative AI in research, now in their fourth version, set out four principles: consider whether AI is the right tool for the task; accountability stays with the researcher; confidentiality is not negotiable; transparency and disclosure build trust. They add that AI systems are neither authors nor co-authors, and that AI must never be used to fabricate, falsify or manipulate research data (European Commission, 2026).
The General Data Protection Regulation applies to any personal data you collect, including what you type into a chatbot.
Four habits that follow from all four layers
You are responsible for every word, number, map and reference you submit, including what AI produced.
Check every fact and every source against the original. AI output can be fluent and wrong.
Keep personal data, interview material and confidential documents out of external AI tools.
Say which tool you used, for what, and how you checked the result.
Where AI helps, and where it does harm
AI is useful at some stages of research and damaging at others. Open each stage of your research to see what to use it for, where to be careful and what not to do.
01+Framing the problemUse · careful · do not
Explain unfamiliar concepts. Suggest angles that you then test against evidence. Play the critic of your problem statement.
AI repeats dominant framings and English-language literature. Dutch planning practice may be thin, outdated or wrong in its answers.
Let it choose your problem. Problems come from the region and your evidence.
02+Finding literatureUse · careful · do not
Brainstorm search terms, synonyms and Dutch equivalents, then search the Library catalogue, Scopus or Google Scholar yourself.
Research assistants that search real databases can still misreport what a paper says.
Ask a chatbot for references and copy them. In one test, 55% of the works cited by GPT-3.5 and 18% of those cited by GPT-4 did not exist (Walters & Wilder, 2023).
03+ReadingUse · careful · do not
Explain a difficult paragraph. Quiz you on a text you have read, to test your understanding.
Summaries flatten arguments and drop the qualifications that matter.
Cite a source you know only from an AI summary. Read and cite the original.
04+Fieldwork and interviewsUse · careful · do not
Draft an interview guide that you then test and revise. Prepare a first translation of a consent form that a person then checks.
Transcription tools and AI notetakers send audio and text to third parties (European Commission, 2026).
Upload recordings or transcripts with personal data to external tools. Follow the Personal Research Data Workflow and your ethics approval.
05+Policy and documentsUse · careful · do not
Translate a Dutch policy text for a first reading. Find where a theme appears in a long document.
Legal and planning terms are easily mistranslated. Check them against official texts on wetten.overheid.nl or IPLO.
Quote a machine translation as if it were the source. Quote the original and give your own translation.
06+Data, code and mapsUse · careful · do not
Help write GIS or data scripts that you understand line by line. Explain error messages.
Code that runs can still calculate the wrong thing.
Let AI produce figures or maps of your region without data. Use real sources such as CBS or PDOK, and cite them.
07+WritingUse · careful · do not
Give feedback on the structure, clarity and grammar of your own draft.
Polished, generic prose can hide a weak argument, and your own voice disappears.
Submit AI-written text as your own, or let AI rephrase a source so that you need not cite it. Both can be fraud (MSc Geomatics for the Built Environment, 2026).
08+Images and designUse · careful · do not
Make quick mood images for a team discussion, labelled as AI-generated.
Images carry stereotypes and invent places that look real.
Present AI images as site photographs or as your design. Mark every AI-generated or AI-altered image in its caption, as the course rules require.
09+ReflectionUse · careful · do not
Check spelling, if your course allows it.
A reflection written with AI help usually reads like everyone else’s.
Let AI write your individual reflection. It is the one part of the report that is yours alone.
Test yourself: six situations
Decide first what you would do, then open the card. The situations are invented, but each one happens every quarter.
01+A chatbot gives you five references on energy poverty in Groningen, perfectly formatted in APA.Think first, then open +
Search each title in the Library catalogue or Google Scholar, open it, and check that it says what the chatbot claims. Chatbots invent references that look real (Walters & Wilder, 2023). If you cannot find a source, it does not exist for your report.
02+You have twelve interview transcripts and want a chatbot to find the themes.Think first, then open +
Transcripts are personal data. Uploading them to an external tool needs the consent of your interviewees and a tool your ethics approval allows (European Commission, 2026; TU Delft Library, n.d.). Even then, coding is your analysis: using AI for it is substantial use that you must disclose.
03+You use Copilot to translate the provincial environmental vision (omgevingsvisie) into English.Think first, then open +
Check the key planning terms, quote and cite the original Dutch text, and say in your AI statement that you used machine translation.
04+AI rewrites a paragraph from a paper so that it no longer looks copied.Think first, then open +
The ideas are still the author’s. Rephrasing them to avoid a citation is plagiarism, and a faculty policy names it explicitly as fraud (MSc Geomatics for the Built Environment, 2026). Cite the source.
05+You ask AI to criticise your research questions and suggest gaps.Think first, then open +
Criticism can sharpen your thinking, as long as the questions stay yours: the course rules do not allow AI to produce your research questions. Rewrite them yourself, and declare the use in your AI use statement; European guidance counts identifying research gaps as substantial use (European Commission, 2026).
06+Copilot corrected the spelling and grammar of your final report.Think first, then open +
European guidance treats spelling, grammar and formatting as minor uses that generally need no disclosure (European Commission, 2026). AR2U088 is stricter: language editing is allowed only if declared, so name it in your AI use statement.
What AI does to research
Rules say what is allowed. Ethics asks what kind of researcher you become, and whose knowledge counts. Four reasons to use AI sparingly and deliberately. Open each card.
Illusions of understanding+Messeri & Crockett, 2024Open
Lisa Messeri and Molly Crockett argue that AI tools can make researchers believe they understand more than they do, and can narrow research into monocultures where a few methods and viewpoints crowd out the rest. Producing more is not the same as understanding more.
Whose knowledge?+Bender et al., 2021Open
Large language models reproduce the views that dominate their training data. Local, informal, non-English and marginalised knowledge is under-represented. In urbanism, the people in the empower corner are the ones a chatbot is least likely to know about. Fieldwork cannot be replaced.
Energy and the right tool+Luccioni et al., 2024Open
Measurements of AI in use show that multi-purpose generative models consume far more energy per task than models built for one task. Before you open a chatbot, ask whether a search, a library database or your own reading would do the job better (European Commission, 2026).
Power and labour+TU Delft Teaching Support, 2025Open
TU Delft’s own guidance lists the power of large technology companies, exploited labour and environmental impact among the ethical concerns of AI. Using a tool is also supporting the business model behind it. See the Ethics page.
Disclose it: your AI statement
Disclosure is how your readers can trust your work, and how your teachers can see your own contribution. In AR2U088 the AI use statement is mandatory and goes in the colophon (Part 02 of the report). The course template gives you Option A (no AI used) or Option B (each tool and version, the task, where in the report, how you checked the output).
What to disclose
Substantial use: AI as a method or a source, interpreting data, carrying out a literature review, identifying research gaps, formulating research aims or hypotheses, translating, drafting text (European Commission, 2026; MSc Geomatics for the Built Environment, 2026).
Minor technical uses that do not affect the substance: spelling, grammar, polishing and formatting (European Commission, 2026). In AR2U088, declare language editing too: the course allows it only if declared. When in doubt, disclose.
What a good Option B says
Adapted from the MSc Geomatics disclosure rules and the European guidelines (MSc Geomatics for the Built Environment, 2026; European Commission, 2026).
| Tool and version | Which tool, which version, when. For example: Microsoft Copilot Chat via TU Delft NetID, March 2027. |
|---|---|
| What for | The tasks you used it for: translating a policy document, feedback on the structure of chapter 3. |
| How | What you asked, with one or two example prompts. Keep a prompt log so that you can show more if asked. |
| How you checked | How you verified the output: against which sources, by whom. |
| What it changed | A short reflection: what you kept, what you rejected, what you learned. |
Invented example: I used Microsoft Copilot Chat (TU Delft NetID, February and March 2027) to translate the provincial environmental vision into English for a first reading, and to give feedback on the structure of chapter 3. I asked it, for example, to list the main claims of each paragraph. I checked every translated passage I quote against the Dutch original, and quote the original in the text. I rejected its suggestion to merge sections 3.2 and 3.3 because it removed the comparison between the two municipalities. I did not use AI for the analysis of interviews.
Cite the tool as software when you used it for a task. Cite a specific chat, with its shareable link, only when readers need to see it. Prompts do not go in the reference list; describe them in your method section or an appendix (McAdoo et al., 2025).
Template: Company. (year). Tool name [Large language model]. URL
Example from the APA Style blog: OpenAI. (2025). ChatGPT [Large language model].
The texts behind this page
Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), Article e2422633122. https://doi.org/10.1073/pnas.2422633122
Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (pp. 610–623). Association for Computing Machinery. https://doi.org/10.1145/3442188.3445922
European Commission. (2026). Living guidelines on the responsible use of generative AI in research (4th version). Publications Office of the European Union. https://research-and-innovation.ec.europa.eu/document/download/2b6cf7e5-36ac-41cb-aab5-0d32050143dc_en Accessed on: 2 November 2026.
KNAW, NFU, NWO, TO2-federatie, Vereniging Hogescholen, & VSNU. (2018). Netherlands code of conduct for research integrity. https://filelist.tudelft.nl/TUDelft/Over_TU_Delft/Strategie/Integriteitsbeleid/Netherlands%2BCode%2Bof%2BConduct%2Bfor%2BResearch%2BIntegrity_2018_UK.pdf Accessed on: 2 November 2026.
Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (Article 1121, pp. 1–22). Association for Computing Machinery. https://doi.org/10.1145/3706598.3713778
Lont, N. (2026, 23 September). The much-anticipated AI policy is nearing publication, say TU Delft policymakers. Delta. https://delta.tudelft.nl/en/article/the-much-anticipated-ai-policy-is-nearing-publication-say-tu-delft-policymakers Accessed on: 2 November 2026.
Luccioni, A. S., Jernite, Y., & Strubell, E. (2024). Power hungry processing: Watts driving the cost of AI deployment? [Conference paper]. ACM Conference on Fairness, Accountability, and Transparency (FAccT ’24), Rio de Janeiro, Brazil. https://doi.org/10.48550/arXiv.2311.16863
McAdoo, T., Denneny, S., & Lee, C. (2025, 9 September). Citing generative AI in APA Style: Part 1—Reference formats. APA Style Blog. https://apastyle.apa.org/blog/cite-generative-ai-references Accessed on: 2 November 2026.
Messeri, L., & Crockett, M. J. (2024). Artificial intelligence and illusions of understanding in scientific research. Nature, 627(8002), 49–58. https://doi.org/10.1038/s41586-024-07146-0
MSc Geomatics for the Built Environment. (2026). MSc Geomatics policy for the use of AI/LLM (Version 2026.09.01). Delft University of Technology. https://geomatics.bk.tudelft.nl/ai/ Accessed on: 2 November 2026.
TU Delft Library. (n.d.). Manage confidential data: Personal data. Delft University of Technology. https://www.tudelft.nl/en/library/data-management/research-data-management/manage-confidential-data-personal-data Accessed on: 2 November 2026.
TU Delft Teaching Support. (n.d.-a). AI in teaching and learning. Delft University of Technology. https://www.tudelft.nl/en/teaching-support/educational-advice/ai-in-education Accessed on: 2 November 2026.
TU Delft Teaching Support. (n.d.-b). Communicate acceptable AI usage. Delft University of Technology. https://www.tudelft.nl/en/teaching-support/educational-advice/ai-in-education/ai-in-assessment/communicate-acceptable-ai-usage Accessed on: 2 November 2026.
TU Delft Teaching Support. (n.d.-c). AI detection tools usage. Delft University of Technology. https://www.tudelft.nl/en/teaching-support/educational-advice/ai-in-education/ai-in-assessment/ai-detection-tools-usage Accessed on: 2 November 2026.
TU Delft Teaching Support. (2025). AI chatbots in unsupervised assessment. Delft University of Technology. https://www.tudelft.nl/en/teaching-support/educational-advice/assess/guidelines/ai-chatbots-in-unsupervised-assessment Accessed on: 2 November 2026.
Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13, Article 14045. https://doi.org/10.1038/s41598-023-41032-5
The AI Assessment Scale is cited through TU Delft Teaching Support (n.d.-b). TU Delft’s rules on AI are changing: check the links above for the current version.