15  Using AI Tools

(Generative) AI tools are increasingly part of the academic workflow and will continue to be implemented in even more diverse ways. The Department of Communication Science at VU Amsterdam generally permits the use of these tools, provided you use them responsibly, ethically, and transparently! What constitutes such use, depends entirely on the nature of your research and the specific type of AI use you are planning to implement.

WarningKey problem: The stochastic parrot

Current AIs, such as chatGPT, Gemini or Claude, are based on an architecture called transformers, which are - simply speaking - large prediction machines. Based on an input (e.g., a question), they deliver the most likely answer. They do not “know”, “reason” or “think” in the human-sense. Their answer thus entirely depends on their training data and word similarities, which can be misleading and plain wrong. Don’t be fooled: these tools may output text that sounds coherent or logical, even convincing, while being based on dubious sources or completely hallucinated!

ImportantThe core principle: Human in the Lead

You are always the one who decides on the content. You are responsible for the quality and originality of your work. Generative AI cannot do the thinking for you — and it should not!

15.1 What is responsible use?

Responsible use means:

  • You decide on the content at all times
  • You are critical of AI output — AI is not a source of truth; it can produce inaccurate, incomplete, or superficial suggestions
  • You understand what you are doing — use AI tools only when you understand the topic well enough to evaluate, improve, or reject what the tool suggests
  • You take full responsibility for the final submitted version

Generative AI can be a useful sparring partner for brainstorming, gaining an overview, structuring your work, or improving text you have already written. It cannot replace the reading, thinking, and writing that constitute your learning process.

There is a deeper reason to stay actively involved in research tasks beyond academic honesty: it is how expertise develops. Searching and reading literature builds a map of what is known and unknown in your field. Cleaning and checking data builds intuition for data quality. Coding qualitative material builds understanding of your participants’ experiences. These tasks may seem routine, but they are precisely how you develop the judgment needed to recognize when something is wrong — and to evaluate whether AI output is accurate (Messeri & Crockett, 2026). Students who delegate these tasks to AI risk not developing the competence to supervise it responsibly.

The following examples illustrate what responsible and irresponsible use look like in practice:

Scenario Assessment
You use Gemini to brainstorm search keywords, then search and read the sources yourself Responsible
You ask ChatGPT “what are the main theories on social media and well-being?” and include those theories without reading the original sources Not responsible — AI can hallucinate or misrepresent findings
You paste an R error into Copilot, read and understand the suggested fix, and verify it works Responsible
You ask Gemini to “write the methods section based on these notes” and submit it with minor edits Not responsible — you are not the author of that text
You upload your own PDFs to NotebookLM to identify sections worth reading more carefully Responsible
You ask an AI to summarize papers you have not read and cite them in your theory chapter Not responsible — you cannot vouch for what you have not read

15.2 Academic misconduct and fraud

  • Never copy-paste AI-generated text into your thesis without attribution. This is plagiarism!
  • Using AI without transparency is academic fraud
  • If I doubt whether you have used AI responsibly, I will discuss this with you; persistent doubts go to the Examination Board

More information: vu.nl/en/student/examinations/academic-integrity

VU rules on generative AI: vu.nl/en/student/examinations/generative-ai-your-use-our-expectations

15.3 VU-licensed vs. non-licensed tools

15.3.2 Non-licensed tools (use at your own risk)

Tools like ChatGPT, Elicit, QuillBot, and others fall outside VU’s license. When using these: - Your data may be used for model training - You are personally responsible for privacy and data protection - Never upload personal research data (names, email addresses, interview transcripts, survey data) - Never upload paywalled articles — this violates copyright

WarningGDPR and data protection

Never (!!!!) upload sensitive or personal data (e.g., data you collected from participants) to any AI tool, not even VU-licensed ones. This includes: names, email addresses, research data, financial records, or company details from internships or partnerships.

THERE IS NO EXCEPTION TO THIS!

15.4 Responsible AI use by thesis phase

Thesis Phase Responsible Use Useful Tools
Topic selection Brainstorming ideas, identifying research gaps, but always verify by cross-checking with relevant literature Gemini, Elicit, Perplexity
Literature review Finding keywords/synonyms, searching articles (although this is very limited and cannot replace a classic literature search), generating summaries (always verify based original source material) Gemini, Elicit, Scite, Research Rabbit, NotebookLM, Undermind, Perplexity
Research design Generating ideas, creating experimental stimuli (justify all choices yourself), cross-validate, develop items (evaluate carefully) Gemini, Qualtrics AI
Proposal writing Feedback on text you wrote; rephrasing, shortening, grammar Gemini, Grammarly, Wordtune
Data collection In some cases, automating data collection (verify quality yourself, implemented validation procedures if possible) Gemini, Google Forms
Data analysis Generating/debugging R code, creating visualizations (verify all output), automated content analysis… Gemini, Orange3
Writing the thesis Improving clarity and structure of your own text Gemini, Grammarly, QuillBot, Wordtune
Revision & feedback Improving academic style, tone, consistency Gemini, Grammarly, Hemingway Editor, Trinka AI
TipSustainability

Every query to a generative AI system requires significant computing energy and thus has a large environmental impact. Use these tools mindfully and purposefully!

15.5 The AI statement (required)

All theses must include a Statement about Generative AI and AI-Assisted Technology usage at the end of the thesis, after the reference list. This statement is not counted in the word count.

Your statement must answer:

  1. Did you use generative AI? (yes/no — if no, the statement ends here)
  2. Which tools did you use, and how? — for each tool, specify the purpose (e.g., “I used Google Gemini to receive feedback on paragraph structure in the introduction and theory sections”)
  3. How did you handle the AI-generated content? — end with a statement taking responsibility for the final work

15.5.1 Example AI statements

“During the process of writing this thesis, I used Perplexity to explore the scientific literature. I always verified the information received via AI searches by reading the original sources. I also used Google Gemini to receive feedback on my own writing regarding paragraph structure and academic phrasing. The final formulations were written and checked by myself. I take full responsibility for the final submitted version.”

“When writing the theory sections, I used Grammarly to improve readability and language. I also used Copilot to debug errors in my R code during analyses. When using these tools, I carefully reviewed and edited the generated content. I take full responsibility for the content of this thesis.”

15.5.2 In-chapter disclosure

If you used AI for specific chapter-level tasks, also note this within the relevant chapter. For example:

  • Theory section: used AI to generate an initial reading list on a topic
  • Method section: used AI to generate experimental stimuli
  • Results section: used AI to generate or refine analysis code

Keep a record of your AI prompts and outputs — you may be asked to show these.

15.6 References

  • Messeri, L., & Crockett, M. J. (2026). The uncritical adoption of AI in science is alarming — we urgently need guard rails. Nature, 653, 675–676. https://doi.org/10.1038/d41586-026-01557-x