Cognitive Sovereignty Self-Audit for University Lecturers

1. When you design essay questions or assignment briefs, how much do you check whether AI can produce a credible submission without actual understanding?

I design assignments the same way I did before AI. I assume students will do the work themselves. I occasionally test an assignment prompt with ChatGPT to see what happens, but I don't change my design based on results. I regularly test assignments with AI tools and adjust prompts or assessment criteria when AI produces too-credible outputs. I systematically design assessments that require reasoning visible in the work itself. I use AI testing as one input, but I also test for conceptual gaps that AI would miss.

2. When you mark student work, how often do you find yourself uncertain whether the thinking shown is genuinely the student's or AI-assisted?

Frequently. I suspect AI involvement in many submissions but cannot prove it, and my uncertainty makes marking harder. Sometimes. A few submissions each term have that distinctive quality that makes me question their origin. Occasionally. I have developed instincts for spotting inconsistencies in argument or false confidence in minor details, which helps me identify suspicious work. Rarely. My assessment tasks make the source of reasoning visible. I can trace how a student arrived at conclusions, whether the work is strong or weak.

3. When you conduct literature reviews for your own research, how much do you rely on AI summaries of papers you have not read in full?

I use Elicit or Semantic Scholar AI to summarise most papers. It is faster than reading full texts. I use AI summaries as a first filter but read full papers for studies that seem central to my argument. I read most papers in full and use AI tools only to extract specific data or check whether I have missed key methodological details. I read all papers I cite fully. I use AI tools only as secondary checks after I have formed my own judgement of a paper's contribution and limitations.

4. How do you currently decide whether to accept an assignment submission that shows signs of AI use but might still contain genuine student work?

I apply a zero-tolerance policy. Any AI use means automatic failure, regardless of what the student actually understood. I have a blanket policy set by my department. I follow it without adapting to individual cases. I use my institution's academic integrity framework as a guideline and hold conversations with students about what they did and why. I distinguish between different kinds of AI use. I assess whether the submission demonstrates the skills the assignment was designed to teach, regardless of the tools used, and make decisions based on that judgement.

5. When you prepare lecture content or teaching materials, how much do you rely on AI to generate or adapt examples, case studies, or explanations?

I generate most of my lecture examples using ChatGPT or Claude. It saves time and keeps materials current. I use AI to draft some materials, then edit them substantially to reflect my perspective and emphasis. I occasionally use AI tools to explore how to explain a difficult concept, but I always rebuild the explanation from my own understanding. I create teaching materials from my own knowledge and experience. I do not use AI to generate content I will then pass to students.

6. How confident are you that your current grading rubrics or marking criteria would catch an AI-generated submission disguised as student work?

Not confident at all. My rubrics assess final outputs, not the reasoning process, so AI work could meet all my criteria. Somewhat unconfident. My rubrics focus on content and structure, but I would struggle to distinguish polished AI output from strong student thinking. Moderately confident. My rubrics include criteria that require students to show their reasoning steps or engage with feedback I gave earlier in their learning. Very confident. My assessment criteria explicitly measure whether a student can justify choices, respond to challenge, or build on prior feedback. These are difficult for AI to fake credibly.

7. When you encounter research that makes you unsure whether findings are robust, how do you investigate that uncertainty?

I ask an AI tool like Perplexity to explain the findings. If it sounds convincing, I move forward. I search for summaries or simplified explanations online. I read the abstract carefully. I review the paper's methods section and check for obvious flaws or limitations. I scan related studies to see if findings are consistent. I read the methods and results sections carefully. I trace the authors' reasoning from data to conclusions. I look for alternate explanations and check whether related work supports or challenges the claims.

8. How much are you actively rethinking what skills you need to certify in your students, given that AI can now perform many tasks your degree was designed to teach?

I have not rethought my curriculum or learning outcomes. My course design reflects what I was taught to teach. I have considered it but I am uncertain what to change. I continue mostly as before while being aware the situation is shifting. I have updated some assignments or assessments to focus on reasoning, judgement, or tasks where AI cannot substitute. But I have not yet redesigned my whole course. I have deliberately redesigned my course outcomes to prioritise reasoning, critique, and synthesis. I have rewritten assessments so that they measure whether students can think, not whether they can produce credible outputs.

Cognitive Sovereignty Self-Audit for University Lecturers

Cognitive Sovereignty: How To Think For Yourself When AI Thinks For You