Katherine Kellen, EdD
Seminole State College of Florida
Adapting Provenance for Student Writing in an AI LLM Age
I teach English, primarily first-year composition online, at an open-door state college. My curricular changes had worked okay-enough in spring 2025. I’d added an “AI LLM Usage Advisory Statement” to the top of my writing assignments, situating each assignment’s learning benefits and explaining if or where an AI LLM might benefit or rob from the learner’s experience. My policy asked students to highlight portions where they’d “danced with an AI LLM” to generate text, but students rarely did so. I’d incorporated Copilot (our institution’s FERPA-promised LLM) into reading assignments, teaching students to prompt it like a “guide on the side” for reading comprehension. Still, my spring 2025 B-term classes had more instances of that icky uncanny valley feeling than my A-term. Student AI LLM usage felt increasingly ubiquitous and decreasingly mentioned.
As I browsed LinkedIn, I clicked on the California Virtual Campus interviewing award-winning photojournalist Santiago Lyon. He introduced me to “The Content Authenticity Initiative (CAI) – Transparency in the Age of AI.”
I learned about this multi-conglomerate collaboration between media platforms and news agencies, ranging from Adobe to the Wall Street Journal. Lyons explained that the CAI grounds itself in provenance, a human practice used for centuries to document items’ origins. The CAI offers an open-source strategy for marking how images and videos came to be—not whether they are trustworthy—called content credentialing. The Content Authenticity Initiative compares content credentialing to a nutrition label, letting users know details relevant to the visual media’s creation.
Although CAI does not provide a provenance system for text, I’d already fallen in love with this application of provenance. I loved that 1) the concept comes from centuries of human innovation and 2) it documents creation processes rather than hunting for illicit behaviors.
Teaching first-year composition at an open-door college, I observe low self-efficacy, motivation, and reading skills in too many students. Online teaching amplifies the ease of unauthorized AI LLM usage and reduces my ability to witness writing generation. In my experience, novice writers sometimes write like an AI LLM. They over-rely on templated rhetoric and echo vocabulary too closely from sources. I didn’t want to accuse someone of unauthorized AI LLM usage who genuinely wrote in a beginner’s imitative style. Just mentioning AI LLM usage could turn writing consultations unpleasant. I’ve been teaching first-year composition for decades; I know how to deliver unflattering news without fully souring our rapport, but the conversations still felt fraught.
Adapting the CAI’s concept of provenance to text for my learners didn’t solve everything or eliminate unauthorized AI LLM usage in my classes. It did, however, blow open how I talk about AI LLM usage in my course, and the changes have been restorative.
In summer 2025, I began my new process by developing a provenance statement of my own. I filmed myself explaining how I designed the course content personally, and that in the few places where I’d used an AI LLM, I documented and explained it. I described provenance, referencing shows like Pawn Stars and Antique Roadshow, and promised that I would not be using an AI LLM to assess or respond to student work. All feedback would come back from me, complete with occasional human error.
Initially, I asked students to use the media recorder in Canvas to explain their writing’s provenance, in addition to my pre-existing AI LLM Usage Advisories. First, they described their writing steps, including brainstorming with a friend, asking a parent for feedback, or using an AI LLM. If they didn’t use an AI LLM specifically, I asked them to share if they turned off or utilized the AI-enabled functions in Grammarly or Microsoft Word. Then, I asked for a short reflection on how their usage impacted their learning, beyond saved time.
I listened to the descriptions of AI LLM usage and writing development from roughly 95 students, for the first five weeks in the term. My mind broadened as students shared several AI LLM use cases that didn’t involve using it to write. They created checklists from my instructions to review their work, pre-graded their submissions against my rubrics, or utilized AI-backed translation tools across alphabets. Other elements of writing collaboration emerged beyond AI LLM usage, too. When students listed where they got guidance, they named parents, partners, co-workers, and older siblings. Even when students used an AI LLM for writing without my recommendation, their transparent descriptions of their process gave me pause; I could see students evaluated writerly decisions and invested cognitive labor. Some students described complex processes, using one AI LLM for one task, switching to two or three different AI LLMs for others.
As the number of writing assignments increased, this method became unsustainable. However, the weeks of listening to students’ descriptions enabled me to draft four levels of AI LLM usage. I switched to students picking which of the four descriptions best captured their usage, along with sharing a provenance and learning reflection.
I’ve tweaked my four descriptions a few times, alone and with a Gemini 2.5 channel I’ve focused on the scholarship of Ellen C. Carillo and Louise Rosenblatt. Increasingly, students post their provenance statements in discussions facing each other, not just where I can see it, demonstrating an appetite for knowing the provenance of the text they read. It’s not about swearing non-usage; some students claim “basic AI LLM usage” and then describe (some of?) what they did. It has become our community practice to share our writing process, including AI LLM transparency. I’ve inferred that some students’ AI LLM usage habits are embedded, so sharing about it is new and requires trust between us for the transparency of usage to grow more detailed.
Admittedly, my provenance statement system crowds my assignment instructions and requires another step in my bookkeeping—if there’s no provenance statement, no points are awarded. I check the assignments each morning and allow late submission without point deduction within 24-hours of the original deadline. With this method, I get provenance statements consistently from all participating students. This decision allows me to have honest, compassionate dialogue with students unhappy with their authentic writing voice, who want to sound more grown up or professional. I share how I worry the AI LLM usage will flatten their voice out of their writing, so maybe use it only in specific situations? I now find myself having nuanced discussions of tone, word choice, audience, and rhetorical situation.
Utilizing provenance broadens the conversation to writing creation instead of AI LLM usage policing. It puts the student writer above all the resources used to create the piece: the instructor, the textbook, the tutor, the older sister, the class peer review process, and yes, often an AI LLM deployment. Students begin to see that much of writing is making decisions, and they share my goal that the AI LLM do only what they decide. In fact, a provenance statement makes students’ practice of noticing their writing decisions explicit. Some students comment that they realize they can do much of what they previously used an AI LLM to do, rolling back their usage a notch. We dialogue about their usage again and again across the term, relating as a team, discussing how to best harness or ignore an AI LLM resource.
Of course, some students lie on their provenance statements, and the incident rate increases across the term. Unscientifically, I share, “Yeah, but it’s the pre-2023 amount.” I’ve never been sure the drafted writing I read came from the person on the roster. Parents do shady things for kids they worry about. Lovers knock out a partner’s paper so the weekend can begin. Some people loathe writing and always have bartered their writing credits rather than earned them. We composition instructors establish our gut instincts through many pages of student prose. Arguably, our practice has created a kind of algorithm in our brains—through repetition, we develop a sense of how an individual writer writes. From that experience, I observe my icky uncanny valley feeling still occurs with this provenance practice, but in those pre-2023 levels, like two or three students out of twenty-five. For me, that’s a significant improvement from spring 2025.
Eventually, I think the tech will rise to meet the moment, offering instructors a kind of user history. The more philosophical, human-sponsored discussion of provenance will be evergreen, though. Documenting the methods by which information got here dovetails not just into information literacy but also workplace practices like CTS coding notations or electronic medical records. Plus, the students and I mostly enjoy it.
Provenance Statement (AI Use Disclosure and Process Statement:):
I’ve shared with colleagues and done some presentations on adapting to AI ubiquity these past few years, so some details have been culled from that material. Otherwise, I wrote this straight into Microsoft Word, with the default writing features enabled. I revised and edited my four descriptions of AI LLM usage with a Gemini 2.5 channel I trained on the scholarship of Ellen C. Carillo and Lousie Rosenblatt.
Provenance Statement Levels for Students:
Select the level of AI LLM usage that best describes your usage in this submission:
Passive AI LLM Usage:
- Aside from standard grammar/spell-check functions available in Word/Google Docs/Grammarly that I did not turn off, I did not seek AI LLM guidance.
AI LLM for Basic Support:
- I used an AI LLM (e.g., ChatGPT, Copilot, Gemini) only for assistance such as instruction clarification, to pre-grade my own writing with the assignment rubric, for minor grammar/punctuation checks beyond standard tools, MLA parenthetical citation or Works Cited assistance, or very occasional word choice suggestions resulting in no more than three words in a row. Essentially, my AI LLM usage is minor and localized. I did not use an AI LLM for any significant drafting or content generation.
AI LLM to Get Started Writing:
- I used an AI LLM for more substantial assistance, such as outlining, developing initial ideas, generating several possible phrases/sentences that I then heavily revised into my own written expression, or using an AI LLM for more extensive editing of my own drafted text. Like the Basic Support described above, I have limited any direct AI LLM word choice suggestions to no more than three words in a row.
AI LLM Usage for Drafting:
- A significant portion of this submission’s text was drafted or generated by an AI LLM, and I then transformed and revised it, demonstrating critical engagement with the AI LLM output, to create my final prose. (For Drafting level of AI LLM usage, please submit the AI LLM original dialogue for me to contrast with your revised and edited submission. This helps me see your cognitive contributions for assessment. Thanks!)
Describe your writing process for this assignment, focusing on the steps you took to produce your prose. Be sure to address the following:
- Resources & AI: What resources did you use at different stages of your writing (e.g., outline, rough draft, revision, editing) including, but not exclusive to an AI LLM? Share the apps you trust, if you ask a friend for feedback or took advantage of librarian or tutoring feedback, like “I checked with Mr. Martin” or “My trusted friend gave me feedback at work,” etc. If you used an AI LLM, specifically identify which one(s) and when and how you integrated it/them into your process, such as, “I used [AI Tool] to brainstorm ideas before I started writing,” or “I drafted my paragraph, then asked [AI Tool] for alternative sentence structures for these two sentences.” Specifically detail the steps and resources utilized in your process producing this work.
- Your Learning & Decisions: Explain what you learned or what insights you gained about your own writing or the topic by using these tools (or by choosing not to use AI). What was your reasoning for the choices you made about tool use in this assignment? (Efficiency/time saving can be mentioned–it’s a reality for all of us–but be sure to emphasize what you learned through the process you chose.)