#18. An AI Meta-Analysis of Postpostmodern Compositionist

In my last blog post, I mentioned that the next two blog posts would feature articles from Composition & Resistance. Those posts are definitely still coming, but I had an interesting idea I have been thinking about over the past few months, and this week I couldn't help myself in deciding to move it to the top of the docket. At the end of the day this is my blog, so I change the rules whenever I want, right? 

Trust me, I have no illusions. I am aware that my blog doesn't get much traffic, if any at all. However, I did notice that Blogger's country-tracking feature shows visitors from over 18 countries, including Kenya, Turkey, and Vietnam. I am not sure how these people are finding my blog or why they are interested in it (perhaps they are rhetoric & composition scholars?), but I also suspect that they might be bots who are scraping the internet for text they can use to train AI models. 

Aside: This is a direct comment for people who might be reading this: If you feel so inclined, leave a comment on this post making your presence known. I genuinely would be curious to know if sentient beings are actually reading this blog. If I see no comments, I am going to assume that these views are simply from bots. Hopefully, I am not just talking to the wall. 

Okay, back to the point about my writing being used to train AI models. No matter how obscure this blog is, it is still publicly available, and I understand that I have no control over what people (or bots) may be doing with my writing. I imagine that this is simply part of maintaining a digital presence in 2026. Again, whether bots are actually scraping my blog posts is complete conjecture. I have no idea of knowing whether they are or not, and perhaps I am paranoid and the views are actually organic views. Regardless, for this reason I have no qualms about running my blog posts through AI, so I thought I would conduct an experiment. 

I copied and pasted all of my blog posts into a Word document and input them into three large language models: 1) Google Gemini (3.6 Flash), 2) OpenAI's ChatGPT (Auto mode), and 3) Anthropic's Claude (Sonnet 5). Along with the blog posts, I also input the following prompt to direct the models: "Here are all of the posts from my scholarly blog. Please identify three patterns you notice throughout these posts. For each pattern, quote or paraphrase small pieces of my writing as evidence (which you will only draw from the text I provided). Limit your response to three short paragraphs." It is important to note that I am using the free versions of all these chatbots. 

One might ask, Casey, why do this? What could AI possibly tell you that you can't figure out yourself? Why give your data freely to these large AI companies? These are all fair questions, but as someone who is becoming a qualitative researcher and who has received a robust methodological training in qualitative methods over the last two years, I have been sold on the affordances of AI as an effective tool to conduct qualitative analysis. I do think that AI does a fairly good job of detecting patterns and trends across a large corpus of texts, and many times when I have run an analysis of my writings through AI, it does allow me to get a fresh perspective and consider blind spots. However, I will emphasize that qualitative analysis is a subjective process and that while AI analysis can augment human analysis, it should never replace it. Though most AI models have come a long way since 2022, in my opinion their outputs often can still feel unremarkable and generic. Again, the limited perspective of a large language model should ideally just be one amongst a variety of other human perspectives. Without much ado, let's get into the AI outputs. 

1. Google Gemini

See Image 1 for Gemini's output. Out of the three large language models, I thought this output was the most idiosyncratic, in that its points had less crossover with the other models. With this said, the first finding is rather unsurprising, as the whole point of this blog is to consider past composition texts against the present day. I did not input the description of my blog and perhaps I should have. At the same time, I do feel that it assuages my worry that I deviate too much from the overall concept of the blog. It appears as if I am mostly conforming to the blog's mission and this is comforting to hear. 

Image 1: Output from Google Gemini

The second point about nostalgia was genuinely intriguing to me and isn't something I would have thought of on my own. I do agree with the notion that this blog does get nostalgic sometimes. Funnily enough, when I was starting the blog I had the idea of calling it "The Nostalgic Compositionist," but decided against this name because I do see nostalgia as being somewhat of a problematic emotion. Maybe problematic is the wrong word, but I do think that too much nostalgia can prevent one from fully living in and appreciating the present. Perhaps it is okay in small doses. Regardless, I still enjoy reading old composition texts of yore, and nostalgia or not, I am going to keep on doing it. 

I thought the last point was maybe the most specific takeaway. There definitely is an undercurrent in my blog so far where I wrestle with the legitimacy of critical pedagogy as a functional approach to teaching writing. However, I did find the connection between critical pedagogy and capitalist realism to be very promising, though the explanation provided by Gemini isn't super thorough. Perhaps my identity as a capitalist realist, which explains my larger political views, somehow informs my skepticism of critical pedagogy? I am not sure, but if I was using AI as a brainstorming tool here, this is an example of a glimmer of an idea that I could take and run with if I was preparing to write a larger piece. 

2. OpenAI's ChatGPT

See Image 2 for ChatGPT's output.The first takeaway from ChatGPT is similar to the first takeaway from Gemini, so this is more verification that I am serving the blog's purpose. One small difference would be that ChatGPT's iteration is more generally about the present, whereas Gemini's is solely connecting my analysis to the current phenomenon of AI. The second point is interesting, and in previous instances running my blog posts through AI, when asking for feedback I have been told that my claims and opinions are too wishy washy and that I'm not making clear cut arguments. Again, I am not too concerned about this, because my blog functions more like a diary rather than an editorial column, though I would argue to the contrary that there are some posts where I take clear and undeniably firm stances. 

Image 2: Output from OpenAI's ChatGPT

I do think there is some crossover between ChatGPT's third point and Gemini's third point. If I was going to identify a difference between the two models so far, ChatGPT does seem to be making more macro-level observations, as here it is focused dually on my musings about critical pedagogy and AI. Gemini was only focused on critical pedagogy, and perhaps this tells us that Gemini is more inclined to have a narrower analytical focus? 

3. Anthropic's Claude

See Image 3 for Claude's output. I will admit that Claude is my preferred tool for most of my AI usage and thus I had high hopes for its performance. Surprisingly, I will say that its output was my least favorite of the three. The first point is similar to ChatGPT's second point, however unlike ChatGPT's more neutral tone, I did feel that Claude here almost comes across as unhappy with the quality of my writing and that it is subliminally trying to offer me feedback. Maybe it's just me, but I get the sense that the framing of this point -- the notion of "uncertainty and revision-in-progress" -- comes with somewhat of a negative connotation. I do often use Claude for writing feedback, so I am wondering if perhaps my personal model has been trained to be more critical. I don't have much to say about the second point, as this theme was detected by all three models and Claude doesn't add anything new. The third point feels similar to Gemini's second point about nostalgia, but in Claude's case, it seems to be emphasizing my tendency to bring personal experiences into my writing. 

Image 3: Output from Anthropic's Claude

Overall, if I had to rank the outputs from best to worst, I would probably say Gemini was my favorite, with ChatGPT being a definite second and Claude being in last place. As I mentioned previously, this might not be completely fair to Claude, because it is the model that I use most for personal use, and this may be influencing its output to be more critical. Overall, this was a fun experiment and it gives me some things to think about as I continue writing this blog. Should I cut down on the nostalgia? Should I figure out what exactly my beef is with critical pedagogy? Who knows. It wasn't my intention to psychoanalyze myself. Until next time!