About This Site

Douglas Eyman


Since the release of ChatGPT in November 2022, there has been a significant amount of interest in large language model (LLM) text generation tools—and I found myself following all of the new and innovative uses, the hype, the outrage, and the humorous responses produced by these so-called AI systems. You'll note that I don't use AI terminology here, because these systems are certainly not intelligent, and I think it is quite easy to use terms like "it thinks" or "it understands" in reference to, for instance, ChatGPT output—but it can perform neither of those actions, and we users would do well to avoid anthropomorphising these sysetems. (For a good overview of how these systems actually work along with a critque of LLMs in general, refer to Bender et al., 2021, "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜")

There are a number of reasons to take a principled stance agains using text generation systems at all, from the black-box nature of the systems that conceal the training sources (and any biases that may be intrinsic to such sources; Mir & De Blanc, 2023), to the environmental effects of running such computationally heavy systems (which in turn use a tremendous amount of energy; Saenko, 2023), but I believe that simply denying or resisting these technologies will be as effective as calls in the late 1980s to not allow students to use computers in their writing classes as doing so would somehow "dehumanize" the experience of writing. My aim in this Disputatio webtext is not to interrogate or investigate LLMs but rather to see what it would be like to put one to use in order to see if this new technology might actually be able to help prospective authors produce webtexts for Kairos.

One of the challenges of editing Kairos is that our authors are often self taught when it comes to the technological infrastructure needed to produce a high quality webtext that is both accessible and usable (and that meets our requirements for using design to forward the argument of the work). My own first publication in Kairos 1.2 (Eyman, 1996)features some terrible coding choices as well as design choices that interfered with the usability of the piece—some 20-odd years later, I'm much better at coding (although still not particularly adept with design), but because the technologies we rely on continue to evolve, there is still always a bit of a learning curve each time I start a new web project based on HTML, CSS, and JavaScript (the three key plain text and open source technologies we recommend as the foundation for any Kairos webtext). When Kairos editors give workshops on how to conceptualize and produce multimodal scholarship, one of the most common questions from participants is asking for advice on how to learn the technical skills needed. We typically suggest a number of resources for learning the basics, but we also recommend finding a collaborator who is versed in programming and/or design to bring on as a coeditor (or as a designer, a designation we sometimes include as separate from "author"). If ChatGPT and other text generation systems are poised to become assistants and collaborators for a wide range of writing tasks, I wondered whether these systems could be leveraged to help authors with the coding and design work of crafting a webtext. To answer this question, I asked ChatGPT to help me build this webtext.


First, I decided that the content for the webtext could also be produced by ChatGPT, so I provided the following prompt:

Please write an essay about how ChatGPT and generative AI in general can help writing studies scholars produce multimedia essays.

I copied the output, including the title provided by ChatGPT, and then selected the "Regenerate" option to see what different approaches it might produce—I did this three times (thus ending up with four versions, including the response to the original prompt). These responses are reproduced under the "Content" link; following the content are links to the additional regenerated versions.

These essays each follow a fairly general structure, close to a traditional five-paragraph essay, with an introduction, three to five body paragraphs, and a conclusion. While quite general, the suggestions for how ChatGPT can help scholars produce multimodal scholarship are not inaccurate and there are some good suggestions presented—as an editor, I would be comfortable with potential authors considering the content produced by this prompt as a first step in considering how and why to produce a webtext (if such were not the case, I would not have included this content here).

The essays are not particularly scholarly, however, so I asked ChatGPT to

Please add citations to your response.

to which it responded

I apologize for the inconvenience, but as an AI language model, I don't have direct access to external sources or the ability to provide citations. However, I can offer general information and insights based on my training data. You can refer to relevant scholarly sources on writing studies, generative AI, and multimedia essays to obtain specific citations to support the points made in the essay.

I elected to continue using just ChatGPT for this exercise rather than exploring other LLM-based systems in order to keep my inquiry focused but also because the text-generation systems that will add citations have a well-documented history of inventing nonexistent citations that sound reasonable (indeed "sounding reasonable" in terms of language produced is the sole ability of these text-generation systems, since they can only provide statistically likely output and have no way to take into account context or audience).

I'll note here that I didn't engage in "prompt engineering"—that is, continual refinement of the prompt to lead to better output; it's possible that higher quality content (or coding or design advice) could have been coaxed from the system, but I was more interested in immediate results. And, after all, if getting good results requires learning a whole new skill, it undermines the premise that ChatGPT could be helpful for users who are not versed in the technical skills required to create a webtext.

Structure (Code)

With the content generated, I next turned to the HTML code—I see the HTML as providing the overall structure for a webtext, whereas CSS handles design. Before generating the actual code, I asked ChatGPT to

Please design a website to present advice written by Chatgpt for an audience of digital media scholars

—this advice is available under the "Structure" heading of this site. Immediately following this query, I asked,

Can you produce basic html for this site?

and ChatGPT immediately output a general HTML template I could copy and use (all of the HTML on this site comes from ChatGPT). You can see a screenshot of this output in the structure section, or use your browser's "view page source" option to see the HTML at work.

ChatGPT reminds me that this is just example code, and I'll need to develop it so it meets my needs.


With the HTML set, I turned to issues of design. As with the structure, I first asked ChatGPT to make suggestions about design decisions (before jumping right into the CSS code). Because this is a webtext for the Disputatio section of Kairos, I asked,

Using kairos.technorhetoric.net as a model, what css design would you suggest?

and ChatGPT provided a series of suggestions (available on the "Design" section of this site). I also asked,

What fonts should I use for this website?

And to get an alternative option, I followed up with

In terms of fonts—what if I want something more edgy?

I then asked a few more design questions:

What colors would look good for this site?

What kind of images would look good for this website?

After which I requested the CSS code:

Please write css for the site

I decided I wanted a bit more, so I asked it to

continue the code

When I put the HTML and CSS together with the content ChatGPT provided, the result wasn't very pretty (or readable), so I iterated a series of requests to try to improve the design:

please add padding to css elements to make the text more readable

please add search styles for nav

please generate css for responsive design

please provide a strong style for the banner text

please provide css for more attractive and readable unordered lists

please provide css for more attractive and readable ordered lists

The solution ChatGPT provided for the ordered lists collided with the CSS for unordered lists and made a bit of a mess. So I asked ChatGPT how to fix it:

This code adds numbers to nested unordered lists within the ordered lists. How do I fix that?

...and ChatGPT provided a CSS fix that did not work. Twice. The final solution it provided ultimately required me to hard code the numbers for the list items (completely defeating the purpose of automatically incrementing list items) and adding excessively complicated CSS that could have worked just fine as a regular ordered list without all the weirdness ChatGPT provided.


The site was looking quite plain at this point, although the HTML code did include a placeholder for a banner, which I thought should be an image. First I asked,

What size should my banner image be?

and then followed up with

What kind of images should I include in my website?

The answer to the above question was excessively generic, so I reframed it as

What specific images should I include in a site that aims to help academics create a multimedia webtext?

Finally, I asked,

What social media icons and links should I use in the footer of my website?

With these answers in mind, I requested some code to help with the banner:

Please produce HTML to display a row of images three across.

ChatGPT helpfully also included the CSS for the row.

I used OpenAI's DALL-E 2 and Stability.ai's DreamStudio to produce images to use for the banner. My initial prompt for DALL-E 2 was

An AI bot helping researchers write multimedia texts

which produced only images of researchers who were men. So I followed up with

An AI bot helping women researchers write multimedia texts

...some of which were a bit problematic. DreamStudio did a little better when asked to produce an image of

A futuristic AI robot helping academic researchers

as the results were certainly more diverse than the images produced by DALL-E 2.

I also had DreamJourney produce the "social media icons" used in the footer of this webtext.

Final Edits

I checked the responsiveness of the site and found one key issue, which I asked ChatGPT about:

when I use a smaller screen, the banner images are not shrinking. How do I get them to scale properly?

It provided some CSS, but that wasn't quite working, so I followed up with

How do I make the banner image containers responsive?

The CSS response to that query did work reasonably well. I finished up with a few tweaks but decided that I didn't want to fine tune the site further in order to demonstrate basic capabilities without excessive tinkering (which in turn requires some knowledge of the underlying technical skills):

How do I center an image in css?

how can I horizontally center an image without using flex?

how can I make the navigation links more visible?


One of the features I really liked about how ChatGPT provided help with code was that it would follow the code output with an explanation of how it worked, along with cautions that these are generic examples that would need to be adjusted or developed more fully. However, the issues with lists shows that ChatGPT may provide incorrect or excessively complex code that might be quite frustrating for someone without a strong grasp of the mechanics of CSS and HTML to attempt to fix. And the final result was itself quite generic, and ChatGPT certainly did not help with the difficult conceptual work of using design to forward an argument. ChatGPT might be helpful as a learning tool, and for providing basic coding as a starting point, but will certainly not be able to actually develop a full-fledged scholarly webtext without a significant amount of training and/or prompt engineering (so much so, I expect, that it would be more efficient to not use it beyond the first iteration of the basic HTML/CSS template).


  1. Bender, Emily M., Gebru, Timnit, McMillan-Major, Angelina, & Shmitchell, Shmargaret. (2021). On the dangers of stochastic parrots: Can language models be too big? 🦜. In FAccT '21: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (pp. 610–623). ACM. https://doi.org/10.1145/3442188.3445922
  2. Eyman, Douglas. (1996). Hypertext and/as collaboration in the computer-facilitated writing classroom. Kairos: A Journal of Rhetoric, Technology, and Pedagogy, 1(2). https://kairos.technorhetoric.net/1.2/binder.html?features/eyman/index.html
  3. Mir, Rory, & De Blanc, Molly. (2012, January 18). Open data and the AI black box. Electronic Freedom Foundation. https://www.eff.org/deeplinks/2023/01/open-data-and-ai-black-box
  4. OpenAI. (2023a). ChatGPT (May 24 version) [Large language model]. https://chat.openai.com/chat
  5. OpenAI. (2023b). DALL-E 2 [Text-to-image generator]. https://labs.openai.com/
  6. Saenko, Kate. (2023, May 25). A computer scientist breaks down generative AI's hefty carbon footprint. Scientific American. https://www.scientificamerican.com/article/a-computer-scientist-breaks-down-generative-ais-hefty-carbon-footprint/
  7. Stability.ai. (2023). DreamJourney [Text-to-image generator]. https://dreamstudio.ai/generate