The growth of template rhetoric in an AI society

Everyone seems to be noticing how bad AI writing is. AI-hypester Ethan Mollick seems to have finally realized it:

I don’t follow Mollick, a professor who studied crowdsourcing before turning to GenAI. But I was shocked that he is only now (in May of 2026) realizing how insufferable online spaces have become after being infested with synthetic AI content. Mollick isn’t the only one. This blog post from Florian Roth a few months ago mentions how every newsfeed is starting to sound the same:

Roth’s post articulates what a lot of people seem to be feeling (as it has over 9K likes on the old standby blog site Medium). These posts are articulating what I’m going to call template rhetoric. (Disclaimer: I wrote an entire academic book that hinges on this concept. It’s free to read the PDF.)

Let me explain.

What would you think if I started this blog post with the following sentence:

“This essay is going to delve into the cultural, political, and economic implications of how AI is templating our society—that is, the ways that AI text is structuring people’s lives.”

I imagine you’d first say it was AI written. It’s not but I tried my best to emulate all of the most obvious AI signals: the word delve, the tripartite list (“cultural, political, and economic implications”), and certainly the em-dash. There are so many AI writing signals that Wikipedia has a page on it now. The writing all feels like templates now, no? Those stylistic features are just the surface of that template.

Notice the subtle, tacit lack of genre awareness of the sentence. When writing that indented sentence, I referred to this blog post as an essay. An essay signals something entirely different: a formal, likely formulaic, piece of text. The term “essay” invokes your English classes from high school or college. There is no use of the first person singular (“I”) in the sentence either; first person singular is a hallmark of a blog’s genre conventions. A good blog is not an essay. I’m allowed to have asides, meander a bit, in a blog.

The missed genre signals (“This essay…”) end up, too, scrambling audience awareness. When I’m writing this blog, I’m functioning as a blogger, which is a strange term nowadays, invoking a halcyon sense of the internet when it wasn’t dominated by dull web design, platform optimization, or AI content. A blogger gets to have an extra weird (dare I write, using my favorite SAT word, addlepated) sense of audience. A blog, after all, is a portmanteau of web-log. That means that it’s essentially a journal, typically with unseen, unknown audiences (unless you, dear reader, leave a “like” or comment). I’m supposed to write in first person. I’m not writing an essay for an English teacher. My English teachers would make me redo this blog post. To wit: A blog gets to be confessional, meaning I get to disregard my audience in many ways. Of course, we all know that a blog is a public web-log, a very strange audience for whom I must work to keep engaged.

And this brings me back to that indented sentence: it’s so shoddy because it’s a template. It restates the first half of the sentence in the second half (fake contrast, to echo Roth’s observations). Notice that the phrase “templating our society” is basically the same thing as “structuring people’s lives.”

People across the Internet (indeed I capitalized it) are starting to realize GenAI machines write rehashes of the same tired templates. We’re reading the skeletons of writing without any of the muscles, the nerve fibers. The parade of buzzwords (“pivotal” or “intricate tapestry”), the construction “Not X but Y”, the lists of three, the plague of commas, and nominalizations (where verbs get turned into nouns) all signal the grammar and sentence structure are better than the ideas within the grammar and sentence structure.

These templated elements, of course, aren’t inherently a signal of AI writing. That’s the problem that these templates make you feel: these features were once human. Everything an AI does is because it learned it somewhere in its human training data. AI writing overuses (technically overindexes) these features because GenAI machines are not trying to communicate; instead, they are processed text, synthetic text derived from mathematical models. The text that emerges is templated, a prefabricated form that is designed to be filled-in and filled-out.

What is a website, anyways?

When we notice these constructions, get annoyed with them, we’re noticing the template rhetoric of large language models (LLMs). In order to add some detail about what I mean by template rhetoric, let me use the concept of a website. What does a website look like to you?

Take a pause now. Imagine a website. (Go ahead. I’ll wait for 30 seconds.)

I’m betting what you imagined was something akin to the beautiful interfaces of whatever site you visit the most, that is, something like YouTube, TikTok, or maybe a news site. Once upon a time, though, the Internet didn’t look at the same. The World Wide Web was once a fertile ground of very oddball, wide-ranging websites (Geocities anyone?). Before corporate platforms enclosed the internet—in the same way that England enclosed its lands during the formation of capitalism—websites weren’t templated. People designed screwball, idiosyncratic websites, often with bad color palettes. Now, though, only the website designs of the big platforms exist (and builder sites like WordPress that, ironically, offer only banal templates). The same thing that happened to web design is happening with writing. Platforms templated web design in the way that LLMs are templating writing itself.

Let me draw the relationship into the nerve root of the problem. Most regular users of websites misperceive what a “website” actually is. Think of the main page of Wikipedia. That link to the main page of Wikipedia is not entirely how the website looks. That’s the front end of Wikipedia. The back end of the website looks something like this:

<noinclude>{{SHORTDESC:Main page of the English Wikipedia}}</noinclude><templatestyles src="Wikipedia:Main Page/styles.css" />
<div id="mp-topbanner" class="mp-box">
<div id="mp-welcomecount">
<div id="mp-welcome"><h1 class="html-heading">Welcome to [[Wikipedia]]</h1>,</div>
<div id="mp-free">the [[free content|free]] [[encyclopedia]] that [[Help:Introduction to Wikipedia|anyone can edit]].</div>
<div id="articlecount"><ul><li>[[Special:Statistics|{{NUMBEROFACTIVEUSERS}}]] active editors</li> <li>[[Special:Statistics|{{NUMBEROFARTICLES}}]] articles in [[English language|English]]</li></ul></div>
</div>
</div>
{{#if:{{Main Page banner}}|
<div id="mp-banner" class="MainPageBG mp-box">
{{Main Page banner}}
</div>

Most of you probably know the above text is HTML. When we imagine websites, we don't see a website like this. We don’t see the templates of the various tags (<div>). We just get a (hopefully) pretty website.

Strangely enough, in my work, I see a lot of website templates. I’ve been working with web-scraped data for a long time. What do I mean by web-scraped? I mean data that is just the raw information on website without any formatting, information as it is stored on the backend. For that’s what a web-browser does: it translates the backend, developer database of information into a pretty website. There is a deeply interesting thing to learn here: our perceptions of websites are not really what the website looks like. Even the HTML isn’t accurate. Below is a screenshot of web-scraped data (online comments) from my research:

A website doesn’t even exist as we typically interact with it until we open the browser. Until we do, the website exists as above, without formatting or aesthetics. Where the analogy with websites breaks down is that it’s often easy for us to ignore the bland repetitive formatting of web design. No one seems to find TikTok or YouTube incredibly boring besides me, despite the awful color and typography, or other felonious web design choices.

All of this is connected to the template rhetoric of LLMs because LLMs do not produce writing in the way a human does, that is, thinking through an argument with their sense of self within the world. LLMs produce sequences of token shaped by statistical patterns in training data, patterns that we can loosely call templates. An LLMs doesn’t know that the word nectarine corresponds to the thing itself in the world, which can be a juicy stonefruit or a mealy overpriced lump. It only sees N-E-C-T-A-R-I-N-E as a string of characters related to other strings. It sees words close to other words; it only sees writing as templates. Nothing is left but templates. With an LLM, there is no connection between text and object. When we (human beings) read the text of an LLM-produced paragraph, we feel something is off because only a template remains. That’s why we’re so bored. We’re drowning in machined textual templates that aren’t filled in with human thought.

Scripts & forms

The Internet did not have to look like it does. It’s just become accepted that web design is a boring, uninteresting commodity foisted upon users. Web design has become a script, a form, to fill in and fill out. The same thing happening with AI writing, which is why people across social media are complaining about the templates they’re seeing repeated over and over again.

I have a broader concern, something I’ve begun calling “the templating of the American mind” or just the scripting of everyday life. Everyone is repeating generic, vague scripts. I see it everywhere online. It’s creeping in organizational discourse too. It’s as though the templated text of LLMs is somehow becoming knowledge. Templated text is a dangerous thing to become knowledge.

For here’s the thing: Knowledge does not objectively exist. It is instead an emergent set of relations developed from information exchanged between human beings. Templates allow for radical simplification of information exchange through a controlled centralized environment. In their guidance, templates are regulatory to our abilities to write and, subsequently, read.

The only way I know how to resist the templating of the American mind is to encourage writing that exists unshackled. To reward thinking away from scripts, to throw away the formulaic forms, burning templates to the ground, rending out insipid genre conventions until only the nerve roots are exposed, revealing writing that is ready to scream, to wail, to howl, when touched.