AI writing seems to be everywhere these days—in classrooms, hiring processes, advertising, literary publishing, and news stories. You may have found yourself questioning whether something you’re reading is human writing or the hallucinations of a bot.
Theories—mostly unsubstantiated—persist about the markers of AI writing. For a minute, the use of em dashes was touted as a tell of AI-generated writing, much to the chagrin of Emily Dickinson stans the world over. (For the record: This LSA writer’s affinity for em dashes, as seen in the previous sentence, is 100 percent human-generated.) Overreliance on the “rule of three” has also recently come under scrutiny as a giveaway that a piece of writing was generated by AI. Both of these popular conjectures have been roundly dismissed by linguists.
Laura Aull, LSA professor in English language and literature and in linguistics, says that identifying the distinctions between human and AI writing is more complex than clocking punctuation, grammar, and cadence. (There’s that rule of three—again, this writer is a human.) Here, Aull talks with LSA about what AI English is, and what AI writing means for human literacy.
LSA: What is AI English, and how did it come to be?
Laura Aull: By AI English, I mean all the English generated by bots—think ChatGPT, LLaMA, Gemini, Claude, DeepSeek—which rely on English large language models (LLMs). How this AI English came to be is directly related to how AI English is different than human English.
AI English came to be in two overlapping ways: Firstly, what corporate AI designers decided to include in LLMs to represent English, and secondly, the most probable and reinforced combinations of words therein. In terms of representation, this especially means British and U.S. English used in standardized school writing, books, and agreeable internet posts—not, for instance, Indian English, or English used in text messages, even though those uses represent millions of English uses and users, every single day.
Probable combinations mean the grammatical and lexical patterns likely within English LLMs. Those grammatical patterns include syntactic density (for instance, regular long noun phrases, and long sentences). The word patterns in AI English include word associations (for instance, the geographical association between Japan and Tokyo, and the biased association between doctor and man).
Of course, human English came to be differently than AI English, and its patterns capture that. Over many centuries, human English has emerged and evolved based on how people have used it all over the globe, in all kinds of ways—as a native and additional language, in speech and writing, in families and schools, in businesses and metro stations. Human English came to be in varied places with varied language users, and its grammar and word combinations are likewise varied.
This means that, overall, AI English is based on what is probable in some English, and AI English patterns favor formal, informational, and native-speaker English. By contrast, human English is based on what is possible in global English, and human English patterns include formal and informal, interpersonal and informational, native speaker and additional language use, all over the world. Put another way: AI English is like a big can of some English, driven by math; human English is a big world of English, driven by all of English’s possibilities and influences.
LSA: Can you talk about some of the distinctions that you’ve identified between AI writing and human writing?
LA: Human English represents more kinds of dialects and genres, and it is less grammatically dense and more humble. AI English is more formal, more impersonal, and more sure of itself.
One way to put these overall differences is this: Human English has more variation and readability than AI English. An example of variation is how human English uses everything from multiple A’s in yaaaay to respond to a friend in a text message, to more forms of subject-verb agreement (e.g., they is as well as they are), and more kinds of cohesion (for example, the use of a word like therefore to link ideas). In terms of stance, human English tends to include a blend of caution and certainty.
AI English has less variety in its usage, cohesion, connection, and stance patterns. To continue these same examples, it is less likely to use innovative forms, it generates fewer kinds of subject-verb agreement, and it uses a smaller range of cohesive forms repetitively. It uses more boosters and counters that show certainty than hedges and concessions.
These usage and stance patterns give people the sense that AI English feels “polished,” but also “assertive” and “insincere,” while human English feels “messy” and “incorrect” to some, but is also praised for having “voice,” “personality,” and “adaptability.”
LSA: To continue that thought, how might the experience of reading AI English be different from the experience of reading human English?
LA: In terms of readability, human English breaks up information in both dense and plain detail, more often. It has, for example, dense sentences, with long noun phrases; but it also has sentences with simpler nouns and verbs. These simpler sentences might provide a concrete example to follow up an abstract, noun-heavy sentence, for instance. That means that with some regularity, human English sentences have subjects and verbs that are close together, making it easy for many kinds of readers (from a range of education levels, language backgrounds, and neural processing) to tell who is doing what. By contrast, AI English uses long noun phrases more often, making it less specific, harder to read, and less accessible overall. These patterns are especially true with AI-generated academic writing.
These readability patterns give people the sense that AI English feels “empty” or “abstract,” while human English feels more “real” and “specific.”
LSA: Where are you seeing AI writing appearing in places we might expect to see human-generated writing? What’s surprising about the context in which AI writing appears? Are the places where we are finding AI writing changing?
LA: The unsolicited ubiquity of AI tools is what strikes me most. Just the day before writing this, I saw that Google scholar has now embedded an AI-generated “quick read” option linked to the right of most articles. For busy readers (and who isn’t busy?), this can incentivize summary reading versus slower, fuller engagement with research materials, and it is clearly not something either a reader or writer opts into (I didn’t opt into it as a reader, and much of my own work has the quick read option, which I never elected).
Literacy is extremely complex. In a single day, most LSA students will need to read grammatically dense texts with specialized terminology; receive and send text messages that follow intricate sociolinguistic expectations; and produce visual data, calculations, and/or written text that follow post-secondary and disciplinary expectations. Literacy has never been simple, in other words—not before nor since AI. It is slow and iterative, and it is best supported with explicit attention and instruction.
Enter AI, which offers a way out of the frustrating slowness and complexity of literacy development, at least in the short term. What I would say is surprising today, then, is the ubiquity and ease with which a person can use a tool that outsources literacy challenges (and is bad for the environment, to boot).
Yet this same scenario is an opportunity. It has always been a good idea to slow down and be explicit and transparent about the unique literacy challenges of new tasks, rather than taking for granted that students—regardless of their background experience with such literacy practices—will know how to begin and persist, just because they are university-level students.
I should also note the widespread skepticism many university students feel toward AI. Studies in multiple countries show that faculty and students alike remain wary about over-reliance on automation and the monovoice of AI-generated language. Research shows that LLM bias, ethical and environmental concerns, and lack of transparency in AI-driven decision-making persist, even when students describe that AI has advantages. At the same time, so far, postsecondary students report more worry and skepticism, while secondary students report greater trust and fewer negative feelings. This suggests we benefit from ongoing, critical conversations with folks of all ages about the possibilities, limitations, and implications of AI.
LSA: What does your research say about how the existence of AI English impacts human literacy?
LA: This is a live question, but right now, I think four points merit our attention. One is how beginning with AI means we engage less of our brains than if we begin with our own ideas. Reading and brainstorming without the use of AI, at least first, means our brains are more active.
Another is that AI English can normalize the more “polished” but “bland” writing such that we lose sight of the value of “messy” but importantly creative, diverse, and adaptive human English use.
A third is that AI needs human oversight, because AI can make probable choices, not all possible ones. It cannot make empathetic decisions, for instance. Those tasks that entail innovative, empathetic, flexible, adaptive language choices are best done by humans, even if they take more time. This is because LLM-driven writing evaluation—such as in student placement or hiring decisions—shows clear bias, against non-native writers and nonstandardized uses of English. Those kinds of reading and writing processes, even if AI is involved, need to be overseen by humans.
A final point relates to whether we care if a human or a robot wrote the writing they are reading. So far, humans care very much. Research shows that even people who use AI English themselves respond badly to finding out something sent to them was generated by AI.
Top image: designed by Matt Vierling using photography from Tierney, MSThiramony, Korkeng, Yanka, and stock Dady/Adobe Stock. Faculty photo: courtesy of Laura Aull
Look to Michigan for the foundational knowledge and experience to ignite purposeful change.
LSA is the place where creative thinkers engage with a complex, diverse, and changing world. See how your support can make an impact on what’s next, for a better tomorrow. Learn more.
