Not too long ago, I came across a post claiming that transition phrases were indicative of AI use. It might have seemed like one more harmless addition to the expanding list of AI "tells", had I not recently returned to school to complete my degree. One of the first observations made during an assessment of my writing was that my prose lacked transitions. One authority was instructing me to develop a convention of competent writing, while people elsewhere were teaching readers to interpret that same convention as evidence that the writing might not be mine. The contradiction placed me in a familiar bind: without transitions, my writing could be judged insufficiently coherent; if I added them, it could become suspiciously machine-like. I was being instructed to make my prose more legible and then told that legibility itself might be evidence that I had not written it.
Transition phrases are only the newest object of suspicion. Updated lists of AI tells have included em dashes, semicolons, polished symmetry, particular sentence structures, and words as common as "landscape". While the machines did not introduce these features into language, they have been gradually detached from their history and returned to us as evidence that a human being may not be present. In an unpublished essay, I asked myself whether people even read for pleasure anymore; I was watching readers stop at one punctuation mark or turn of phrase, and abandon whatever encounter the text might have offered. The search for meaning appeared to be giving way to a search for provenance, and reading began to resemble a policing exercise in which every sentence was evaluated for what it might betray.
A reader may reasonably care whether a student completed an assessment without prohibited assistance, whether reporting was written by the journalist whose name appears above it, or whether a commissioned work satisfies the conditions under which it was purchased. Also, a reader doesn't owe a writer an encounter, nor do they owe attention proportional to the writer's labor. However, freedom from obligation does not make every reading practice equally rewarding, nor does it free us from having to consider what kind of reader the detection culture produces.
I believe a text benefits from good-faith consideration. But, this is not a request for charity on the writer's behalf. Good faith is a precondition through which the text can be encountered at all, and the reader is its primary beneficiary. By remaining open for a little longer, the reader preserves the possibility of surprise, recognition, disagreement, pleasure, and transformation. While the writer benefits incidentally, the greater reward belongs to the person whose world remains capable of being enlarged. Good faith does not mean credulity; a reader can remain skeptical and eventually reject the entire work. What good faith suspends is premature finality. It gives the text enough time to establish what kind of object it is, what it is attempting, and whether the feature that initially caused suspicion carries the meaning the reader assigned to it. The minimum is modest: enough attention to ensure that a local feature does not become a total judgment before the work has had a chance to establish itself. The difference is between noticing and sentencing.
When someone reads for AI tells, an em dash or transition phrase is rarely treated as one feature among many. It becomes jurisdictional. Once detected, it claims authority over everything else. The reader no longer asks what the sentence is doing because they believe they have discovered what the sentence is. This is how the search for provenance replaces the search for meaning. The problem with this is the question arrives early enough to prevent other questions from becoming available. Instead of asking what the writer tried to convey or whether their argument holds weight, the reader begins examining the prose for additional confirmation of a conclusion already reached. No amount of accommodation from the writer can repair this relationship once suspicion becomes self-sealing. If the writer removes the em dash, suspicion migrates to the semicolon; if they avoid "landscape", it moves to transitions; if they simplify the prose, genericness becomes incriminating; they make it more idiosyncratic, and the irregularity can be interpreted as a machine attempting to imitate humanity. The criteria continuously changes to preserve the verdict, leaving the writer unable to satisfy the assessment.
The writer may owe clarity, honesty, and whatever the form reasonably promises. But only the reader can stop converting every possible response into further proof. The dynamic reminds me of dating. Nobody owes a particular person intimacy or continued attention. Yet, someone who chooses to date has an interest in approaching the person in front of them with enough openness to perceive them. We all bring histories into new encounters: betrayals, insecurities, and reasons for no longer wanting to try. While those histories may be justified, they can also become so authoritative that the new person is never permitted to become particular. Every silence suddenly resembles abandonment. Every delay confirms indifference. Still, every unfamiliar habit activates an older danger. The person in front of us suddenly turns into a surface onto which previous injuries are projected. The person may not be entitled to our openness. Still, without some willingness to suspend the verdict, we may deprive ourselves of whatever the encounter could have become.
Readers also arrive with histories. Someone may have encountered enough generic, deceptive, or unrewarding AI-generated prose to develop genuine vigilance. The reader does not need to discard the history, but they need to prevent it from speaking both first and last. This becomes especially important when the surrounding discourse claims to protect authentic and original writing. Originality requires risk. A writer may need to attempt an unusual rhythm, retain an unfashionable word, follow a metaphor beyond the point at which its purpose is immediately evident, or construct a form whose value cannot be recognized through familiar categories. They may write a passage whose polish appears unlikely from someone without an established reputation. They may repeat a structure deliberately, or use precisely the phrase a circulating list has declared suspicious, because it remains the phrase that best carries the thought. If every deviation from the new norm can trigger disqualification, writers learn that the safest proof of humanity is conformity. A culture demanding authentic human writing may, therefore, create conditions under which writers are afraid to take the risks authenticity requires. The writer begins choosing language not according to whether it expresses what they mean, but according to whether it will survive a reader determined to infer its production history.
The result, inevitably, is defensive prose. The pattern feeds itself: readers identify recurring features as signs of generic AI writing; writers remove those features to preserve credibility; the range of publicly acceptable expression contracts; more people begin writing within the same low-risk corridor; readers encounter increasing sameness and attribute it to AI; another set of tells emerges, and the corridor becomes narrower, still. Instead of merely detecting genericness, the policing helps produce it. The consequences are not distributed evenly. An established novelist can use an em dash, invent a word, repeat a structure, or publish an unnervingly polished paragraph and have the peculiarity interpreted as style. A writer without accumulated authority may have the same choice interpreted as evidence. The people most in need of experimentation to discover a voice may be the ones least able to survive experimenting publicly. What is lost here is larger than stylistic variety.
One of the most beautiful possibilities of reading is that it allows one consciousness to host another temporarily. A reader can enter someone else's fears, reservations, contradictions, and longings long enough to leave changed in ways they did not anticipate. Writing can provide a window into another arrangement of thought and feeling, and through it, into parts of the human psyche that ordinary interaction may never make visible. That possibility depends upon the reader granting that there may be something here to inhabit. Once the text becomes primarily forensic material, its features perform different functions. The reader may believe they are protecting authentic human interiority while damaging the very capacity through which authentic interiority becomes perceptible. Over time, this can diminish a reader's ability to empathize through text.
At first, this may appear to be an individual aesthetic preference. After all, readers are entitled to these preferences, just as they are entitled to dislike first-person narration, long paragraphs, or stories written in the present tense. But preferences do not always remain private. Visibility has become a primary currency for writers and content creators. It affects readership, income, commissions, publication, employment, invitations, and even the reputation someone carries into nondigital spaces. When writers notice that particular turns of phrase attract accusations or appear to reduce engagement, they learn to police themselves whether or not anyone issues a direct instruction. No institution needs to publish a style guide forbidding transition phrases. A writer sees comments ridiculing them, watches another person publicly exposed, and understands that the feature has become dangerous. One person is punished so that thousands of silent observers can learn what not to sound like. The preference becomes materially binding without ever becoming epistemically sound.
The question is what writers can do in response. It would be easy to ask them to resist, retain every suspicious feature, and accept whatever visibility is lost in defense of expressive freedom. But, courage is easier when one can afford invisibility. Emerging writers cannot be expected to martyr their work so that readers may eventually develop better habits. Writers with established platforms, however, possess another kind of power. They can disendorse suspicion rather than pretending it carries more epistemic weight than it does. They can explain that machines learned these conventions from human writing, restore the history of allegedly suspicious features, distinguish aesthetic dislike from evidence of provenance, and refuse to amplify public exposure based on weak tells. They can say the real thing: removing useful transitions, punctuation, and structure to satisfy a detection culture damages writing rather than protecting it. This does not mean writers should dictate what readers enjoy. But, those with enough authority to shape literary norms should stop teaching readers that suspicion is knowledge.
The kind of reader I hope to write for is not naïve. They know AI-generated writing exists. They may recognize recurring patterns and care about how a work was produced. But when suspicion appears, they hold it provisionally while continuing. Their posture is something like this: I recognize this feature. Let me see whether the rest of the text changes what I think it means. That small willingness preserves the encounter. The suspicion may remain after the text is finished. The work may confirm it, complicate it, or render it irrelevant to what the reading provided. The reader may still leave dissatisfied. But the judgment will emerge from the experience rather than replacing it. I know the alternative because I have watched the discourse alter my own attention. At first, I would encounter the familiar tells and continue reading because I derived value from the work.
Eventually, the surrounding commentary poisoned certain kinds of prose for me. The dislike became emotional enough that it arrived before conscious reasoning. I would encounter a piece and immediately recoil. Only close inspection revealed the machinery beneath the reaction. The lists had not merely described my preference; they had also helped create the preference they could later cite as evidence that this kind of writing was naturally intolerable. I have recovered some openness, but not by making the associations disappear. Reading prose saturated, in my mind, by everything unfavorable I had heard about AI-adjacent writing remained uncomfortable. The task was to sit with that discomfort without allowing it to issue the verdict. The surrounding polemic could remain audible while the particular text was given a chance to speak.
The good reader, rather than someone who never experiences suspicion, is someone who can sit beside suspicion without immediately handing it interpretive authority. Inside the interval between reaction and judgment, is the freedom to encounter a text on its own terms. That does not guarantee insight; some writing remain generic, tedious, manipulative, or simply irrelevant. But the possibility of being enriched reenters the picture. Something superficial no longer has automatic authority to prevent an experience that might become generative. Reading has never required innocence, as we bring every prior disappointments to the page. But if we allow those things to decide what a text is before it has had time to become itself, we may protect ourselves from deception by becoming unreachable. The writer might lose an encounter, but the reader risks losing a world.