If AI can imitate my writing, how can I show readers that I made choices, checked sources, and remain responsible for the article?
1. I want to retain more than a recognizable style
The original title said AI could capture the average but not you. Looking back, that turned an aspiration for writing into a guarantee about AI’s limits.
I still care whether an article sounds like me. But familiar expressions, sentence lengths, and uneven rhythms cannot settle that question. Surface features can be imitated, and my writing changes too.
I would rather explain why I chose a question, which evidence changed my mind, and which claim I left out. These are choices I ask myself to account for, not secrets declared permanently inaccessible to machines.
2. The “average face” is only a metaphor
The original called generated portraits and articles average, then used that impression to explain their lack of individuality. That went too far.
Finding some outputs similar is a viewing or reading experience. It does not establish that images average everyone’s faces or that text generation always picks the most probable word. Nor does it show that more capable models inevitably produce less individuality.
I can use average as a metaphor for work I dislike. Explaining how generation works requires different evidence. This article no longer treats the metaphor as a technical explanation.
3. Style can be learned; choices still need checking
I no longer argue that whatever can be copied is style and whatever cannot is essence. That definition can exclude anything AI achieves without answering the question.
A more useful boundary is that something sounding like me may still cite a false source, invent an experience, or make a commitment I do not accept. Natural prose alone is not enough for my signature.
If I keep accepting the same templates, openings, and edits, my articles may increasingly resemble the template. That is a working habit to watch, not a rule that every AI tool makes every author alike. AI can also suggest different approaches. The choice should return to meaning and readers’ needs.
4. A detection score cannot identify the author on its own
Can an AI detector tell who wrote an article? I would first ask about the language, data, version, and meaning of its score. A percentage should not automatically be treated as the definite probability that a person used AI.
Liang and colleagues’ 2023 study tested seven then-used detectors on 91 human-authored TOEFL essays and 88 US eighth-grade essays, finding substantial misclassification in the former group. This concerns historical tools and English samples. It warns about false accusations but does not supply an error rate for every current tool or Chinese article.
I therefore recommend against deciding someone cheated or impersonated another person from a single detection result. Grading, hiring, or public accusations should also consider applicable rules, writing history, other evidence, and an opportunity to explain. This is my practical recommendation, not a claim of one legal prohibition across all settings.
5. Discussing the process helps more than guessing an “AI flavor”
If asked how an article developed, I want to explain the topic, evidence, revisions, and trade-offs. That comes closer to my concern than asking whether it sounds like AI.
Answering follow-up questions or presenting drafts is not a foolproof way to identify a human author. I see it as context for understanding the article, not a new authenticity ritual.
Likewise, “every article went through a dozen revisions” cannot vouch for how all my historical writing was produced. A better practice is to explain an article’s sources, tool assistance, and what I actually checked when relevant. That concerns responsibility, not a competition to use no tools.
6. Four questions before signing
I would focus on the article rather than proving an essence that can never be imitated:
- Why did I choose this question and these examples, and what is only rhetoric or personal feeling?
- How far did I check important facts and quotations, and which experiences can I not vouch for on behalf of other people or historical versions?
- Which tool suggestions did I adopt or reject, and do the reasons return to meaning and readers’ needs?
- After signing, who answers challenges, corrects errors, and explains what remains unknown?
These are my writing checks, not a test of human identity or a guarantee against imitation or misclassification.
For me, writing is a practice of keeping judgment in my own hands. AI can help and imitate convincingly. I still need to know why I accept a sentence and what I will do if it proves wrong.
Sources and reading limits (checked through November 4, 2026)
- Liang et al., GPT detectors are biased against non-native English writers, arXiv v3, July 10, 2023; Abstract, Results, and Discussion. Its tools, English samples, and misclassification setting do not establish performance across all detectors, languages, or future versions.
- Similarity of generated work, personal style, and writing values are the author’s observations and positions. Portrait generation and authorship-identification experiments were not rerun; revision counts across historical articles were not verified.
Revision note: This version withdraws claims that average faces and maximum-probability words explain generation, guarantees of inimitability, insufficiently verified quotations, and the medical-identification example. It retains concerns about false accusations and responsibility for choices, checking, and corrections.


💬 Comments
Loading...