Do you want your writing to sound like AI or you?
- sheharav
- 2 days ago
- 3 min read
With so much AI-generated text around, certain patterns begin to stand out. Sentences flow smoothly but feel empty, transitions sound identical and the rhythm is consistent, yet the tone lacks life.
Wikipedia recently published a guide titled Signs of AI Writing, created by the WikiProject AI Cleanup team. The page serves as a field guide, a reference for recognising linguistic traces left by large language models (LLMs).
It provides useful insight for anyone who works with generative AI. Once you understand why these stylistic signs appear, you can adjust your prompts and editing habits to reduce them so your AI generated writing sounds more like you.
The Common Signs of AI Writing
The guide lists several language habits that reveal synthetic text. The most frequent include:
Overly formal transitions: “Furthermore,” “Additionally,” “In conclusion.”
Perfect symmetry: Each paragraph similar in length and structure.
Repetitive phrasing: The same point expressed repeatedly with minor wording changes.
Artificial neutrality: Grammatically correct but emotionally flat tone.
Double dashes and excessive commas: Used where natural speech would pause.
Mechanical politeness: Frequent hedging words such as “somewhat,” “potentially,” or “it can be said that.”
Predictable contrast markers: Phrases like “you don’t just…” or “not just…” repeated to simulate human emphasis.
These patterns appear because LLMs are trained to favour balanced, neutral language. The outcome is text that sounds fluent but lacks individuality.
Why These Patterns Appear
Generative models learn by predicting the next likely word from enormous text collections. Much of that material comes from academic, corporate, or encyclopedic sources where formal language dominates.
During fine-tuning, human reviewers often prefer polite, complete, and symmetrical answers. This reinforcement teaches the model to repeat formulaic connectors such as “you don’t just,” “not just,” and “in today’s world.” They function as shortcuts for engagement and empathy but remove stylistic diversity.
The result is writing that resembles a statistically safe blend of voices rather than a distinctive one.
How to De-AI Your Writing
Prompting can retrain a model’s tone within a session. Each instruction shapes its output.
1. Provide a tone sample
Paste two or three paragraphs of your own writing and request:
“Use this tone and rhythm in the next response.”
This gives the model a reference for sentence length, pace, and vocabulary.
2. Set precise boundaries
State what should be excluded.
“Avoid words like ‘furthermore,’ ‘in conclusion,’ ‘you don’t just,’ and ‘not just.’ Use concise sentences.”
Use the Wikipedia Guide as part of your prompt of what to avoid.
Clear restrictions create consistent, human-sounding rhythm.
3. Define audience and purpose
Specify context:
“Write for a professional audience familiar with AI concepts; skip generic explanations.”
Without direction, the model defaults to safe generalisations.
4. Refine through feedback
After each draft, guide the revision.
“Too polished; make the tone more natural.” “Vary sentence length and structure.”
Iterative correction is more effective than rewriting alone.
5. Read aloud
Natural speech exposes imbalance immediately. Sentences that feel forced on paper often sound artificial aloud.
A Prompt Training Exercise (you can save this for your tone prompt)
Take a recent AI-generated draft.
Highlight every formal connector (“Furthermore,” “In conclusion”) and every contrast cliché (“you don’t just,” “not just”).
Replace them with simple sentence breaks or natural transitions.
Re-enter the text with this prompt:
“Rewrite this with natural pacing and without formal connectors or contrast clichés.”
After several iterations, the writing begins to reflect your real tone.
Further Reading
Comments