Student writing got more polished after ChatGPT. Here's what writers can learn.
Since late 2022, student reports at a major UK university have trended more formal, more sophisticated, and more upbeat-without any jump in grades. An analysis of 4,820 reports (17 million words) across 10 years shows a clear inflection after generative AI tools became common on campus. Up to 88% of students report using ChatGPT for assessments. Credit: University of Warwick.
The core takeaway for working writers: AI nudges style fast, but substance still wins. If you use AI, you need guardrails that keep your voice distinct and your thinking sharp.
What changed after ChatGPT's launch
- Sentiment shifted positive, often independent of content. This mirrors how many GenAI systems default to polite, constructive phrasing.
- Formality and vocabulary range increased in a way you'd expect after years of practice-not months. That suggests tool influence, not sudden skill growth.
- "AI tells" spiked: words like "delve" and "intricate" rose until 2024, then fell in 2025-likely a response to AI scrutiny.
- When older reports were rewritten by ChatGPT, the same shifts appeared. That's strong evidence the tools steer tone and style.
Grades didn't move. That's a signal.
Despite smoother prose, marks and examiner feedback stayed steady. The likely reason: assessment still rewards critical reasoning, interpretation, and argument-depth over surface. Translation for you: polish is table stakes; clear thinking and original framing make the difference.
The homogenization trap
AI often trends optimistic and formal. Useful for clarity. Risky for voice. If every paragraph sounds politely enthusiastic, your work starts to blur with everyone else's.
There's a second risk: constant positivity can blunt critique. If your drafts feel agreeable by default, push harder on analysis, counterpoints, and limits.
Practical steps to keep your voice (while using AI)
- Draft the spine first: Write your thesis, angle, and outline in your own words before you touch a model.
- Constrain the model: Prompt for structure, examples, or alternatives-not tone. If you do ask for tone, specify yours: "short sentences, dry humor, no fluff, no platitudes."
- Create a "ban list" for your brand: Words you never use (e.g., "intricate," "utilize," "plethora"). Add any AI tells you see creeping in.
- Flip the sentiment switch: Ask for downsides, failure modes, trade-offs, and what would change your mind. Insert at least one strong counterargument.
- Audit for sameness: Compare new drafts to three bylines you admire and three you don't. If it reads interchangeable, cut or rewrite.
- Shorten to sharpen: Cut 15-25% of sentences. Remove hedges, filler adverbs, and "polite noise." Keep the point, lose the padding.
- Keep a voice file: A living doc of your phrases, rhythms, and rules. Feed it to the model as context. Update monthly.
- Do a word-frequency pass: Scan for repeats every few pieces. If the same adjectives keep showing up, swap them or cut them.
- Human final pass: Read aloud. If you wouldn't say it that way, change it.
For editors and content leads
- Set style guardrails: Approved tone, banned phrasing, and examples of "on-voice" vs. "off-voice."
- Assess thinking, not gloss: Score argument quality, evidence, and originality higher than surface polish.
- Design prompts once, reuse often: Provide team-wide prompt templates that enforce your voice and structure.
- A/B for clarity vs. sameness: If AI improves readability but flattens voice, blend: human open, AI-structured middle, human close.
Why this matters
AI can make average writing clean, fast. It doesn't make thinking deeper. The study's pattern-smoother language, same grades-backs that up. If you want work that sticks, pair AI's speed with your own judgment, original angles, and a voice readers recognize instantly.
Source: Computers and Education: Artificial Intelligence (study DOI)
If you're building an AI-assisted workflow for copywriting, these curated tools can help: AI tools for copywriting.
Your membership also unlocks:
