AI for Writing
AI can speed up every stage of writing, from a blank page to a final polish. Used carelessly, it also makes everyone sound the same. Here is how to get the speed and keep the voice — and when you need to say you used it.
Read ten AI-written LinkedIn posts in a row and you start to notice it: the same eager openings, the same three-part lists, the same tidy uplifting endings. Nothing is wrong, exactly; it is just nobody in particular. The opposite problem is just as real — the report you have been putting off for a week because the first sentence will not come. Good use of AI for writing sits between these two: let it break the blank page and do the heavy lifting on structure and polish, but make sure that what goes out still sounds like you and says what you mean.
Where AI helps in the writing process
Writing is not one task but several, and AI is more useful at some stages than others. Tap each stage.
Brainstorming and outlining
Brainstorming is where AI is at its most generous: it never tires of producing another list. The trick is to ask for range and to give it your constraints, so the ideas are usable.
For outlining, give the model your messy notes and ask for structure — then argue with it.
<notes>
[paste your notes]
</notes>
Drafting without losing your voice
The single best way to keep your voice is to show the model your writing. Paste two or three samples of things you have written and liked — emails, posts, reports — and ask it to match them. This is few-shot prompting applied to style (see Prompt Craft).
<samples>
[paste 2–3 samples]
</samples>
<facts>
[the facts of the delay]
</facts>
Then edit hard. A useful rule: change at least one thing in every paragraph, cut the first sentence if it is throat-clearing, and replace every vague phrase with a specific one. Watch for the tell-tale signs of untouched AI text: lists of exactly three, sweeping openers, heavy use of certain words such as “delve”, “crucial” and “landscape”, and conclusions that summarise what was just said.
Most importantly, the ideas should be yours. If the model supplied the argument, the examples and the conclusion, you have not written the piece — you have approved it.
Editing, tone and summarising
Editing is where AI gives the most value for the least risk, because you start from your own text. Ask for suggestions with reasons instead of a silent rewrite, so you learn and stay in control.
Tone shifts are a common job: firmer, warmer, more formal, shorter for mobile. Be precise about the direction and the reader.
For summaries, say who the summary is for and what they will do with it. “Summarise for a manager deciding whether to approve the budget” produces something very different from “summarise”. Always check that the summary has not added a conclusion the original did not reach — a common and subtle error.
Translating between Bahasa Melayu and English
Modern assistants translate between Bahasa Melayu and English fluently for everyday purposes, often more naturally than older translation tools As of 2026. The weak points are nuance and register:
- Forms of address. Choosing between anda, awak, kamu or a title such as Tuan/Puan, and handling honorifics such as Yang Berhormat or Dato’, depends on the relationship and setting. Models often pick one and apply it throughout.
- Malaysian versus Indonesian. The languages are close but not the same; models sometimes slip in Indonesian words or spellings.
- Idioms and peribahasa. Proverbs and culturally loaded phrases may be translated literally, losing the meaning or the respect intended.
- Official terms. Government, legal and technical terms often have an established Malay equivalent; a model may coin its own instead.
For anything official, legal, medical or public-facing, have a fluent speaker review the result. A useful self-check is back-translation: ask a different chat to translate the result back into English and see whether the meaning survived — it catches gross errors, though not subtle ones.
Disclosure and academic integrity
Debated When should you say that AI helped? There is no single rule. Norms differ between workplaces, publishers, universities and even individual lecturers, and they are still changing As of 2026. Some points are widely agreed: an AI tool cannot be listed as an author, because it cannot take responsibility for the work; and passing off AI-generated work as your own where that is forbidden is dishonest, whatever the tool.
- Disclose any use: some journals, courses and organisations ask you to state any AI assistance, including editing, so readers can judge the work.
- Disclose substantive use only: many people treat grammar checking and light editing like a spell-checker, and disclose only when AI generated content, ideas or analysis.
- Prohibit in assessment: some assessments ban AI altogether because the point is to show what the student can do unaided.
I’m a student. Can I use AI for my assignment?
Read your institution’s policy and the specific assignment instructions; if they are unclear, ask your lecturer before you start. Many Malaysian universities and schools now publish AI guidelines, and permitted use varies from “not at all” to “allowed with acknowledgement”. When in doubt, keep a short log of how you used AI and what you changed.
Is AI-assisted writing plagiarism?
Plagiarism means presenting someone else’s work as your own. Using AI in a way your course or employer forbids, without disclosure, is widely treated as academic misconduct even if no human source was copied. Separately, AI can reproduce phrases from its training data or invent references, so you remain responsible for every sentence and citation.
Can AI detectors prove someone used AI?
No. Consensus AI-text detectors give probabilities, not proof, and produce false positives. OpenAI withdrew its own AI-text classifier in 2023 because of its low accuracy, and a 2023 study found popular detectors often flagged writing by non-native English speakers as AI-generated. A detector score alone should never decide a misconduct case. More on detectors in Cutting Through the Noise.
What should a disclosure say at work?
Something short and factual: “Drafted with the help of an AI assistant; facts and figures checked by me.” Follow your organisation’s policy — some require disclosure for external documents, some do not. See AI in the Workplace.
What is the most effective way to keep your own voice in an AI-assisted draft?
Samples show your style; hard editing makes the result genuinely yours.
Which editing request keeps you most in control?
Suggestions with reasons let you accept or reject each change — and learn from them.
A model translates a formal school letter into Malay. What is the most important next step?
Back-translation catches gross errors only; public-facing text needs a fluent human check.
An AI detector says an essay is “92% likely AI”. What does that establish?
Detectors produce false positives, especially for non-native writers. Evidence and dialogue matter more.
// Pause and reflect
- Which stage of writing do you find hardest — starting, structuring, drafting or polishing? Which prompt on this page would help most?
- What are the disclosure rules where you work or study? If there are none, what would you consider fair?
- Read something you wrote with AI help recently. Does it sound like you? What would you change?
Your notes are saved privately in this browser.
Choose three things you wrote yourself and are happy with (no confidential content). Use the “Match my style” prompt to draft a new, non-sensitive piece — a club newsletter item, a social post, a thank-you note. Then edit it until you would sign it. Finally, compare: how many sentences survived untouched? That number tells you how much of the piece is really yours.
Key terms on this page
Sources & further reading
- Committee on Publication Ethics (COPE), position statement “Authorship and AI tools” (2023).
- Liang, W. et al., “GPT detectors are biased against non-native English writers”, Patterns (2023).
- OpenAI, “New AI classifier for indicating AI-written text” (2023; updated July 2023 to note its withdrawal).
- Anthropic, Prompting best practices (accessed 2026).
- Google, Prompt design strategies (accessed 2026).