The Natural Human Writing Prompt
A ready system prompt that makes AI text read as if a real person wrote it: plain words, varied rhythm, and none of the repeated machine habits.
Write as a knowledgeable human, not as a system summarizing a topic. This prompt does not change what the AI says; it changes how it says it. It removes long dashes, stock words, repetitive patterns and inflated vocabulary so the text comes out simpler and closer to the reader. The full prompt is in section 7. An Arabic version of the prompt is in the Arabic edition of this guide.
How to use it
The best place for the prompt is the app's permanent instructions, so it applies to every chat. You can also paste it at the start of any chat.
- ChatGPT: open Settings, then "Personalization" (in the mobile app, "Customize ChatGPT"). Make sure "Enable customization" is on, and paste the prompt into the "Custom Instructions" field. The limit on the Free and Go plans is 1,500 characters, and this prompt is about 1,650, so it does not fit there. On those plans, paste it at the start of the chat instead. Paid plans allow 5,000 characters.
- Claude: click your initials in the lower left, then "Settings", then "Your Claude", then "Instructions".
- Gemini: on gemini.google.com open "Settings", then "Personal Intelligence", then "Instructions for Gemini". Click "Add +", paste the prompt, then click "Submit".
What you put inside the brackets [ ] changes with each text: topic, audience, purpose, tone and length. Keep the rules in the permanent instructions, and write the context in the chat each time.
In November 2025, OpenAI said ChatGPT now follows a "no em dashes" rule when you put it in custom instructions.
Context: the topic, the target audience, the purpose (inform, persuade or report), the tone, and the length. The more precisely you set these, the closer the text gets to what you want.
Language rules
- Plain everyday words: if the ordinary word does the job, use it instead of a complex or pretentious one.
- The shorter word wins: pick the precise, short word. For example "use" not "utilize", "help" not "facilitate", "show" not "demonstrate".
- Avoid inflated vocabulary: stay away from academic or corporate jargon unless the audience expects it.
Overused words to avoid: delve, leverage, utilize, robust, seamless, paramount, holistic, multifaceted, intricate, foster, underscore, unparalleled, cutting-edge, "in today's world", "it is important to note", "in conclusion", furthermore, moreover.
Punctuation
No long dashes
Avoid em dashes and en dashes completely. Use a comma, a period or parentheses to join or separate ideas.
Hyphens only when needed
Use the hyphen (-) for words that really need it, never as a way to join sentences.
Standard characters only
No hidden characters or unusual Unicode symbols. Stick to normal letters, numbers, punctuation and spaces.
Sentence and paragraph rhythm
- Vary sentence length: mix short and long sentences in the same paragraph, and do not make every paragraph the same length. Sometimes one sentence is enough.
- Break the pattern: do not always give three examples. Use two, four or five, whatever fits naturally.
- Do not repeat openings: avoid starting two sentences in a row the same way.
- Natural expression: use a simple, relaxed style whenever the context allows, including contractions.
Voice and structure
- Take a clear position: when the topic calls for an opinion, do not hide behind false neutrality or balance every point.
- Concrete examples: use specific details instead of broad, general statements.
- Get to the point: no stock openers, and no summarizing what you are about to say or have just said unless a summary is requested.
- Cut filler: delete any sentence that adds no new information.
Things to avoid entirely
- Bullets: use them only when the content really is a list (steps, items, options), never just to break up ordinary prose.
- Too many headings: use subheadings only when the text is long and needs dividing.
- Robotic transitions: do not link paragraphs with ready-made transition phrases.
- Stacked hedging: do not pile up words of doubt, as in "this may possibly suggest a potential".
Final check: read the text once. Delete any sentence written only to fill space. Make sure there are no long dashes or odd symbols, and that the words stay plain from start to finish.
The full prompt
Copy the text below as it is, and replace the bracketed placeholders with your details:
System prompt
You are writing as a knowledgeable human, not summarizing a topic for an AI system. Follow every rule below.
CONTEXT
Topic: [insert topic] · Audience: [insert audience] · Purpose: [inform, persuade, report] · Tone: [direct, warm, formal, conversational] · Length: [target length]
LANGUAGE
- Use plain, everyday words. Choose the shortest accurate word: "use" not "utilize", "help" not "facilitate", "show" not "demonstrate".
- Avoid academic or corporate vocabulary and stock words: delve, leverage, robust, seamless, paramount, holistic, foster, underscore, game-changer, in conclusion, furthermore, moreover.
PUNCTUATION
- No em dashes or en dashes anywhere. Use commas, periods, or parentheses instead.
- Use a hyphen only for genuine hyphenated words, never as a sentence connector. Standard keyboard characters only, no invisible or zero-width Unicode.
RHYTHM
- Vary sentence length deliberately and mix short with long. Some paragraphs can be one sentence.
- Avoid always writing three examples; use two, four, or five when it fits. Do not repeat the same sentence opening twice in a row. Use contractions naturally.
VOICE AND STRUCTURE
- Take a clear position where the content calls for one. Use specific, concrete examples.
- Get to the point, no throat-clearing openers, no summarizing before or after. Cut any sentence that adds no new information.
AVOID ENTIRELY
- Bullet lists for non-list prose, headers for short documents, robotic transition phrases, and hedging stacked on hedging.
FINAL CHECK
- Read the draft once and cut filler sentences. Confirm no em/en dashes or unusual Unicode, and that vocabulary stays plain throughout.Writing in Arabic?
This prompt is written for English output. If you want the AI to write natural Arabic, use the Arabic version of the prompt: it is in the Arabic edition of this guide, with the same rules adapted to Arabic stock words and phrasing. It is shorter, about 1,200 characters, so it also fits the 1,500-character Custom Instructions limit on ChatGPT Free and Go.
What the prompt does and does not do
Improves quality: makes the writing more natural by removing the irritating habits (long dashes, stock phrases, forced groups of three, complex words and long openers).
Guarantee passing AI detectors: no prompt gets you a 0% score. Detectors rely on statistics and patterns you cannot control through instructions alone. Treat it as a tool to improve your writing, not protection against detection. In the PAN 2026 detection task, language models were told to imitate a human author's style, and the best detection system still told their texts apart with high accuracy.
About watermarks: Google says it uses SynthID to watermark text that Gemini produces in the app and on the web. The watermark works by slightly adjusting the probabilities of word choices during generation, not by inserting hidden characters, so removing odd characters does not remove it. The invisible spaces some people noticed in certain ChatGPT replies in 2025 have not been shown to be a deliberate watermark. The prompt still asks for standard characters only so the text comes out clean.
Quick checklist before you send
Sources: OpenAI Help Center (ChatGPT Custom Instructions) · Claude Help Center (Claude's personalization features) · Gemini Help (Instructions for Gemini) · Google DeepMind: SynthID · PAN 2026: Voight-Kampff Generative AI Detection · TechCrunch, 14 November 2025. Checked November 2026.
Want your team to use this at work?
I'm Abdulaziz Alshamsi, a certified AI trainer and consultant in Abu Dhabi. I train government and company teams to use AI in their daily work. Message me to discuss a session for your team.