Banned Words and Style Rules for Your Prompts

Asking an AI tool to "sound more human" rarely works. The instruction is vague, so the model guesses, and its guess usually adds a cheery tone and an exclamation mark. Specific rules work much better. This lesson gives you a list you can paste into any prompt today.

Rules beat adjectives

Compare two instructions:

Adjectives Write in a natural, engaging, human tone.

Rules Use contractions. Keep most sentences under 20 words, and include at least one under 8 words per paragraph. Don't start sentences with "Moreover" or "Additionally".

The first leaves everything to the model's idea of "natural", which is the average of its training data. The second gives it things it can actually do and you can actually check.

Word swaps

These are the words most often flagged as AI-overused in research and in Wikipedia's editor guide. The right-hand column is what people tend to write instead.

Instead of	Try
delve into	look at, dig into, cover
leverage	use
utilize	use
crucial, pivotal, vital	important, key, or say why it matters
robust	strong, solid, reliable
seamless	smooth, easy, or describe what happens
foster	build, encourage, grow
landscape	market, field, situation, or name it
navigate the complexities of	deal with, handle
a testament to	shows, proves
plays a crucial role in	helps, matters for, drives
in today's fast-paced world	delete it
embark on a journey	start
showcase	show
tapestry, realm, myriad	delete or rewrite the sentence

The paste-in list

Copy this block into the top of your prompt, or save it as custom instructions in your AI tool so it applies to every chat.

Style rules. Follow them strictly. Never use these words or phrases: delve, tapestry, testament, realm, landscape, crucial, pivotal, vital, robust, seamless, leverage, utilize, foster, navigate, embark, journey, showcase, underscore, myriad, plethora, holistic, synergy, game-changer, cutting-edge, "in today's world", "it's important to note", "plays a role", "not just X but Y". Never start a sentence with: Moreover, Furthermore, Additionally, Ultimately, In conclusion, Overall. Writing rules: - Use contractions (it's, don't, we're). - Vary sentence length. Each paragraph needs one sentence under 8 words. - No em dashes. Use commas, periods or parentheses. - Don't group things in threes by habit. Use one or two items unless there really are three. - No summary paragraph at the end. - Don't invent facts, statistics or quotes. If something is missing, leave [GAP] and keep going. - No intro like "Sure, here's" and no closing offer to help further.

Why models still slip

Even with this list, you'll see an occasional banned word in the output. Models follow rules better near the start of a response and drift as it gets longer. Two things help:

Ask for shorter pieces. Draft section by section instead of a whole article at once.
Ask for a self-check. End the prompt with: "Before you answer, check your draft against the rules and fix any breaks."

The AI detector catches what's left. Paste the draft in and swap whatever's highlighted.

Don't overdo it

It's possible to ban so much that the writing goes stiff. If every useful word is off-limits, the model starts reaching for odd synonyms, and you trade one kind of strangeness for another. Keep the list focused on words you would never use yourself. If "important" is a word you use all the time, don't ban it.

Merge this list with your voice brief and you've got the start of a personal style sheet.

Try this

Run the same request twice in a new chat: once plain, once with the paste-in list on top. Paste both into the AI detector and compare the number of highlighted phrases. Then add two words of your own to the list, ones that annoy you every time you see them.