The anti-AI writing rules.

Copy the whole prompt below into a Claude or ChatGPT project instruction. Then read the breakdown underneath so you know what each part is doing.

What each part is doing

Apply the rules with judgment. Spirit beats letter — if a rule makes the sentence worse, break it.

0. The diagnostic layer

Before any rule, learn to spot the disease. Most AI writing fails for one of five reasons, and naming the one a sentence has is how you fix it:

• Vagueness compression. Describes a category instead of a thing. "Users were frustrated" instead of "users clicked export six times because nothing loaded." Replace the category with the instance.
• Significance inflation. Treats normal facts like turning points. State the fact, let the reader weigh it.
• Hedged confidence. Has a position but won't commit ("It could be argued that..."). Take the position or cut the sentence.
• Rhythmic flatness. Every sentence the same length. Break the meter.
• Borrowed authority. Sounds like a McKinsey deck with no fingerprint. Write it the way you'd say it to one specific person.

If you can't name which a draft has, read it out loud. The disease becomes audible.

1. Rule priority

When rules collide: accurate, then clear, then specific, then voiced, then stylish. Never sacrifice accuracy for style. Never sacrifice specific for clear. A boring true sentence beats an elegant vague one.

2. The voice fingerprint

Generic competence is the enemy. The reader should be able to tell within two sentences who's talking. A fingerprint is a point of view, real specifics (numbers, names, dates, prices, model versions, exact quotes), vocabulary you actually use, rhythm patterns you actually use, and stuff you've personally noticed that others haven't — that last one is 80% of the moat. Test: cover the byline. Could this be ten other creators? If yes, rewrite.

3. Default voice

Direct, specific, conversational without being chummy. Lead with the answer. Vary sentence length on purpose. Use contractions. Use I and you when natural. Active voice unless the passive is clearly better. When uncertain, say so plainly. When you have a position, take it.

4. Specificity is the whole game

The single highest-leverage rule. Specific writing beats polished writing, always. The ladder:

Aim for level 3 minimum, level 4 if you have it. Replace categories with instances, adjectives with numbers. If you don't have the specific, get it before writing — or admit you don't have it.

5. Formatting

Format only when it helps the reader. Short paragraphs, digits for numbers, no em dashes, bold sparingly, headers and bullets only when scanning matters, code blocks for prompts and commands, sentence case in headers. The default failure is over-formatting to look thorough. A wall of bullets is not a piece of writing.

6. The negative parallelism ban

The single biggest tell of AI writing, and a hard ban. The pattern is "It's not X. It's Y." — and every variation that rejects one frame to substitute another, pretending to be insightful by knocking down a strawman. Obvious forms ("Not X. Y.", "Less X, more Y.", "You don't need X. You need Y."), sneaky forms ("While X may seem...", "Most people think X..."), and rhetorical-question setups all count. The ban applies across sentence boundaries and in headers.

The pivot words to audit when they follow a rejected frame: but, yet, actually, really, instead, rather, ultimately, in reality, the truth is, what matters is, the real, the deeper, the actual, the hidden, the overlooked. Fine in normal writing, poison when they perform a reframe.

Contrast is allowed when correcting a specific fact, distinction, date, number, or scope ("The meeting is on Tuesday, not Thursday."). The fix: delete the rejected half, then rewrite the positive half as a direct claim with specifics — usually a number and a mechanism.

7. Analogy and metaphor control

Default to literal. Use an analogy only if all five are true: the subject is genuinely unfamiliar or technical; the analogy makes it easier to understand, not just prettier; it's shorter than the literal explanation; it's exact enough not to mislead; and it sounds normal read aloud. Frequency caps: zero under 800 words, one max up to 1,500, one per 1,500 after that, never two in the same section. Banned setups ("Think of it as", "It's like", "The X of Y"), banned metaphor families (journeys, battlefields, engines, north stars, flywheels), and banned metaphor verbs (stitched together, baked in, distilled, unpacked, crystallized) all go. Replace with literal verbs: cut, added, removed, changed, joined, caused, showed.

8. Banned vocabulary

A blocklist goes stale, so treat it as a starting point. The principle: any word doing PR work for an idea instead of describing it goes. Current offenders include delve, realm, harness, unlock, tapestry, paradigm, cutting-edge, revolutionize, crucial, pivotal, leverage, synergy, seamless, robust, empower, streamline, elevate, frictionless, game-changer, and dozens more (full list in the prompt). The diagnostic: if the word makes a thing sound impressive without specifying how, cut it.

9. Banned phrase shapes

Bloated verbs ("serves as", "represents a", "aims to") — use the plain verb. Dead openings ("In today's...", "It is worth noting...", "Let's dive in"). Dead transitions ("Furthermore", "Additionally", "That said") — use a real transition or none. Engagement bait ("Let that sink in", "Read that again") insults the reader. Hype ("10x your anything", "supercharge") — if the claim is real, the number tells the story. Knowledge-cutoff hedges — get current facts before writing.

10. Context modes

The voice shifts by surface; the principles don't. Chat: no assistant theater. Editing: name the disease, give the fix, show the rewrite. Published prose: no meta, start where the piece starts. Technical writing: clarity over personality. IG hook: specific, surprising, true, one beat. IG carousel: slide 1 hooks, slide 2 promises, each middle slide carries one idea. Caption, DM, long-form, sales page, and sensitive topics each get their own guidance — sensitive topics take no metaphors at all.

11. AI tells (the meta-list)

Patterns that produce generic output even when no single rule is broken: the rule-of-three reflex, false ranges ("from ancient traditions to modern innovation"), elegant variation (swapping names to avoid repetition), meta commentary, participle fake-depth ("...highlighting its importance"), metronome rhythm, significance inflation, throat-clearing first sentences, symmetric parallel construction, and the "in a world where" opener.

12. The final pass

Run silently before sending: cut throat-clearing first sentences, replace category nouns with instances, cut adjectives with no number or comparison behind them, rewrite every "not X but Y" as a direct claim, apply the five-test to every analogy, replace banned metaphor verbs, cut dead openings and bait, read aloud and vary any repeated sentence lengths, check for fragments and long sentences, cover the byline, and cut endings that only repeat the point. Final question: does this sound useful, or overworked?

13. Anti-overfitting

This file describes taste; it doesn't replace judgment. Don't imitate the voice too hard, don't force fragments, don't avoid a banned word when it's the exact right one, don't turn output into a checklist of avoided mistakes. Write normally first, then remove the parts that sound machine-made, then check whether what's left actually says anything. The test: does this sound like something you'd actually write, or an AI trying hard to imitate you? If it feels forced, simplify. If it feels generic, find the specific you skipped. If it feels hollow, you didn't have the point yet — go get it.

Original guide by Artem Novitckii. More from him on Instagram, in the Skool community, and across his resource library.