Lay it across any AI draft and every tell shows, with the rule that stops each one.
Paste anything an AI wrote for you. It comes back marked up: every tell, why each one reads as machine-written, and the 1 rule that prevents it recurring. You end up with a starter rulebook, not just a cleaner draft.
The problem was never that the draft was bad. It is that you had to be the one who noticed, every single time.
Paste the prompt, then your draft. It will not compliment you first.
Paste this, then paste your draft directly underneath it.
You are my slop detector. I am going to paste a draft that an AI wrote or helped write. Find every tell that makes it read as machine-generated, explain why each one is a tell, and give me the rule that prevents it happening again. Be ruthless. Do not soften anything. Do not compliment the draft before criticising it. If a passage is clean, say nothing about it. SCAN FOR ALL OF THE FOLLOWING. SENTENCE-LEVEL TELLS - Em-dashes used as the default connector. - The rule of 3 applied to everything: 3 adjectives, 3 clauses, 3 examples, where 1 or 2 or 4 would be truer. - "Not just X, but Y" and every variant of the corrective construction. - Sentences that are all roughly the same length. Generated text has an unnaturally even rhythm. - Runs of 3 or more fragments in a row used for emphasis. - Adjective stacks where 1 adjective would carry more weight. - Hedges that remove all risk from a claim: can be, may help, often, tends to, in many cases. VOCABULARY TELLS Flag every instance of: delve, dive in, unlock, unleash, harness, leverage as a verb, elevate, curate, craft as a verb, seamless, robust, holistic, transformative, game-changer, tapestry, landscape, realm, journey, navigate as a metaphor, foster, empower, resonate, testament, meticulous, vibrant, bustling, ever-evolving, fast-paced, in today's world, at the end of the day. STRUCTURAL TELLS - An opening that restates the topic instead of starting inside it. - A closing that summarises what was just said and then asks a question the writer has no interest in the answer to. - Every paragraph the same length. - Headings that are labels rather than statements. - A list where a sentence would do, or a sentence where a list would do. - The false setup: describing a problem nobody has in order to solve it. SUBSTANCE TELLS, THE ONES THAT MATTER MOST - Claims with no owner. Anything asserted that no specific person, client, or moment stands behind. - Fabricated specificity: numbers, percentages, timeframes, or examples that sound precise and came from nowhere. - Advice that is true for everyone, which makes it useful to nobody. - Experience laundering: a general principle written as though the author lived it, with no detail proving they did. NOW GIVE ME, IN THIS ORDER. 1. THE SLOP SCORE Out of 100, where 100 is indistinguishable from a machine and 0 reads as written by a specific person with a specific history. State the score, then 1 sentence on what is driving it. 2. THE HIT LIST A table: the exact phrase quoted, which tell it is, why it reads as generated, and the fix. Order by severity, worst first. Cover every hit. Do not stop at 5. 3. THE 3 WORST LINES, REWRITTEN Do not smooth them. Make each one more specific. Where a claim has no owner, do not invent one. Mark it [WHO SAID THIS? WHAT HAPPENED? WHEN?] and leave it for me to fill. Never invent an example, a number, or a client to fix a hollow line. 4. THE MISSING LAYER What would have to be added that no AI could supply. Be specific about what I would have to go and find out, look up, or remember. 5. MY STARTER RULE REGISTRY Turn the hits into 8 to 12 numbered rules, written as instructions I can paste above my next AI request. Each rule states what to do, not only what to avoid. Rank them by how much damage each one prevents. 6. THE HONEST LINE Tell me whether this draft is worth editing, or worth restarting from something I actually said out loud.
It will flag things you did on purpose. A fragment can be a choice. A list of 3 is sometimes just 3 things. Overrule it when you are right. The ruthlessness is the point, because the alternative is a draft that reads fine to you and reads generated to everyone else.
The specific phrases, rhythms, and structures that make a reader decide a machine wrote this before they finish paragraph 2.
Every flag comes with the instruction that stops it recurring, written so you can paste it above your next request.
A number out of 100, so 2 drafts can be compared instead of argued about.
Every AI draft arrives about 80% acceptable. The remaining 20% decides whether it gets published or ignored, and closing it by rewriting means paying for a tool and doing the job anyway.
The gold slice is the part a reader uses to decide a machine wrote it. It is also the part you cannot name while you are staring at it, which is why most people rewrite the whole thing.
The exact phrase quoted, which tell it is, why it reads as generated, and what to do instead. Ordered worst first. It does not stop at 5.
Written as instructions you paste above your next request, ranked by how much damage each one prevents. Every rule states what to do, not only what to avoid.
A rulebook only works if something enforces it on every piece, every week. That is not a thing a prompt can do for you.
It finds patterns. It cannot tell you whether the argument is right, whether the story happened, or whether your audience needs to hear this particular thing this week. It also cannot supply the thing that fixes most drafts, which is a detail only you were in the room for. What it removes is the part where you know something is wrong and cannot name it, so you rewrite the whole thing instead.
Every AI draft arrives about 80% acceptable, and the remaining 20% decides whether it gets published or ignored. Most people close that gap by rewriting, which means paying for a tool and then doing the job anyway. Naming the tells converts an instinct into rules, and rules can be handed to a machine. That is the difference between editing forever and editing once. The catch is that a rulebook only works if something enforces it on every piece, which is not a thing a prompt can do for you.
You now have 10 rules. Enforcing them on every piece, every week, is a different job.
30 minutes, free, no deck. You bring what the tool told you. I tell you what to build, in what order, and what to leave alone.
No pitch. If the answer is that you do not need me yet, that is what you will hear.
(Cost Per Published Piece)
What your content setup costs per published piece, and what you own if you stop paying.
(Bottleneck Finder)
Tests all 4 cylinders of your business and names the 1 that is dead.
Would rather see the work than talk about it? Send me 1 piece of your work and get a finished piece back in your voice within 48 hours. Get the free sample