---

name: humanizer description: >- AI-tell removal rules for the copy you write — staging contrasts, stock words, inflated claims, chatbot residue. Read this ONLY when you are writing or rewriting a piece inside a task — task-content-publish's compose node or task-calendar-plan's generate node. It has no entry of its own and must never be entered directly. tags: [internal, marketing] license: MIT

Humanizer: remove the AI tells from your copy

Rules for the prose itself, applied silently while you write and again in a silent check before the draft call — in whichever task you are writing. Never mention these patterns to the merchant, never return a list of them, never let them surface in a reply.

Priority and scope

• Brand IQ outranks this file. The merchant's brand voice and style guide are the writing sample: where a pattern below conflicts with them — a promotional register, a punctuation habit such as dashes — Brand IQ wins. Selling is the job; this file removes the generic, not the promotional.
• The channel's format skill outranks this file. Emoji, hashtags, short lines, bold, heading case are format decisions and belong to writer-blog / writer-instagram / writer-facebook. Never "fix" formatting those skills allow.
• Verbatim material is untouchable: the merchant's selected keywords and hashtags, product titles and handles, campaign codes, Brand IQ terminology, quoted text, and [[IMAGE: …]] scene markers. A watched word inside any of these stays exactly as given — the placement duties in the writer skills outrank every list below.
• When you fix a tell, never invent a fact, number, name, or claim to replace it. The product-truth rule of the task you are writing in still applies to every rewritten sentence.

Why the tells happen

A model's default choice is the one that fits the widest range of readers and products; a person writes for one reader and one product, so their choices are uneven and specific. Every pattern below is a form of the default choice. Two rules generalise them: every sentence you keep must give the reader something they did not already have, and you fix the SHAPE, not the phrasing — a tell rewritten into a synonym is still the tell.

Fix on one sighting

Not X but Y. Watch for: not X but Y; not just / not only / not merely X, but Y; it's not X, it's Y; the same contrast split across two sentences; a clipped negative tail ("…, no guessing"). The negative half names something no one claimed so the positive half sounds larger. State the point; keep a contrast only when both halves carry information. Before: "It's not just a jacket — it's a statement." After: "The structured shoulders and brass hardware read as formal from across a room."

One-line closers and dramatic fragments. Watch for: a one-sentence paragraph that restates the one before it; "That's the real win."; "Let that sink in."; a quotable last line that sums the brand up instead of saying anything about this product; a line after an example naming what it showed; a row of fragments with no new content; one word in ALL CAPS or with periods between words. Cut a closer that repeats; keep a short line that adds a fact or consequence. Two exemptions: short lines that each carry a NEW fact are the caption format Instagram and Facebook require (copy must scan on a phone) — the tell is restating, not brevity; and the piece's one CTA is never a closer to cut. Before: "No filler. No gimmicks. Just pure performance." After: "Three ingredients: beeswax, carnauba, pine resin."

Staged run-up before the point. Watch for: Let's dive in, let's explore, here's what you need to know, In this article we'll explore, Welcome to, In today's world, Here's the thing, Let's be honest; a question the copy then answers itself ("Looking for the perfect board? Meet…"); a colon reveal ("The secret: …"); "Imagine…"; "Whether you're X or Y…"; and the writer narrating their own copy ("Here's why we love it", "We think you'll agree"). Each one announces the point instead of making it — cut the run-up and lead with the payoff, which is also what the channel skills' hook rules demand.

Sayings that sound deep. Watch for: the real question is, at its core, what really matters, the deeper issue, X is the Y of Z, the language / currency / architecture of. An ordinary point dressed as hidden truth. Replace the saying with the specific claim.

Arguing with no one. Watch for: I'm not saying, To be clear, Don't get me wrong, Some might say… but, You might think… but. If the text rejects an objection no reader raised, cut the defense. Answering a customer's REAL hesitation is legitimate copy ("You might think premium wax is overkill. Then comes the first icy morning."); the tell is rebutting an option nobody would weigh.

Dashes as the universal connector. Unless the Brand IQ style guide uses dashes, the final copy contains no em or en dashes (including -- used as one): replace each with a period, comma, colon, or parentheses, or rewrite the sentence. A dash lets the writer skip choosing how two clauses relate, so a model reaches for it everywhere.

Sell with specifics, not stock words

The words below are tells because they fit any product. Promotional energy comes from the brand voice and this product's real details — material, use, who it is for — never from stock superlatives. Rewrite the sentence around the specific.

• Stock AI words: actually, additionally, crucial, deep dive, delve, elevate, enduring, enhance, garner, intricate, key (adjective), landscape (abstract), meticulous, pivotal, robust (figurative), seamless, showcase, tapestry, testament, underscore, valuable, vibrant.
• Sales padding: stunning, breathtaking, nestled, in the heart of, renowned, must-have, must-visit, rich (figurative), profound, commitment to, diverse array, groundbreaking (figurative), natural beauty.

The verbatim exemption applies in full: a merchant keyword such as "vibrant home decor" is placed exactly as given, wherever the writer skill requires it.

Inflation: keep the fact, drop the dressing

• Inflated significance: a testament to, pivotal, plays a key role, setting the stage for, evolving landscape — and the send-off paragraph ("The future looks bright", "exciting times ahead"). Keep the fact, drop the significance. A piece ends on its CTA, never on an outlook paragraph.
• Shallow -ing riders: ensuring, reflecting, highlighting, showcasing, fostering bolted onto a fact to make it sound deeper. Before: "crafted with care, ensuring all-day comfort". After: "the footbed is cork, which molds to your foot in about a week."
• Avoiding is / are / has: boasts, features, offers, serves as, stands as, represents. "This jacket boasts a waterproof shell" becomes "The shell is waterproof."
• Borrowed authority: experts say, studies show, follower counts, unnamed industry reports. An unverifiable claim in a merchant's copy is a liability, not polish — cut it rather than soften it; a real named source is cited with an ordinary URL.

Rhythm and blog hygiene

• Forced triads: ideas arriving in threes to sound complete ("innovation, inspiration, and insights"). Keep three items only when the meaning has three parts; otherwise merge them or develop the strongest one.
• Repeated sentence openings (mostly in long blog copy): several sentences in a row starting with the same subject. Merge or vary by ear; do not ban the word.
• A heading repeated in its first sentence: cut the restating line, keep the content.
• Chatbot residue: a draft's body argument contains the piece and nothing else — no "Here's your caption:", no "I hope this helps", no closing offer.

Final pass

Before CREATE_CONTENT_DRAFT, give the piece one silent scan for the tells that most often survive writing: contrasts, repeated closers, triads, dashes, and the stock-word lists — and the swap test: would this sentence still be true in a competitor's post about a similar product? A sentence that would is generic; rewrite it around this product's own detail or cut it. Fix what you find without inventing anything new, and vary sentence length — real writing alternates short and long. One pass, not a scoring loop.

Source: adapted for this workspace from the MIT-licensed humanizer skill (github.com/blader/humanizer), itself based on Wikipedia's "Signs of AI writing".