# Content Anti-Patterns: Your Brand This file catches what your content must **never** do. It derives from `context/voice/core.md` and `context/brand-strategy.md` - when this list and the voice profile disagree, the voice profile wins. The goal isn't zero matches; it's that every match is either rewritten or a deliberate, voice-sanctioned choice. **This file is a HARD GATE, not a style suggestion.** No draft ships and no copy goes on the site until it clears the checklist below. Treat the countable patterns (em dashes, negative parallelism, banned words) like a build step: actually count them, do not eyeball them. Eyeballing is exactly how the antithesis tell slips through a piece whose individual sentences all look fine. Include this file in any drafting or review context. **What counts as a violation:** a single pattern used once might be fine. Flag when (a) several different tropes appear in one piece, (b) one trope repeats, or (c) a pattern contradicts the voice profile's banned lists or register. The stakes are high here: the brand sells *trust over hype*, and the blog is a live demo of the engine. Slop or hype on it falsifies the entire pitch. --- ## Run this checklist against every draft A reviewer should be able to answer "no" to all of these. Any "yes" is a rewrite or a deliberate, justified exception. 1. Does it contain an em dash, anywhere? (Banned. See below.) 2. Does it contain banned agency jargon or hype/overpromise language? 3. Does it get loud, brash, all-caps, or "make some noise"? 4. Is anything glib about patients, mental illness, addiction, or clinical outcomes, or does it use suicide- / self-harm-adjacent idiom or analogy (e.g. "end of their rope," "talk off the ledge")? 5. **Antithesis count first.** Count the negative-parallelism ("not X, it's Y" and its variants, see below) constructions, then divide the word count by 800. Over that budget and it reads AI no matter how good the points are. Count, do not eyeball. Then scan for the other tells: generic opener, listicle padding, hedging, self-posed questions, fake-conversational filler. 6. Does it cite a metric we don't actually have, or attribute a claim to a source it can't name? 7. Is it stuffing keywords, or naming locations as keyword filler? 8. Does it state a compliance specific without a `[NEEDS VERIFICATION]` flag and a real source? 9. Does it break an AEO guardrail (agent-incentive scheme, cloaking, non-identical markdown)? 10. Could this copy be pasted onto any other agency's site unchanged? (If yes, it has no point of view - kill it.) --- ## 1. Banned agency jargon (instant rewrite) Pulled from `voice/core.md` and expanded. Any of these is an automatic rewrite: synergy, move the needle, growth hacking, full-funnel, best-in-class, thought leadership (as a self-label - we produce it, we don't label ourselves with it), ninja / guru / rockstar / wizard, circle back, leverage (as a verb), "solutions," elevate your brand, "in today's fast-paced world," unlock, supercharge, game-changer, secret sauce, drive results (vague - say *which* results), delve, utilize, robust, streamline, harness, tapestry, landscape (as filler), paradigm, ecosystem (as buzzword), bandwidth (as availability), "at the end of the day," "low-hanging fruit," double-click (as "look closer"), "north star" (as filler), boil the ocean. **The deeper rule:** if a sentence needs a buzzword to sound impressive, the idea underneath is thin. Say the plain thing instead. ## 2. Hype and overpromise (the cardinal sin for this brand) the brand's lead pillar is *trust over hype*. Overpromising doesn't just sound off, it makes us the thing we position against. Banned: guaranteed, #1, explode your growth, 10x overnight, skyrocket, crush it, blow up, dominate, game-changing campaign, huge results, results overnight, "we'll make you go viral," "the only X you'll ever need," "transform your business," "revolutionary," "cutting-edge," "next-level," "unprecedented." **The honest alternative (voice-prescribed):** name the big-promise pattern the reader has already been sold, then set the honest expectation - consistent quality that compounds over time. Measured contrast, never a takedown, never a counter-promise that's just a louder claim. ## 3. Loud / brash register (banned) No all-caps shouting. No agitprop, no "make some noise," no "rouse demand" energy (the name is a *wink*, not a volume setting). No exclamation-point stacking. No brutalist or bold register. The confidence is calm; the substance does the work. Get this wrong and the brand breaks: too hype-y and we *become* the slick agency we're positioning against. ## 4. Glib about the mission (banned, no exceptions) The irreverence in the brand's voice points at **us and the marketing industry** - never at the people clients serve. Never joke about, minimize, or be flip about mental illness, addiction, patients, clinicians, or clinical outcomes. Behavioral health is the serious register, always. This is the single tension the whole voice rests on; breaking it is the worst failure on the list. **No suicide- or self-harm-adjacent language, including as idiom or analogy (added 2026-07-15).** Do not reach for figures of speech that evoke suicide, self-harm, or a person at a breaking point - "at the end of their rope," "the far end of their rope," "talk them off the ledge," "jump off a cliff," "dead on arrival," "a cry for help" - anywhere in behavioral-health content, even when the phrase is not describing a patient. In a field where the reader may be a parent or clinician in an acute moment, casual crisis imagery is careless at best and harmful at worst. When one slips in during drafting, cut it and say the literal thing ("exhausted and out of options," "stalled," "unlikely to work"). This holds across every engagement (across every engagement). ## 5. Em dashes (BANNED) Use a hyphen surrounded by spaces ( - ) for every pause, aside, or pivot an em dash would carry. This is a deliberate AI-tell mitigation and house style across our repos: it should read as typed, not typeset. Also banned for the same reason: Unicode arrows, smart/curly quotes, and other characters you can't type on a standard keyboard. --- ## 6. AI-tells and slop patterns to avoid These are the writing defaults that make content read as machine-generated. Because the brand's blog is proof that "AI in the right hands" beats a slop mill, every one of these on our own site is self-refuting. ### Generic openers Never open with "Welcome to," "In today's fast-paced / competitive / digital landscape," a mission statement, company history, or a list of services. **Voice-prescribed opener:** name the reader's frustration or a sharp, honest claim, then the calm contrast, then the proof. ### Negative parallelism (the antithesis - COUNT IT) The single most common AI tell, and the one that makes a whole piece read machine-generated even when every sentence is individually fine. It is a structure, not a word, which is why it slips past keyword checks. Learn to see all its forms: - "It's not X, it's Y." / "It isn't X. It's Y." - "X, not Y." / "not because X, but because Y." - "The question isn't X. The question is Y." - "Instead of X, Y." / "rather than X, Y." / "X without being Y." - Paired antithesis across two sentences: "One does X. The other does Y." / "A polished site reads as distance. A human one reads as care." - Balanced conditionals: "If X... If Y..." **The rule is a count against a word budget, not a vibe.** Budget roughly one deliberate antithesis per 800 words: count them, divide the word count by 800, and that is the ceiling. An 800-word piece gets about one; a 1,600-word piece about two; a 4,000-word piece about five. The voice profile's *acknowledge-the-hype-then-offer-the-honest-alternative* move is the sanctioned use. Modestly over budget is a warning; at roughly double the budget the piece reads AI no matter how good the points are. Density counts too: even inside the budget, do not cluster them - two in one paragraph is the repeat-trope failure regardless of length, and the survivors should be varied in form (a metaphor pair, an aphorism, a thesis line, a closing parallel) and spread out, never the same "isn't X, it's Y" beat over and over. The title and a CTA get one punchy contrast for free; they sit outside the body budget. When in doubt, rewrite the antithesis as a plain declarative and spend the space you save on a concrete, specific detail. Simple negations ("she is not weighing levels of care") do NOT count - the tell is the *paired contrast*, the rhetorical see-saw, not the word "not." ### Self-posed rhetorical questions "The result? Predictable." "The worst part?" Asking a question nobody asked, then answering it. Cut it. ### Listicle padding and bold-first bullets Padding a thin point into a numbered list, "The first... The second... The third..." as fake prose, or every bullet opening with a **bolded label:**. Use real sentences. Bullets earn their place by carrying distinct content, not by formatting decoration. ### Hedging and filler transitions "It's worth noting," "it bears mentioning," "importantly," "interestingly," "notably," "arguably," "in many ways." They signal nothing. Also the magic adverbs: "quietly," "deeply," "fundamentally," "remarkably." Say the thing or cut it. ### Fake-conversational filler "Here's the thing," "here's the kicker," "here's where it gets interesting," "let's unpack this," "let's dive in," "let's break this down," "think of it as," "imagine a world where." Patronizing teacher-mode and false suspense. The reader is a busy operator, not a student. ### Tricolon and anaphora abuse A single rule-of-three is fine; three back-to-back is a pattern failure. Repeating the same sentence opening in quick succession (anaphora) is the same failure. ### Stakes inflation and the signposted conclusion "This will fundamentally reshape everything." No. And no "In conclusion," "To sum up," "To wrap up" - just land the point. ### The enumerated wrap-up (the "two takeaways" tell) A summary list bolted onto the end that restates what the piece just said: "Two takeaways: X, and Y." "Three things to remember." "The lesson here is." "So what does this mean?" It is the signposted conclusion wearing a list, and it is a strong AI tell because a model reaches for a tidy recap instead of trusting the body to have landed. Make the point once, in the body, and stop. If the ending needs a line, make it a single concrete declarative, not an itemized recap. Especially fatal in short formats (a social caption has no room for a recap of three sentences). ### The "serves as" dodge and tacked-on analysis Replacing "is" with "serves as / stands as / represents." And tacking a present-participle phrase onto a sentence to fake insight ("...underscoring its role as a dynamic hub"). Both add words and remove meaning. --- ## 7. Fabricated or unsourced metrics (banned) **Never claim a number we don't have.** No invented Lighthouse scores, traffic lifts, ranking movements, client counts, percentages, or "studies show." The voice profile is explicit: if we don't have the number yet, we don't claim it. This is non-negotiable for a brand whose whole pitch is provable competence; one fabricated figure poisons every real one. - **No vague attributions.** "Experts agree," "research shows," "studies have found" without a named source. If you can't name it, you don't have it. Cite the actual source (a Google Search Central post, a named study, a real client result with permission). - **Real numbers, used sparingly.** A few hard, concrete figures to *prove* a claim, then move on. Not a data dump. - **Client results need permission and accuracy.** Only use outcomes we can stand behind and are cleared to share. ## 8. Keyword stuffing and location filler (banned) the brand's SEO play is depth and high-intent long-tail, not volume (see `seo-clusters.md`). So: - No repeating a target keyword unnaturally to hit a density. Write for the reader; the ranking follows credible depth. - No "we proudly serve [city], [city], [city], and surrounding areas" - service-area info belongs in structured data and the footer, not body copy. - No "[City] is known for..." or "Located in the heart of [City]..." openers. The reader knows where they are. - No FAQ that answers its own questions with keyword-stuffed reformulations. Use questions real prospects actually ask in sales conversations. --- ## 9. AEO guardrails (the legitimate-format-optimization line) the brand's AEO play is serving clean markdown so AI assistants can read and cite us accurately - a legitimate format optimization. It stops there. These are hard lines, not preferences: - **No agent-incentive schemes.** Never embed instructions, incentives, or manipulation aimed at an AI agent or its model. We optimize the *format*, never bribe the reader (human or machine). - **Never cloak.** Never serve different content to Googlebot, an AI crawler, or any bot than we serve to a human. No alternate content, no hidden text. - **Serve identical markdown.** The `.md` version of a page and its `llms.txt` entry must say the same thing the human page says. The markdown is a clean *representation* of the real content, not a separate optimized payload. Breaking any of these would make the brand exactly the kind of manipulative operator the brand exists to be the opposite of. The integrity of the AEO play *is* part of the pitch. --- ## 10. The faceless-vendor test (final gate) If a paragraph could be pasted onto any other agency's homepage and still fit, it has no point of view and it fails. Your content is specific, proof-forward, and unmistakably ours, or it doesn't ship. --- **Source:** Derived from `context/voice/core.md` and `context/brand-strategy.md`. Updates: when the reviewer keeps catching a new pattern, add it here. When a pattern stops appearing, mark it deprecated rather than deleting it (it may return). Voice profile always wins.