[{"_1":2,"_82":-5,"_83":-5},"loaderData",{"_3":4,"_19":20},"root",{"_5":6,"_13":14},"env",{"_7":8,"_9":10,"_11":12},"PUBLIC_APP_ENV","production","PUBLIC_APP_FQDN","www.theadpharm.com","PUBLIC_TURNSTILE_SITE_KEY","0x4AAAAAADD1R3ybTHbU1ywg","preferences",{"_15":16,"_17":18},"sidebarWidth",240,"theme","system","routes/_app.insights.tag.$tag",{"_21":22,"_23":24},"tag","copywriting","posts",[25],{"_26":27,"_28":29,"_30":31,"_74":75,"_76":77,"_78":79,"_80":81},"slug","claude-opus-anti-slop-playbook","category","ai-and-tech","frontmatter",{"_32":33,"_34":35,"_36":37,"_28":29,"_38":39,"_44":45,"_47":48,"_49":48,"_50":51,"_52":53,"_54":55,"_61":62},"title","How to get Claude Opus 4.7 to write copy that doesn't sound like AI","description","A working playbook for stopping Claude from defaulting to LinkedIn-thinkpiece prose: the leverage points, the banned-words block, the three-pass workflow, and the before/after.","publishedAt","2026-04-03","tags",[40,41,22,42,43],"claude","prompting","anti-slop","opus-4-7","authors",[46],"ben-honda","draft",false,"featured","provenance","ai-research-with-human-curation","ogEyebrow","AI & Tech","keywords",[56,57,58,59,60],"claude opus 4.7 prompting","anti-slop prompt","claude custom instructions copywriting","humanizer claude","style transfer claude","faq",[63,68,71],{"_64":65,"_66":67},"question","What is the single biggest leverage point for stopping Claude from sounding like AI?","answer","Move the rules out of the prompt and into Claude's Custom Instructions, the Project's knowledge, or a Custom Style. A pasted 200–600-word writing sample anchored in a Style outperforms a 500-word prompt. The prompt should carry the brief, not the guardrails.",{"_64":69,"_66":70},"Does naming a copywriter in the prompt actually change Claude's output?","On its own, only marginally. About ten per cent of the gain. The other ninety comes from pasting 200–600 words of that writer's actual prose and asking Claude to list five observed voice characteristics before drafting. The 'list five first' step is the active ingredient.",{"_64":72,"_66":73},"Why won't a single prompt fix AI tells?","Claude cannot reliably catch its own slop in one pass. The pros run three: draft without self-editing, critique against an explicit list of tells (banned vocabulary, em dashes, tricolons, negative parallelism), then rewrite against the critique. It costs about three times the tokens and is the difference between a draft you have to fix and a draft you would ship with a light edit.","html","

\n
Don’t put guardrails in every prompt: install them once in Custom Instructions or a Project. Paste 200–600 words of a chosen writer’s prose, not just their name. Run three passes on every draft (write, critique, rewrite). Fix the negative-parallelism, tricolon, and em-dash trio first; it’s most of the tell.
\n

A tactical playbook for stopping Claude from sounding like a LinkedIn thought leader on espresso. Industry-agnostic, copy-paste, edit. The whole thing assumes you already know how to write; you just want the model to stop performing.

Why Claude defaults to slop

Claude is not trying to sound like a tech-bro. It is a probability machine. Give it a vague brief and it lands on the statistical centre of all marketing prose ever scraped, which means LinkedIn, SaaS landing pages, and Medium thinkpieces. That centre is the slop.

Wikipedia’s editors maintain a 15,000-word document called Signs of AI writing cataloguing the tells. The blader/humanizer skill on GitHub (13.8k stars) distils it to twenty-nine patterns. Here is the short version of what actually trips the alarm:

Vocabulary. Delve, dive into, navigate (figurative), underscore, bolster, foster, harness, leverage, unpack, shed light on, pave the way. Pivotal, groundbreaking, cutting-edge, transformative, game-changing, innovative, robust, comprehensive, seamless, intricate, nuanced, multifaceted, holistic. Testament, landscape (figurative), realm, tapestry, ecosystem. Discover, Unlock, Elevate, Transform, Empower, Reimagine, Supercharge. The copula avoiders (serves as, functions as, stands as, acts as) used in place of plain is.

Structure. Negative parallelism (“It’s not just X — it’s Y”). Tricolons (“innovative, transformative, groundbreaking”). Mirror sentences (“The tool is a catalyst. The assistant is a partner. The system is a foundation.”). Em dashes used for pseudo-emphasis where a comma would do. Bold-term-colon bullet lists where every item has the same shape. Setup-then-pivot openings (“In today’s fast-paced world…”). Question-then-answer rhythm (“What does this mean? It means…”). Inflated symbolism via -ing analyses (“symbolising… reflecting… showcasing…”).

Tone. Sycophantic preamble (“Great question!”). Wrap-up coda (“In conclusion, the future looks bright”). Promotional travel-brochure register (“nestled at the intersection of…”). Hedging stacks (“could potentially possibly”). Throat-clearing (“It’s worth noting that…”, “At its core…”).

Format, which Opus 4.7 still does without instruction. Title Case Headings. Excessive bolding. Emoji headers. A three-bullet summary at the end of everything.

If you only fix one thing, fix the negative-parallelism, tricolon, and em-dash trio. Those three patterns alone produce roughly seventy per cent of the obvious tell.

The single best leverage point

Do not put your guardrails in every prompt. Put them once, where Claude reads them automatically. There is a hierarchy:

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

Layer	Where	What goes there
Global	Settings → Profile (Custom Instructions)	Banned-words list, voice defaults
Per-client	Projects → Project Knowledge + Custom Instructions	Brand voice doc, three to five writing samples, audience, tone rules
Per-task	Custom Style (Create & Edit Styles)	A specific stylistic preset, e.g. “Plain-spoken landing copy”
Per-message	Prompt	The brief: what, who, length

In Opus 4.7 the Style feature is the one most people ignore. Anthropic’s own docs confirm 4.7 needs less anti-slop scaffolding than 4.5 or 4.6, but only if it has something to anchor to. A Style with two pasted writing samples beats a 500-word prompt every time.

The drop-in Custom Instructions block

Paste this into Settings → Profile, or the top of any Project. Adapted from Will Francis’s widely-shared block, the blader/humanizer pattern list, and lguz/humanize-writing-skill. It is the baseline; layer voice on top.

## Voice\nBe direct. Have opinions. State the point first, then support it. Trust the reader.\nDon't label things \"important\" — show why. Use specific nouns and verbs, not abstractions.\n \n## Banned words (do not use)\ndelve, dive into, navigate (figurative), underscore, bolster, foster, harness,\nleverage, unpack, shed light on, pave the way, pivotal, groundbreaking,\ncutting-edge, transformative, game-changing, innovative, robust, comprehensive,\nseamless, intricate, multifaceted, holistic, testament, landscape (figurative),\nrealm, tapestry, ecosystem, vibrant, empower, elevate, unlock, supercharge,\nreimagine, journey (figurative), at its core, in today's world\n \n## Banned phrases\n- \"In today's [fast-paced / rapidly evolving / digital] world…\"\n- \"It's worth noting that…\" / \"It's important to note…\"\n- \"When it comes to…\" / \"At the end of the day…\"\n- \"This is where X comes in.\" / \"Let's break it down.\"\n- \"Plays a crucial role in…\" / \"It cannot be overstated…\"\n- \"Whether you're X or Y…\"\n- Any sentence beginning with \"Imagine…\"\n \n## Banned structures\n- \"It's not just X — it's Y.\"\n- \"Not only X, but also Y.\"\n- \"This isn't about X. It's about Y.\"\n- \"No X. No Y. Just Z.\" (the three-fragment kicker)\n- Question → answer pairs as a rhythm device\n- Three-item lists where two would do; never stack three adjectives\n- Mirror sentences: \"The tool does X. The system does Y. The platform does Z.\"\n \n## Punctuation & format\n- Maximum one em dash per page. Prefer commas, periods, parentheses.\n- No emojis in headers. No Title Case in headers — sentence case only.\n- Don't bold key terms inside bullets. Don't open every bullet with a bold colon.\n- Use contractions. \"It's,\" \"don't,\" \"won't.\"\n \n## Prose discipline\n- Vary sentence length on purpose: short, then medium, then a longer one that\n  takes its time and works through the idea, then short again.\n- Don't open with a sweeping context statement. Start on substance.\n- Don't end with a wrap-up or \"the future looks bright\" coda. Stop when the\n  idea stops.\n- Cut copula avoidance: \"is\" beats \"serves as\" 95% of the time.\n- No unsolicited preamble or sign-off.\n \n## Self-check before returning\n1. Read it aloud. Anything sound like a press release? Rewrite.\n2. Is the first sentence a setup? Delete it. Start at sentence two.\n3. Is the last sentence a summary? Delete it.\n4. Count em dashes. Reduce to one.\n5. Count tricolons. Replace at least one with a pair or a single item.

This block alone removes about eighty per cent of the obvious slop. The remaining twenty per cent is voice.

Pick a copywriter and paste their prose

Naming a copywriter in the prompt nudges Claude, but vaguely. Naming a copywriter and pasting 200–600 words of their actual prose is what shifts the output. Opus 4.7 does style transfer well from samples and badly from names alone.

A short list of distinguished references that produce a clear style shift in testing:

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

Reference	What it gives you	Use when you want
David Ogilvy (Ogilvy on Advertising, the Rolls-Royce ad)	Confident specifics, factual claims, no fluff	Authoritative, considered B2B or premium consumer
Bill Bernbach (VW “Think Small,” Avis “We Try Harder”)	Wit, self-deprecation, brevity, respect for the reader	Brands that want charm and an underdog stance
Dave Trott (Predatory Thinking)	Punchy short paragraphs, plain English, story-led	Editorial brand writing without sounding “punchy AI”
Joe Sugarman (The Adweek Copywriting Handbook)	Slippery-slide rhythm, conversational	Long-form sales pages, founder letters
Ann Handley (Everybody Writes)	Warm, plain, useful, lightly literary	Content marketing that isn’t “content marketing”
George Orwell (“Politics and the English Language”)	Plain words, active verbs, no ornament	Anything where you want the model to tighten up

The Orwell trick is worth its own line. Paste this into your system prompt:

\n
Apply Orwell’s six rules from “Politics and the English Language.” Especially: never use a long word where a short one will do; if it is possible to cut a word out, cut it out; never use the passive where you can use the active.
\n

The output tightens noticeably. Citing the essay is the cheapest single intervention in this whole playbook.

The three-sample rule

Naming a writer does about ten per cent of the work. Pasting samples does the other ninety. A template that drops cleanly into a Project or Style:

You are writing in the voice of [NAME or \"this brand\"]. Below are three short\npassages. Study them for: sentence-length variance, word choice, what they\nrefuse to say, where they break rhythm, where they use specifics instead of\nabstractions, what their default verb tense is, and whether they use first\nperson.\n \nDO NOT imitate the topic. Imitate the voice.\n \n<sample_1>\n[paste 150–300 words]\n</sample_1>\n \n<sample_2>\n[paste 150–300 words]\n</sample_2>\n \n<sample_3>\n[paste 150–300 words]\n</sample_3>\n \nBefore drafting, list the five voice characteristics you observed. Then write.

The “list the five characteristics first” step is the active ingredient. It forces Claude to extract the style explicitly instead of regressing to its mean. Skip it and you lose about half the gain.

Words to remove from your own prompt

These are the things people put in briefs that reliably degrade the output. Strike them:

Engaging. Pulls Claude toward exclamation-marked enthusiasm.
Compelling. Triggers superlative stacking.
Punchy. Triggers fragment-spam and three-word sentences.
Modern, fresh, dynamic. Activates the SaaS-landing-page register.
Concise but powerful. Produces gnomic Yoda-LinkedIn.
Use vivid imagery. Triggers metaphor inflation.
Optimised for conversion. Generic AIDA bilge.
Professional yet approachable. The most AI-coded phrase of all; produces the exact register you’re trying to avoid.
Add some personality. Triggers forced quirkiness.
Em dashes in your own prompt. Claude mirrors your punctuation.

Replace each with a concrete constraint:

Instead of “engaging,” write “the reader should want to read sentence two after reading sentence one.”
Instead of “punchy,” write “average sentence length 14 words; vary between 4 and 28.”
Instead of “professional yet approachable,” write “the way [specific writer] writes; see samples.”
Instead of “compelling,” write “include one specific number, one named thing, and one concession.”

The three-pass workflow

A single prompt cannot fix slop. The pros run three passes, each with one job. This is the structure used by both blader/humanizer and lguz/humanize-writing-skill, and by every working agency workflow.

Pass 1. Draft, do not optimise.

Write a [page type] for [brand] aimed at [reader]. The reader's situation:\n[one sentence]. The thing we want them to do: [one verb + object].\nConstraints: [length], [must-mention facts]. Voice: see Project samples.\n \nDon't self-edit. Don't add a conclusion. Don't apologize. Stop when the idea stops.

Pass 2. Critique. Claude finds its own slop.

You are a hostile editor whose job is to detect AI tells. Read the draft below\nand produce a numbered list of every instance of:\n \n1. Negative parallelism (\"not X, it's Y\")\n2. Tricolons / three-adjective stacks\n3. Em dashes (count them)\n4. Banned vocabulary (delve, leverage, transformative, etc.)\n5. Sentences that begin with \"In today's…\" or \"Imagine…\" or \"Whether you're…\"\n6. Copula avoidance (\"serves as,\" \"stands as,\" \"acts as\")\n7. Bold-term-colon list items\n8. Wrap-up codas in the final paragraph\n9. Sentences that could be cut entirely without losing meaning\n10. Any place where a specific noun, name, or number could replace an abstraction\n \nQuote each offender verbatim. Do not rewrite yet.\n \n<draft>\n[paste pass 1]\n</draft>

Pass 3. Rewrite against the critique.

Now rewrite the draft, fixing every issue in your critique. Additional rules:\n \n- Replace at least three abstractions with specific nouns or numbers.\n- Vary sentence length: include at least one sentence under 6 words and one over 25.\n- Cut the opening sentence if it's setup. Cut the closing sentence if it's a summary.\n- Keep the meaning. Don't soften the claims.\n \nReturn only the rewritten copy.

This three-pass loop, run inside one Claude conversation, is the difference between “AI draft I have to fix” and “draft I’d put my name on with a light edit.” It costs about three times the tokens. Worth it.

The faster alternative is to install the blader/humanizer skill in Claude.ai (Customize → Skills → Upload, point at the GitHub zip), then end any prompt with use the humanizer skill. You get pass 2 and pass 3 free.

A drop-in template that works for most briefs

Eight templates is too many to remember. The one that earns its keep is the distinguished-copywriter prompt:

Draft this in the spirit of Dave Trott writing for [BRAND]. Trott's voice:\nshort paragraphs, plain English, one idea per paragraph, anecdote-led, no\nadjectives where a verb will do, never sells directly, lets the reader reach\nthe conclusion themselves.\n \nBefore drafting, list five things Trott would NOT do that a typical SaaS\nlanding page does. Then write.\n \nBrief: [BRIEF]\nLength: [WORDS]\nMust include: [FACTS/PROOF POINTS]

Swap Trott for Bernbach, Ogilvy, Sugarman, Wiebe, Handley as the brief calls for. The “list five things they would NOT do first” step is the active ingredient. Same mechanism as the three-sample rule, applied to a single named writer.

Holding the voice over a long session

Even with all the above, Claude drifts back toward its default register over a long session. Three techniques hold the line. First, re-anchor every four or five turns by pasting a reference sample again with the instruction recalibrate; the voice should match this sample, not the previous outputs; identify three ways the last response drifted, then redo it. Second, run the two-voice test: ask Claude to write the same paragraph twice, once in the brand voice and once in default Claude voice, then list every difference. It forces the model to notice and protect the voice in subsequent turns. Third, repeat critical rules as the last line of the prompt, not buried in the system prompt; the position weights more there.

Before and after, so you can see it

Brief. Hero copy for a project management tool aimed at small agency owners.

Default Claude Opus 4.7, no instructions:

\n
Reimagine How Your Agency Delivers
\n
The all-in-one platform that empowers creative teams to streamline workflows, foster collaboration, and deliver standout work — every time. It’s not just project management; it’s the operating system your agency has been waiting for.
\n
Get Started Free
\n

Same model, with the Custom Instructions block, a Trott-style anchor, and the three-pass workflow:

\n
Your projects don’t run late. Your replies do.
\n
Most agency tools track tasks. This one tracks the bottleneck — the email you owe a client, the brief still sitting in your drafts, the approval nobody’s chased. Six agencies tested it for ninety days. Average reply time fell from 31 hours to 4.
\n
Try it on one project
\n

Same product. Same model. Different scaffolding. The second one wouldn’t embarrass you.

A note on detection scores

Most of this is stable: the Wikipedia Signs of AI writing taxonomy, the blader/humanizer pattern list, Anthropic’s own prompt-engineering docs (which now mention the AI slop aesthetic by name and confirm Opus 4.7 needs less anti-slop scaffolding than 4.5 or 4.6, but still benefits from style anchors). Skills, Projects, and Styles are documented features in claude.ai today.

What is contested is detection. Multiple humaniser-tool vendors claim their output bypasses GPTZero and Originality; independent tests put Claude’s own self-rewrite at around 48% AI on GPTZero¹. If your goal is evading detection, the prompting techniques here are not sufficient on their own. If your goal is not embarrassing yourself in front of clients who know good copy, they are more than enough.

Bookmark the page, paste the Custom Instructions block once, and come back when output starts drifting.

\n
Independent comparison reported in Can Claude Humanize Text? We Tested Anthropic’s AI Against 3 Detectors, thehumanizeai.pro. Claude’s best self-rewrite scored 48% AI on GPTZero (compared to ChatGPT’s best of 61%); both are still failing scores by GPTZero’s threshold. ↩
\n

","rawMarkdown","\n> TL;DR: Don't put guardrails in every prompt: install them once in Custom Instructions or a Project. Paste 200–600 words of a chosen writer's prose, not just their name. Run three passes on every draft (write, critique, rewrite). Fix the negative-parallelism, tricolon, and em-dash trio first; it's most of the tell.\n\nA tactical playbook for stopping Claude from sounding like a LinkedIn thought leader on espresso. Industry-agnostic, copy-paste, edit. The whole thing assumes you already know how to write; you just want the model to stop performing.\n\n## Why Claude defaults to slop\n\nClaude is not trying to sound like a tech-bro. It is a probability machine. Give it a vague brief and it lands on the statistical centre of all marketing prose ever scraped, which means LinkedIn, SaaS landing pages, and Medium thinkpieces. That centre is the slop.\n\nWikipedia's editors maintain a 15,000-word document called [*Signs of AI writing*](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing) cataloguing the tells. The [blader/humanizer](https://github.com/blader/humanizer) skill on GitHub (13.8k stars) distils it to twenty-nine patterns. Here is the short version of what actually trips the alarm:\n\n**Vocabulary.** Delve, dive into, navigate (figurative), underscore, bolster, foster, harness, leverage, unpack, shed light on, pave the way. Pivotal, groundbreaking, cutting-edge, transformative, game-changing, innovative, robust, comprehensive, seamless, intricate, nuanced, multifaceted, holistic. Testament, landscape (figurative), realm, tapestry, ecosystem. Discover, Unlock, Elevate, Transform, Empower, Reimagine, Supercharge. The copula avoiders (*serves as*, *functions as*, *stands as*, *acts as*) used in place of plain *is*.\n\n**Structure.** Negative parallelism (\"It's not just X — it's Y\"). Tricolons (\"innovative, transformative, groundbreaking\"). Mirror sentences (\"The tool is a catalyst. The assistant is a partner. The system is a foundation.\"). Em dashes used for pseudo-emphasis where a comma would do. Bold-term-colon bullet lists where every item has the same shape. Setup-then-pivot openings (\"In today's fast-paced world…\"). Question-then-answer rhythm (\"What does this mean? It means…\"). Inflated symbolism via -ing analyses (\"symbolising… reflecting… showcasing…\").\n\n**Tone.** Sycophantic preamble (\"Great question!\"). Wrap-up coda (\"In conclusion, the future looks bright\"). Promotional travel-brochure register (\"nestled at the intersection of…\"). Hedging stacks (\"could potentially possibly\"). Throat-clearing (\"It's worth noting that…\", \"At its core…\").\n\n**Format, which Opus 4.7 still does without instruction.** Title Case Headings. Excessive bolding. Emoji headers. A three-bullet summary at the end of everything.\n\nIf you only fix one thing, fix the negative-parallelism, tricolon, and em-dash trio. Those three patterns alone produce roughly seventy per cent of the obvious tell.\n\n## The single best leverage point\n\nDo not put your guardrails in every prompt. Put them once, where Claude reads them automatically. There is a hierarchy:\n\n| Layer | Where | What goes there |\n|---|---|---|\n| Global | Settings → Profile (Custom Instructions) | Banned-words list, voice defaults |\n| Per-client | Projects → Project Knowledge + Custom Instructions | Brand voice doc, three to five writing samples, audience, tone rules |\n| Per-task | Custom Style (Create & Edit Styles) | A specific stylistic preset, e.g. \"Plain-spoken landing copy\" |\n| Per-message | Prompt | The brief: what, who, length |\n\nIn Opus 4.7 the Style feature is the one most people ignore. [Anthropic's own docs](https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering) confirm 4.7 needs less anti-slop scaffolding than 4.5 or 4.6, but only if it has something to anchor to. A Style with two pasted writing samples beats a 500-word prompt every time.\n\n## The drop-in Custom Instructions block\n\nPaste this into Settings → Profile, or the top of any Project. Adapted from [Will Francis's widely-shared block](https://willfrancis.com/how-to-stop-claude-writing-like-an-ai/), the [blader/humanizer](https://github.com/blader/humanizer) pattern list, and [lguz/humanize-writing-skill](https://github.com/lguz/humanize-writing-skill). It is the baseline; layer voice on top.\n\n```\n## Voice\nBe direct. Have opinions. State the point first, then support it. Trust the reader.\nDon't label things \"important\" — show why. Use specific nouns and verbs, not abstractions.\n\n## Banned words (do not use)\ndelve, dive into, navigate (figurative), underscore, bolster, foster, harness,\nleverage, unpack, shed light on, pave the way, pivotal, groundbreaking,\ncutting-edge, transformative, game-changing, innovative, robust, comprehensive,\nseamless, intricate, multifaceted, holistic, testament, landscape (figurative),\nrealm, tapestry, ecosystem, vibrant, empower, elevate, unlock, supercharge,\nreimagine, journey (figurative), at its core, in today's world\n\n## Banned phrases\n- \"In today's [fast-paced / rapidly evolving / digital] world…\"\n- \"It's worth noting that…\" / \"It's important to note…\"\n- \"When it comes to…\" / \"At the end of the day…\"\n- \"This is where X comes in.\" / \"Let's break it down.\"\n- \"Plays a crucial role in…\" / \"It cannot be overstated…\"\n- \"Whether you're X or Y…\"\n- Any sentence beginning with \"Imagine…\"\n\n## Banned structures\n- \"It's not just X — it's Y.\"\n- \"Not only X, but also Y.\"\n- \"This isn't about X. It's about Y.\"\n- \"No X. No Y. Just Z.\" (the three-fragment kicker)\n- Question → answer pairs as a rhythm device\n- Three-item lists where two would do; never stack three adjectives\n- Mirror sentences: \"The tool does X. The system does Y. The platform does Z.\"\n\n## Punctuation & format\n- Maximum one em dash per page. Prefer commas, periods, parentheses.\n- No emojis in headers. No Title Case in headers — sentence case only.\n- Don't bold key terms inside bullets. Don't open every bullet with a bold colon.\n- Use contractions. \"It's,\" \"don't,\" \"won't.\"\n\n## Prose discipline\n- Vary sentence length on purpose: short, then medium, then a longer one that\n takes its time and works through the idea, then short again.\n- Don't open with a sweeping context statement. Start on substance.\n- Don't end with a wrap-up or \"the future looks bright\" coda. Stop when the\n idea stops.\n- Cut copula avoidance: \"is\" beats \"serves as\" 95% of the time.\n- No unsolicited preamble or sign-off.\n\n## Self-check before returning\n1. Read it aloud. Anything sound like a press release? Rewrite.\n2. Is the first sentence a setup? Delete it. Start at sentence two.\n3. Is the last sentence a summary? Delete it.\n4. Count em dashes. Reduce to one.\n5. Count tricolons. Replace at least one with a pair or a single item.\n```\n\nThis block alone removes about eighty per cent of the obvious slop. The remaining twenty per cent is voice.\n\n## Pick a copywriter and paste their prose\n\nNaming a copywriter in the prompt nudges Claude, but vaguely. Naming a copywriter and pasting 200–600 words of their actual prose is what shifts the output. Opus 4.7 does style transfer well from samples and badly from names alone.\n\nA short list of distinguished references that produce a clear style shift in testing:\n\n| Reference | What it gives you | Use when you want |\n|---|---|---|\n| **David Ogilvy** (*Ogilvy on Advertising*, the Rolls-Royce ad) | Confident specifics, factual claims, no fluff | Authoritative, considered B2B or premium consumer |\n| **Bill Bernbach** (VW \"Think Small,\" Avis \"We Try Harder\") | Wit, self-deprecation, brevity, respect for the reader | Brands that want charm and an underdog stance |\n| **Dave Trott** (*Predatory Thinking*) | Punchy short paragraphs, plain English, story-led | Editorial brand writing without sounding \"punchy AI\" |\n| **Joe Sugarman** (*The Adweek Copywriting Handbook*) | Slippery-slide rhythm, conversational | Long-form sales pages, founder letters |\n| **Ann Handley** (*Everybody Writes*) | Warm, plain, useful, lightly literary | Content marketing that isn't \"content marketing\" |\n| **George Orwell** (\"Politics and the English Language\") | Plain words, active verbs, no ornament | Anything where you want the model to tighten up |\n\nThe Orwell trick is worth its own line. Paste this into your system prompt:\n\n> *Apply Orwell's six rules from \"Politics and the English Language.\" Especially: never use a long word where a short one will do; if it is possible to cut a word out, cut it out; never use the passive where you can use the active.*\n\nThe output tightens noticeably. Citing the essay is the cheapest single intervention in this whole playbook.\n\n## The three-sample rule\n\nNaming a writer does about ten per cent of the work. Pasting samples does the other ninety. A template that drops cleanly into a Project or Style:\n\n```\nYou are writing in the voice of [NAME or \"this brand\"]. Below are three short\npassages. Study them for: sentence-length variance, word choice, what they\nrefuse to say, where they break rhythm, where they use specifics instead of\nabstractions, what their default verb tense is, and whether they use first\nperson.\n\nDO NOT imitate the topic. Imitate the voice.\n\n\n[paste 150–300 words]\n\n\n\n[paste 150–300 words]\n\n\n\n[paste 150–300 words]\n\n\nBefore drafting, list the five voice characteristics you observed. Then write.\n```\n\nThe \"list the five characteristics first\" step is the active ingredient. It forces Claude to extract the style explicitly instead of regressing to its mean. Skip it and you lose about half the gain.\n\n## Words to remove from your own prompt\n\nThese are the things people put in briefs that reliably degrade the output. Strike them:\n\n- *Engaging.* Pulls Claude toward exclamation-marked enthusiasm.\n- *Compelling.* Triggers superlative stacking.\n- *Punchy.* Triggers fragment-spam and three-word sentences.\n- *Modern, fresh, dynamic.* Activates the SaaS-landing-page register.\n- *Concise but powerful.* Produces gnomic Yoda-LinkedIn.\n- *Use vivid imagery.* Triggers metaphor inflation.\n- *Optimised for conversion.* Generic AIDA bilge.\n- *Professional yet approachable.* The most AI-coded phrase of all; produces the exact register you're trying to avoid.\n- *Add some personality.* Triggers forced quirkiness.\n- Em dashes in your own prompt. Claude mirrors your punctuation.\n\nReplace each with a concrete constraint:\n\n- Instead of \"engaging,\" write \"the reader should want to read sentence two after reading sentence one.\"\n- Instead of \"punchy,\" write \"average sentence length 14 words; vary between 4 and 28.\"\n- Instead of \"professional yet approachable,\" write \"the way [specific writer] writes; see samples.\"\n- Instead of \"compelling,\" write \"include one specific number, one named thing, and one concession.\"\n\n## The three-pass workflow\n\nA single prompt cannot fix slop. The pros run three passes, each with one job. This is the structure used by both [blader/humanizer](https://github.com/blader/humanizer) and [lguz/humanize-writing-skill](https://github.com/lguz/humanize-writing-skill), and by every working agency workflow.\n\n**Pass 1. Draft, do not optimise.**\n\n```\nWrite a [page type] for [brand] aimed at [reader]. The reader's situation:\n[one sentence]. The thing we want them to do: [one verb + object].\nConstraints: [length], [must-mention facts]. Voice: see Project samples.\n\nDon't self-edit. Don't add a conclusion. Don't apologize. Stop when the idea stops.\n```\n\n**Pass 2. Critique. Claude finds its own slop.**\n\n```\nYou are a hostile editor whose job is to detect AI tells. Read the draft below\nand produce a numbered list of every instance of:\n\n1. Negative parallelism (\"not X, it's Y\")\n2. Tricolons / three-adjective stacks\n3. Em dashes (count them)\n4. Banned vocabulary (delve, leverage, transformative, etc.)\n5. Sentences that begin with \"In today's…\" or \"Imagine…\" or \"Whether you're…\"\n6. Copula avoidance (\"serves as,\" \"stands as,\" \"acts as\")\n7. Bold-term-colon list items\n8. Wrap-up codas in the final paragraph\n9. Sentences that could be cut entirely without losing meaning\n10. Any place where a specific noun, name, or number could replace an abstraction\n\nQuote each offender verbatim. Do not rewrite yet.\n\n\n[paste pass 1]\n\n```\n\n**Pass 3. Rewrite against the critique.**\n\n```\nNow rewrite the draft, fixing every issue in your critique. Additional rules:\n\n- Replace at least three abstractions with specific nouns or numbers.\n- Vary sentence length: include at least one sentence under 6 words and one over 25.\n- Cut the opening sentence if it's setup. Cut the closing sentence if it's a summary.\n- Keep the meaning. Don't soften the claims.\n\nReturn only the rewritten copy.\n```\n\nThis three-pass loop, run inside one Claude conversation, is the difference between \"AI draft I have to fix\" and \"draft I'd put my name on with a light edit.\" It costs about three times the tokens. Worth it.\n\nThe faster alternative is to install the [blader/humanizer](https://github.com/blader/humanizer) skill in Claude.ai (Customize → Skills → Upload, point at the GitHub zip), then end any prompt with `use the humanizer skill`. You get pass 2 and pass 3 free.\n\n## A drop-in template that works for most briefs\n\nEight templates is too many to remember. The one that earns its keep is the distinguished-copywriter prompt:\n\n```\nDraft this in the spirit of Dave Trott writing for [BRAND]. Trott's voice:\nshort paragraphs, plain English, one idea per paragraph, anecdote-led, no\nadjectives where a verb will do, never sells directly, lets the reader reach\nthe conclusion themselves.\n\nBefore drafting, list five things Trott would NOT do that a typical SaaS\nlanding page does. Then write.\n\nBrief: [BRIEF]\nLength: [WORDS]\nMust include: [FACTS/PROOF POINTS]\n```\n\nSwap Trott for Bernbach, Ogilvy, Sugarman, Wiebe, Handley as the brief calls for. The \"list five things they would NOT do first\" step is the active ingredient. Same mechanism as the three-sample rule, applied to a single named writer.\n\n## Holding the voice over a long session\n\nEven with all the above, Claude drifts back toward its default register over a long session. Three techniques hold the line. First, re-anchor every four or five turns by pasting a reference sample again with the instruction *recalibrate; the voice should match this sample, not the previous outputs; identify three ways the last response drifted, then redo it*. Second, run the two-voice test: ask Claude to write the same paragraph twice, once in the brand voice and once in default Claude voice, then list every difference. It forces the model to notice and protect the voice in subsequent turns. Third, repeat critical rules as the last line of the prompt, not buried in the system prompt; the position weights more there.\n\n## Before and after, so you can see it\n\n**Brief.** Hero copy for a project management tool aimed at small agency owners.\n\n**Default Claude Opus 4.7, no instructions:**\n> ### Reimagine How Your Agency Delivers\n> The all-in-one platform that empowers creative teams to streamline workflows, foster collaboration, and deliver standout work — every time. It's not just project management; it's the operating system your agency has been waiting for.\n>\n> **Get Started Free**\n\n**Same model, with the Custom Instructions block, a Trott-style anchor, and the three-pass workflow:**\n> ### Your projects don't run late. Your replies do.\n> Most agency tools track tasks. This one tracks the bottleneck — the email you owe a client, the brief still sitting in your drafts, the approval nobody's chased. Six agencies tested it for ninety days. Average reply time fell from 31 hours to 4.\n>\n> **Try it on one project**\n\nSame product. Same model. Different scaffolding. The second one wouldn't embarrass you.\n\n## A note on detection scores\n\nMost of this is stable: the Wikipedia [*Signs of AI writing*](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing) taxonomy, the [blader/humanizer](https://github.com/blader/humanizer) pattern list, Anthropic's own [prompt-engineering docs](https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering) (which now mention the AI slop aesthetic by name and confirm Opus 4.7 needs less anti-slop scaffolding than 4.5 or 4.6, but still benefits from style anchors). Skills, Projects, and Styles are documented features in claude.ai today.\n\nWhat is contested is detection. Multiple humaniser-tool vendors claim their output bypasses [GPTZero](https://gptzero.me) and [Originality](https://originality.ai); independent tests put Claude's own self-rewrite at around 48% AI on [GPTZero](https://gptzero.me)[^claude-self-rewrite]. If your goal is evading detection, the prompting techniques here are not sufficient on their own. If your goal is not embarrassing yourself in front of clients who know good copy, they are more than enough.\n\nBookmark the page, paste the Custom Instructions block once, and come back when output starts drifting.\n\n[^claude-self-rewrite]: Independent comparison reported in *Can Claude Humanize Text? We Tested Anthropic's AI Against 3 Detectors*, [thehumanizeai.pro](https://thehumanizeai.pro/articles/can-claude-humanize-text). Claude's best self-rewrite scored 48% AI on GPTZero (compared to ChatGPT's best of 61%); both are still failing scores by GPTZero's threshold.\n","readingTimeMin",12,"url","/insights/claude-opus-anti-slop-playbook","actionData","errors"]