Direct response copy

Adapted from the direct-response-copy skill by Rob Palmer (https://github.com/robpalmer99/claude-code-copywriting-skills), used under CC BY 4.0.

scripts/check-copy/rules/formats/direct-response.mjs is the actual gate for this format. This skill explains what it checks, why each check exists, and what craft produces copy that clears it on the first pass instead of the fourth. Reading this list is never a substitute for running the script. A sales page ships because the process exited 0. An agent reading it back to itself and liking it is a different thing.

Invoke the gate on this format with:

node scripts/check-copy/index.mjs <file> --format direct-response

Scope

This format covers LONG-FORM persuasive copy whose job is a measurable action from the reader: sales pages, VSL and TSL scripts, email copy, headlines, and the CTAs that close them. Two neighbours are deliberately out of scope. A short bridge page, where the constraint is compression rather than sustained persuasion, belongs to landing-page-copy and its --format landing-page contract. Meta ad copy, where character caps and platform policy do most of the enforcing, belongs to ad-copy and --format ad. A single post on X belongs to tweet-copy. Pick the format that matches where the copy ships, because each module checks a different thing.

What the gate checks

Five rules fire from the direct-response module. Four are errors and one is a warning, which tells you what the format is strict about: making a specific claim, and asking for the action.

• format/missing-cta (error). No call-to-action verb appears anywhere in the body. The pattern looks for the verbs that actually ask: click, tap, call, order, buy, get, grab, start, try, book, claim, download, join, subscribe, reply, register, reserve. Direct response is defined by asking for a specific action; copy that describes an offer and then trails off is brand copy wearing a sales page's clothes. Name the action.
• format/missing-specificity (error). Copy running over 100 words with no number anywhere in it: no price, no timeframe, no result, no count. This is an error rather than a warning because specificity is the discipline. Every other rule in this module can be argued against from some register or another, but unquantified copy has no mechanism by which it could persuade. "Cleans almost all bacteria" and "cleans 99.6% of bacteria" are the same claim; only one of them is believed. A page with zero numbers is asserting things it never grounds, and no amount of rhythm fixes that. Warning severity would let it ship.
• format/vague-benefit (error). Nine phrases that occupy the slot where a concrete outcome belongs. Four are self-flattery with no referent: best-in-class, world-class, state-of-the-art, industry-leading. Two duck the question of who the product is for: businesses of all sizes, achieve your goals. Three are stock aspiration filler: the "next level" escalator, the "potential" one about to be released, and the "new heights" one about to be reached. The module's VAGUE_BENEFITS array carries the exact patterns; read it there rather than trusting this paraphrase. Each fires per occurrence with a sentence-grain excerpt. These are not weak words to be strengthened. They are placeholders that were never filled in, and the fix is always the same shape: put the concrete outcome and the number behind it into the slot.
• format/missing-proof (warning). Copy over 150 words with no testimonial, named source, quoted result, or cited study. The check looks for attribution verbs (said, says, told, according to), the words study, survey, customers, clients, users, or any quoted run of 20 or more characters. Proof carries the middle of long-form copy. A specific claim with nothing behind it is still just a claim, and past a certain length the reader has had time to notice. This is a warning because a genuinely short offer email can be honest without a testimonial block; weigh it against the brief rather than padding a quote in to silence it.
• format/unqualified-claim (error). The shared claims list, run on this format because the legal exposure here is real. It fires on guaranteed outcomes ("guaranteed you'll", "guarantee to"), absolute risk-free promises ("100% risk-free", "completely guaranteed"), "zero risk", results presented as typical, make-money-fast, get-rich-quick, no-effort-required framings, cure claims ("cures your", "cures any"), and specific weight-loss claims ("lose 30 lbs"). Ad networks suspend accounts over these and regulators fine over them. Qualify the claim or cut it. Note that this same list runs on --format ad, so a claim that fails here fails there too.

What this module deliberately does not do

The universal contracts always run. tells-detector and prose-quality fire on direct-response copy exactly as they fire on everything else, and this module never relaxes them. No copy type is licensed to sound more like an LLM because of where it ships.

That matters here more than in the other formats, because direct response legitimately uses devices that look like defects from a distance. Short punchy fragments are a rhythm tool in this tradition. So are repeated openers across consecutive lines. Sentence-length variance runs wide on purpose. The module's answer is not to loosen prose/* for this format, but to add the two checks those rules cannot make: whether the copy is concrete, and whether it closes.

So when a genuine rhythm device trips a universal rule, that is raised to the user, per the closing section below. It is never silently worked around, and it is never used as a reason to reword until one regex stops matching while the underlying prose stays the same.

The core principle

Good direct response sounds like a person who figured something out and wants to tell you about it. A marketing team sounds like something else, and so does a guru.

Write as though you are explaining something to a smart friend who is skeptical but curious. Back every claim with a specific. Make the transformation something the reader can picture. Everything below is mechanics in service of that.

Headlines

The headline does most of the work. Caples measured one headline outpulling another by 19.5x on the same product and the same offer. Ogilvy's number: on average five times as many people read the headline as read the body.

The master formula: action verb, plus specific outcome, plus a timeframe or a contrast.

• "Ship your startup in days, not weeks"
• "Save 4 hours per person every single week"
• "Build a $10K/month business in 90 days"

The contrast form ("days, not weeks") builds a before and after inside six words.

The story headline. Caples again, with the most famous ad headline ever written: "They Laughed When I Sat Down at the Piano... But When I Started to Play!" That is a complete story in fifteen words. Embarrassment, then triumph. The pattern: "They [doubted] when I [action]... But when I [result]..."

The specificity headline. Ogilvy's Rolls-Royce line: "At 60 miles an hour, the loudest noise in this new Rolls-Royce comes from the electric clock." It never says "quiet car". The reader reaches that conclusion alone, which is why it holds. The pattern: a specific number or metric, plus an unexpected comparison or detail.

The question headline. "Do You Make These Mistakes in English?" ran for forty years. The reader thinks "what mistakes?" and self-selects. The pattern: "Do you [common struggle]?" or "What if you could [desirable outcome]?"

The transformation headline. "From Broke Musician to $100K/Year Music Teacher." Before and after in one line, with the reader placing themselves in the before. The pattern: "From [bad state] to [good state]".

Why headlines fail: reaching for clever instead of clear; forgetting the reader's self-interest; vague claims where a specific belongs (this is what format/vague-benefit catches); and telling the whole story so that nothing is left to discover.

Caples' working advice was to write at least five headlines. One of them will beat the others by somewhere between 2x and 10x, and it will not be the one you expected.

Opening lines

The first sentence has exactly one job: earn the second sentence.

• Direct challenge. "You've been using Claude wrong." Stops the scroll, creates tension, self-selects readers who suspect it might be true.
• Story. "Last Tuesday, I opened my laptop and saw a number I couldn't believe: $47,329 in one day." The reader is inside the scene before noticing this is sales copy.
• Confession. "I'll be honest with you. I almost gave up on this business three times." Vulnerability disarms skepticism.
• Specific result. "In 9 months, we did $400k+ on a vibe-coded website using these exact methods." Numbers buy credibility, and the reader wants the how.
• Question. "Have you ever stared at a blank page, knowing you need to write something that sells... and just froze?" If the question matches their reality, they are already reading.
• The short sentence. Sugarman's approach: "It's simple." "Here's the truth." "This works." Near-zero friction to begin. They are in paragraph two before deciding to be.

Openings to avoid: the pace-of-the-modern-world scene-setter, the "are you ready to" question with a stock aspiration attached, the greeting that welcomes the reader and thanks them for arriving, the table of contents that announces what this article will teach, and the invitation to dive in. These are generic enough to head any document about anything, which is the problem. Several of them also trip tells/banned-phrase or format/vague-benefit outright, so they fail the gate as well as the reader.

Curiosity gaps and open loops

The brain wants closure. Open a loop and the reader keeps going to close it.

Creating the gap. "10 Tips for Better Writing" tells you exactly what you get, so there is no reason to read on. "I tested 47 headlines. One pattern beat everything else by 3x." raises a question the reader now has to answer.

Seeds of curiosity. End paragraphs with a pull into the next one: "But that's not even the best part." "Here's where it gets interesting." "Let me explain why." "Which brings me to the real secret." "Now here's the thing..." Two to four per page. Every paragraph ending on a tease reads as a tic.

The partial reveal. "The formula has three parts. The first one is obvious. The third one is counterintuitive. But the second one? That's where the magic happens." The reader now needs the second part specifically.

Closing loops. Every loop you open must close. Tease "the one thing that changed everything" and never deliver it, and you have spent trust you will not get back. Close small loops within one to three paragraphs; close big ones by the end of the piece.

Flow: the slippery slide

Sugarman's image: the reader should be so compelled that they cannot stop until they have read all of it, as if sliding down a slippery slide. Every element exists to deliver the next element.

Bucket brigades. Short connectives that smooth the seam between paragraphs: And. So. Now. But. Look. Here's why. Truth is. Turns out. The result? Think about it.

Without one: "Most landing pages focus on features. Benefits are what customers care about." With one: "Most landing pages focus on features. Here's the thing: benefits are what customers care about."

The stutter technique. Repeat a word from the previous sentence in the first sentence of the next paragraph.

"Now we're going to look at a more sophisticated technique.

A technique used by professional writers, but often overlooked by copywriters."

"Technique" bridges the gap, which is smoother than starting cold.

Short first sentences. The opening sentence of any section should be almost effortless to read. "It's simple." "Here's the problem." "This works." Momentum builds from a standing start, so make the start easy.

Vary paragraph length. Uniform paragraphs read as monotone. Short. Then a medium one that expands, adds the detail, gives the eye somewhere to rest. Then short again. This is also what keeps prose/low-burstiness quiet, since that rule measures the same thing from the other side.

Momentum killers: jargon the reader must stop to decode; long paragraphs with no break; tangents off the main thread; weak transitions that jar; and the same sentence structure used too many times running.

Pain quantification

Vague problems feel overwhelming. Quantified problems feel solvable, and a quantified problem is something the reader can weigh against a price.

Do the arithmetic in front of them:

4 hrs to set up emails + 6 hrs designing a landing page + 4 hrs to handle Stripe webhooks + 2 hrs for SEO tags + ∞ hrs overthinking...

= 22+ hours of headaches.

There's an easier way.

Seeing "22+ hours" is what lets a reader decide whether eliminating it is worth $99.

The other approach is the scenario that makes them feel it: "Imagine the scene: you and your team get an urgent email, so you rapidly reply. But just after you hit send, your team replies as well. In the best case, you look disorganized. In the worst case, you contradict each other." They have been there. Now the problem is felt rather than acknowledged.

The "So what?" chain

The default failure of machine-written copy is stopping at the first layer of benefit. "Saves time." "Increases productivity." "Helps you grow." All true, all inert.

For every feature, ask "so what?" until you hit something emotional or financial:

Feature: Fast database. So what? Functional: Queries load in milliseconds. So what? Financial: Users don't bounce, revenue doesn't leak. So what? Emotional: You stop waking up stressed about churn.

The bottom of the chain is where the copy lives. Not "saves 4 hours" but "close your laptop at 5pm instead of 9pm". Not "automates outreach" but "wake up to replies instead of a blank inbox". Three levels down, then write.

Rhythm and alternation

This is where generated copy most often gives itself away. It comes out either all choppy fragments or all flowing paragraphs. Real writing alternates.

Short sentence. Impact. Then a longer one that breathes, adds the context, and sounds like an actual person talking.

Two working examples of the poles. The punchy register: "Customers do NOT buy code. Customers buy a life transformation." Declarative, repeated structure, no ornament. The conversational register: "Once upon a time, you had a job. You traded hours for dollars, clocked in and out, and waited for the weekend. Your skills were confined to a cubicle and your ambitions to an annual review and a 4% raise." Longer, built on parallel structure.

Both work. The craft is knowing when to punch and when to breathe. The shape to repeat: hook (short, sharp), expand (breathe, add context), land it (a kicker that punctuates).

The founder story

Almost every high-converting creator page carries a first-person story. The shape is humble origins, struggle, discovery, success, offer, and the arc underneath it is vulnerability, then credibility, then shared journey.

When writing for a founder, get their actual story. This is not optional decoration. On most pages it is the single highest-trust element, and it cannot be invented on their behalf.

Testimonials

"Great product!" carries no persuasive weight at all. Structure a testimonial as a miniature case study: before state, action taken, specific outcome, timeframe, emotional reaction.

• "I shipped in 6 days as a noob coder. It would have taken me months. I wanna cry 🥲"
• "I managed to exit and sell for 5 figures in a few weeks. Best investment I've made in so long."
• "We were able to buy our first business within 4 months of joining."

The specifics do all of it. "4 months" is believable. "Helped me succeed" is not. This is also the material that answers format/missing-proof, and it only counts if the quote is real.

Disqualification

Telling some readers they are not a fit feels backwards and converts anyway:

You're a good fit for this if: ✅ You know this is a tool, and you'll need to use it ✅ You're willing to reassess your existing ideas

You're NOT a good fit if: ❌ You equate success with just buying a course ❌ You're not willing to do the unsexy work required

It flips the frame from "please buy" to "prove you're a fit", which is the velvet-rope effect. It also pre-filters the customers most likely to ask for a refund.

CTAs

Weak CTAs command an action. Strong CTAs describe what the reader gets by taking it.

Underneath the button, add friction reducers on the pattern of risk reversal, then social proof, then speed: "$199 once. Join 2,600+ marketers. 2 minutes to install."

Note that both columns satisfy format/missing-cta, since the rule only asks whether an action verb is present at all. Clearing the rule is the floor. The right column is the job.

Internet-native voice markers

Signals that a person wrote this rather than a marketing department:

• Revenue transparency, meaning specific numbers a corporate reviewer would soften
• Honest limitations, naming what the product does badly
• Strategic emoji, used sparingly and on purpose
• In-group language, the words the audience already uses with each other

The full sequence

Building a complete long-form page, in order: hook (outcome headline with a specific number or timeframe), problem (quantify the pain), agitate (a scenario or story that makes it vivid), credibility (founder story, authority, proof numbers), solution (what it does, framed as transformation), proof (testimonials with specific outcomes), objections (an FAQ or a fit/not-fit block), offer (pricing with value justification), urgency (only where it is real), and a final CTA with friction reducers beneath it.

Video scripts

When the deliverable is a VSL, TSL, or video ad that hands off to an editor, the format matters as much as the words. The editor is a creative collaborator with judgment of their own.

Deliver the voiceover as the primary artifact, with minimal direction. Editors read bracketed notes as mandatory rather than suggestive. A script carrying [CUT TO X], [B-ROLL Y], [TRANSITION Z] on every line reads as a contract, so instead of cutting, the editor books a call to settle the directions first. The script that looks more thorough is the one that delays the edit. Most editors will pick better B-roll, transitions, music and pacing than you would have described in brackets anyway.

The default is voiceover only. Plain paragraphs of what the narrator says, broken where they should breathe. No "Hook / Problem / Promise" headers inside the script itself; if those labels are needed at all, they go in a one-line note at the top.

Add a bracketed visual direction only when one of these is true:

• A number, claim, or quote must appear on-screen for the viewer to register it (price, statistic, guarantee terms, URL)
• A product shot, screen recording, or specific supplied asset has to land at a precise moment, and the audio alone does not make that obvious
• A comedic or dramatic beat depends on visual timing: a punchline reveal, a hard cut, a freeze frame
• A piece of B-roll is non-obvious enough that the script reads differently with it than without it

Over-directed, which triggers a clarification meeting:

[CUT TO: extreme close-up of clock ticking] [B-ROLL: stressed man at desk, papers flying] "Most people waste 4 hours every morning..." [TRANSITION: hard cut to bright kitchen] [B-ROLL: woman smiling, sipping coffee] "...but there's a faster way to start the day."

Right amount, where the editor moves immediately:

Most people waste 4 hours every morning... but there's a faster way to start the day.

[ON-SCREEN: "4 hours = 1,460 hours/year"]

The on-screen text earns its bracket because the number needs visual anchoring. Imagery, mood and cuts stay the editor's call.

Conventions: voiceover as plain text with no surrounding quotes, and paragraph breaks where the narrator breathes. Direction cues bracketed, all-caps tag, one line: [ON-SCREEN: "$47/month"], [B-ROLL: founder at whiteboard], [PRODUCT SHOT]. A cue that needs a paragraph of explanation belongs in a conversation with the editor before they start, somewhere other than the script.

If the user asks for a full shot list or a shooting script, that is a different deliverable from a VSL handoff. Confirm before producing one, because the lean voiceover-first script is usually what was wanted.

Classic frameworks

Eugene Schwartz: the five levels of awareness. The headline and the approach have to match where the reader already is.

1. Unaware. No knowledge that they have a problem. Lead with identity or emotion rather than the problem itself.
2. Problem-aware. They feel the problem, but do not know solutions exist. Name the problem vividly, then reveal that it is solvable.
3. Solution-aware. They know solutions exist but not yours. Show what makes your specific mechanism different.
4. Product-aware. They know your product and have not bought. Handle objections, add proof, create urgency.
5. Most aware. They want it and need a push. Lead with the deal.

The rule that falls out: the less aware they are, the longer the copy has to be.

Claude Hopkins, Scientific Advertising. Advertising is salesmanship in print, and the only purpose is to make sales. Reason-why copy does not merely claim, it explains why the product works. Specificity creates belief, which is why "cleans 99.6% of bacteria" beats "cleans almost all bacteria". Test everything with coupons, codes, and splits. Headlines do the heavy lifting.

David Ogilvy. On average five times as many people read the headline as read the body copy. Write the way you talk, naturally. Tell the truth, but make the truth fascinating. Give facts, because readers remember facts rather than adjectives.

Gary Halbert, The Boron Letters. Find a starving crowd first, because the offer matters more than the copy. Write to one person, using "you" and "I". Tell stories instead of pitches, since stories disarm skepticism. Read the copy aloud, and fix whatever comes out lumpy. AIDA still holds: attention, interest, desire, action.

John Caples, Tested Advertising Methods. One headline can outpull another by 19.5x on an identical product and offer. Self-interest beats cleverness. Specifics beat generalities. Test at least five headlines, because one will land somewhere between 2x and 10x better than the rest.

Joseph Sugarman, the slippery slide. Every element has one job, which is to get the reader into the next element. Seeds of curiosity end paragraphs by pulling forward. His list of psychological triggers runs to 31; the heaviest are honesty, proof, specificity, familiarity, and story. The buying environment (layout, price anchoring, logical flow, a clear action) is part of the copy.

Robert Collier, the six essentials of every letter. An opening that grabs attention. Description or explanation. An argument for buying. Persuasion to buy now. A risk-free offer. A clear call to action. And underneath all six, the line everything else in this skill is a technique for: enter the conversation already taking place in the customer's mind.

Before you run the gate

Read it aloud, then ask:

1. Does this sound like a person talking, or like someone "writing copy"?
2. Would I say this to a friend?
3. Is every claim backed by a specific number or a named proof?
4. Does the rhythm alternate between punch and breath?
5. Is it about the reader's transformation, or about my product?
6. Are there open loops pulling forward, and does each one close?
7. Does the ending carry momentum into the action?

Then run the script, because that list is judgment and the script is the gate.

When a violation looks wrong

Direct response owns devices that resemble defects. A deliberate fragment run trips sentence-level prose checks. A repeated opener used as anaphora across four lines looks like the structural monotony prose/low-burstiness exists to catch. A page-level rhythm built on triplets meets tells/rule-of-three-density.

When a device is genuinely load-bearing and a universal rule fires on it, that is a mismatch between the brief's register and this core's default general-audience contract. Raise it to the user per hedgehog-copywriting-loop's rule on extending the contract. Do not reword until the regex stops matching while the prose underneath is unchanged, and do not treat a format module as license to relax tells/* or prose/*, because it never does.

The error-severity format rules are a different case, and none of them have this escape hatch. Copy with no action verb is not direct response. Copy with no number is not making a claim anyone can check. A vague benefit phrase is an empty slot. An unqualified claim is legal exposure. Fix those at the root.