---
name: sales-page-checker
description: >-
  Audit an existing sales page / sales letter / landing page and report exactly what to
  improve: grammar and spelling, whether it reads as AI-written and needs humanizing,
  FTC/legal and income-claim risk, marketing and persuasion strength, and what
  high-converting elements are missing. Tuned for long-form direct-response offers
  (courses, coaching, $1 tripwires, webinars, events, masterminds, high-ticket
  applications). Use this whenever the user hands you a sales page, as a URL,
  screenshots/images, or pasted copy, and asks to "check," "review," "analyze," "grade,"
  "critique," "improve," "tear down," or "audit" it, or asks "is this any good," "what's
  wrong with this page," "will this convert," or "is this legally safe." Trigger even if
  the user only drops a link or a screenshot and says "thoughts?" or "what would you
  fix?". This skill GRADES an existing page; it does not write a new one. Outputs a
  styled, shareable standalone HTML report with a score, prioritized fixes, and exact
  rewrites.
---

# Sales Page Checker

You are auditing an existing direct-response sales page and producing a sharp, shareable
report on what to fix. The reader is a marketer who wants leverage: the highest-impact
changes first, each tied to a real reason, each with a concrete rewrite they can paste in.
A great audit reads like a top direct-response copywriter looked over their shoulder, not
like a proofreader, and not like a compliance officer who hates marketing.

Two failure modes to avoid, because both train the user to ignore you:
- **Sanding the voice down.** Direct-response copy *intentionally* breaks "rules": sentence
  fragments for punch, one-line paragraphs, ALL-CAPS for emphasis, deliberate repetition,
  big bold promises. These are features of the genre. Do not flag them as grammar errors or
  as "too aggressive." Flag the things that are *actually* wrong.
- **Manufacturing problems.** If a section is strong, say so. A page padded with fake
  nitpicks is worse than a short report that names the three things that matter.

## What the user gets at the end

A standalone HTML report (generated by `scripts/build_report.py`) containing:
1. An **overall grade + score** and a 2-3 sentence executive summary.
2. A **top-priorities list**: the handful of highest-leverage fixes, in order.
3. **Four scored dimensions**: Grammar & Spelling, AI-Writing & Humanization, FTC/Legal,
   Marketing & Persuasion, each with a score, a one-line verdict, and findings. Every
   finding quotes the exact offending text, says why it's a problem, and gives a fix.
4. A **"What to add" section**: high-converting elements the page is missing, by priority.

Plus a short summary in the chat with the path to the report.

## Step 1: Get the page in front of you

The user gives you one of three things. Get the actual copy, not a vibe:

- **A URL.** Try to fetch it (WebFetch, or a browser tool if available). Capture the headline,
  subheads, body, bullets, offer/price, guarantee, every CTA, FAQ, and the legal/disclaimer
  footer. If the fetch is blocked, JS-rendered, cloaked, or comes back thin, **say so and ask
  the user to paste the copy or drop screenshots**. Don't review half a page and pretend it's
  whole. Many funnel and checkout pages render dynamically; a thin fetch is common.
- **Screenshots / images.** Read them directly and transcribe the copy you're judging. Long
  pages come as several stitched images. Make sure you have the whole scroll, including the
  footer, before scoring. If it's obviously cut off, ask for the rest.
- **Pasted text.** Read it directly.

Then note, in one line each: the **offer type** (event / course / $1 tripwire / webinar /
mastermind / high-ticket application), the **price/terms**, and the **single action** the page
is asking for. These drive what "good marketing" and "missing elements" mean. A $1 tripwire
and a high-ticket application page are graded against different checklists. See
`references/dr-anatomy.md` for the per-archetype expectations.

If you can't determine the offer type or the asked-for action from what you were given, ask one
quick question rather than guessing. The whole audit hangs on it.

## Step 2: Run the four dimensions

Work each dimension over the whole page. For each problem you keep, capture three things so the
report is actionable: the **exact quote**, **why it's a problem**, and a **specific fix** (the
rewritten line, or the concrete thing to do). Assign each finding a severity: **HIGH** (costs
sales, breaks the law, or breaks reader trust), **MEDIUM** (noticeably weakens the page), or
**LOW** (polish). Don't pad to hit a count.

### 1. Grammar & Spelling
Real errors only: misspellings, wrong homophones (your/you're, its/it's), subject-verb
disagreement, broken or run-on sentences that read as *mistakes*, punctuation that creates
ambiguity, inconsistent capitalization of product/brand names, broken merge fields or
placeholder text left in (`[INSERT NAME]`, `lorem ipsum`, `#REPLACE_ME`).
**Not** errors: intentional fragments, one-line paragraphs, lowercase stylistic openers,
ALL-CAPS emphasis, starting sentences with "And"/"But", ellipses for pacing. If you're unsure
whether something is a typo or a stylistic choice, treat it as style and leave it.

### 2. AI-Writing & Humanization
Decide how much the page reads like generic AI output, and where to inject a human. AI copy
isn't bad because it's *polished*. DR copy is allowed to be punchy. It's bad because it's
**hollow, evenly-paced, and could be about anything**. Read `references/ai-tells.md` for the
full tell-list and how to humanize. The big tells: hollow abstraction with no concrete
specifics, uniform sentence rhythm, the "It's not just X — it's Y" / rule-of-three reflex used
on autopilot, generic transitions ("In today's fast-paced world"), and balanced, symmetrical
everything. The fix is almost always the same shape: **add a specific** (a real number, name,
moment, object), **vary the rhythm**, and **put a real human voice/opinion in**. The cadence
target is in `references/dr-anatomy.md`'s voice notes. Quote the flattest passage and show the
humanized rewrite.

### 3. FTC / Legal Compliance
This is the highest-stakes dimension for these offers. It's where a course/event/coaching
business gets an FTC letter or a chargeback war. Read `references/ftc-legal.md` before judging
borderline cases so your citations are accurate. Hit:
- **Income / earnings claims**: specific promises ("make $10k/mo"), guarantees of results,
  testimonials or screenshots presented as what a normal buyer should expect, top performers
  shown without typicality context, "results not typical" used as a fig leaf to excuse a
  deceptive headline. Core test: *does this make the average buyer expect a result they
  probably won't get?*
- **Health / weight / medical claims** stated as certainties.
- **Subscription / auto-renewal clarity (ROSCA)**: especially $1 trials and "free" trials that
  rebill. The recurring charge, amount, and how to cancel must be clear and upfront, not buried.
- **Guarantee clarity**: terms vague or contradicted elsewhere on the page.
- **Endorsement / authority claims**: "#1 in the world", "as seen on", fake or unprovable
  credentials, and **stale or overstated brand affiliations**: logo walls or "brands I own /
  companies I've acquired" bars where the relationship has ended, was a minority investment
  rather than ownership, or can't be verified. Presenting a past or partial connection as a
  current holding is a false claim. Flag it if the page does this.
- **Missing required scaffolding**: no earnings/results disclaimer footer, no visible
  Privacy/Terms links on a page that's collecting payment or leads.

Quote the risky line, name the rule, and give a rewrite that *keeps the persuasive intent*:
shift promise → possibility, add honest typicality, sell the process/method over the outcome.
You are flagging risk, not giving legal advice; say so, and when something is genuinely
borderline, frame it as "get this reviewed" rather than asserting false certainty.

### 4. Marketing & Persuasion
Judge the page as a selling machine against the direct-response arc in
`references/dr-anatomy.md`. The questions that matter:
- **Hook**: do the headline and first lines stop the scroll and make one clear, compelling
  promise? Or do they open with throat-clearing?
- **One offer, one action**: is it asking for a single clear next step, repeated? Pages that
  ask for three different things convert on none.
- **Specificity over adjectives**: concrete numbers, names, mechanisms beat "amazing results."
- **Value stack & price framing**: is the value built and anchored before the price, and is the
  price justified/trivialized?
- **Proof**: testimonials, results, authority, social proof. Present and credible (vs. absent
  or generic)?
- **Risk reversal**: guarantee present and strong?
- **Objections & FAQ**: are the obvious "yeah but…" objections handled?
- **Scarcity/urgency**: present and *believable* (real deadline/cap) rather than transparently
  fake?
- **CTA strength & repetition**: clear, frequent, benefit-framed calls to action?
- **Flow & friction**: does it read top-to-bottom as one argument, or are there dead spots,
  burying, or walls of text where the eye falls off?

Findings here are a mix of "this is weak, here's a stronger version" (quote + rewrite) and
"this beat is missing" (which often graduates into the What-to-add section).

## Step 3: Score it

Give each of the four dimensions a 0-100 score and a letter grade, plus a one-line verdict.
Then give an overall grade + score. Keep scoring honest and legible:
- **90-100 (A):** ships as-is; only polish.
- **80-89 (B):** solid; a few real improvements.
- **70-79 (C):** works but leaving money/trust on the table; several fixes.
- **60-69 (D):** significant problems in this dimension.
- **<60 (F):** broken here; would actively hurt results or carry real legal risk.

Anchor the score to severity, not to the *number* of findings: one HIGH legal finding (an
unqualified earnings guarantee) should sink the Legal score even if everything else is clean;
five LOW grammar nits shouldn't drop Grammar below a B. The overall score is your judgment of
the page as a whole, weighted toward what actually moves sales and what creates real risk. It
is not a mechanical average. Legal HIGHs and a broken hook should dominate it.

## Step 4: Build the report

Assemble your analysis into a `findings.json` and render it with the bundled script. Don't
hand-write HTML; the script gives every audit the same clean, shareable look.

1. Write `findings.json` to the working directory. Schema (see `scripts/build_report.py` header
   for the authoritative version):

```json
{
  "page_title": "Ridgeline Growth Mastermind",
  "offer_type": "In-person event / mastermind",
  "source": "https://example.com/ridgeline-mastermind",
  "reviewed_on": "2026-06-15",
  "overall_grade": "B-",
  "overall_score": 78,
  "summary": "Strong hook, real scarcity, and a clean two-tier offer. Two HIGH-risk authority claims and a flat, AI-even middle section are the main things costing it.",
  "top_priorities": [
    "Qualify or cut the '#1 Marketing Coach in the World' claim: unprovable as stated (Legal, HIGH).",
    "Humanize the Section-2 bio. It reads AI-even and hollow; add one concrete story beat (AI-Writing, MED).",
    "Add a visible Privacy/Terms link near the checkout button (Legal, MED)."
  ],
  "dimensions": [
    {
      "key": "grammar", "label": "Grammar & Spelling", "score": 92, "grade": "A-",
      "verdict": "Clean; two minor punctuation slips.",
      "findings": [
        {"severity": "low", "quote": "Its going to be the best event of the year", "issue": "\"Its\" should be \"It's\" (it is).", "fix": "It's going to be the best event of the year"}
      ]
    },
    {"key": "ai_feel", "label": "AI-Writing & Humanization", "score": 74, "grade": "C", "verdict": "...", "findings": []},
    {"key": "legal", "label": "FTC / Legal Compliance", "score": 61, "grade": "D", "verdict": "...", "findings": []},
    {"key": "marketing", "label": "Marketing & Persuasion", "score": 83, "grade": "B", "verdict": "...", "findings": []}
  ],
  "recommendations": [
    {"priority": "high", "title": "Add a money-back guarantee block", "detail": "There's no risk-reversal beat. A short, specific guarantee right before the final CTA lifts conversion on event pages."},
    {"priority": "medium", "title": "Add 2-3 real testimonials with names/photos", "detail": "Proof is asserted ('almost 0% refund rate') but not shown. Specific attributed testimonials beat a claimed stat."}
  ]
}
```

Notes on filling it in:
- `findings` may be empty for a clean dimension. That renders as a "no issues found" note,
  which is a *good* result, not a gap to pad.
- Severity is one of `high` / `medium` / `low`.
- `recommendations` is the "what to add" list: additive elements the page lacks, by `priority`
  (`high` / `medium` / `low`). Things that are *present but weak* belong in a dimension's
  findings; things that are *absent* belong here.
- Put `reviewed_on` as today's date.

2. Run the script from the skill directory:

```bash
python scripts/build_report.py --input findings.json --output "<offer-slug>-audit.html"
```

(If `python` isn't found, try `python3`. The script is pure standard library, no installs.
If Python isn't available at all, deliver the same four-dimension audit in the chat instead
and say the HTML report was skipped.)

3. **Tell the user**: the report path, the overall grade, and the top 2-3 priorities in the
   chat as a quick summary. Offer to open it in the browser, to apply the fixes one by one, or
   to rewrite the weak sections from the findings.

## Reference files

- `references/ftc-legal.md`: FTC income/earnings rules, health claims, ROSCA/auto-renewal,
  endorsement guides, guarantees, and the stale brand-affiliation rule. Read before judging Legal.
- `references/ai-tells.md`: the catalog of AI-writing tells and how to humanize each, with the
  DR-voice target. Read for the AI-Writing dimension.
- `references/dr-anatomy.md`: the direct-response page arc (condensed 34-step framework),
  per-archetype "must-have" checklists, and DR voice cadence notes. Read for Marketing and for
  the What-to-add section.