landing-builder · git:20261104.217b23e · 2026-11-04 · sha256 a2d38f19899708a5

landing-builder git:20261104.217b23eA

Immutable. This exact content never changes and is served at /api/v1/blob/a2d38f19899708a5.

No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_emkzcy4fe66q7sxa/a2d38f19899708a582f3aea7ff5eb4a7e3594158b6c1fa2e2dfddaaeac70fcfa.json

Then /plugin install landing-builder@markdownregistry-emkzcy4fe66q7sxa in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.

---
name: landing-builder
description: >-
  Write the structure and copy for a selling landing page: hero, proof,
  features as outcomes, pricing, FAQ and a final call to action, with
  variants per audience and an A/B test plan. Trigger on "write a landing
  page", "landing for", "homepage copy", "page for teams", "page for
  bookmakers", "B2B landing", "rewrite the hero", "A/B test the landing", or
  when a product or feature needs a page whose job is to sell.
---

# Landing builder

A landing page is an argument with one conclusion: the primary action. Each
section answers the next objection a visitor would raise. Write it in that
order, not in the order the product was built.

## Step 1: brief (ask, max four questions)

1. **Audience**: who lands here and from where (search, a social post, an ad,
   a partner link). One page per audience that differs in need, not per
   channel.
2. **Primary action**: one (start free, start trial, book a demo, install).
3. **Proof available**: what can be shown and checked today.
4. **Constraints**: brand voice, legal claims that are off-limits (for a
   betting-adjacent product: no promises of winnings, age and jurisdiction
   notices where required).

## Step 2: the section order

| # | Section | Job | Rule |
|---|---------|-----|------|
| 1 | Hero | What, for whom, what next | Outcome headline under 10 words, subline naming the audience, one primary button, one proof point, a real product visual |
| 2 | Proof strip | Believe it | The strongest checkable proof (public track record, numbers with dates, logos with permission) |
| 3 | How it works | Understand it | Three steps, each an action the user takes |
| 4 | Outcomes | Want it | Features written as results: "Know the grid before qualifying" not "Qualifying model v3" |
| 5 | Deep proof | Trust it | A real example: last race's prediction next to the result, misses included |
| 6 | Pricing | Afford it | Free and paid side by side; follow `paywall-and-pricing` |
| 7 | FAQ | Remove the last objection | Five to eight real questions from support, sales or reviews |
| 8 | Final call to action | Act | Repeat the primary action with the objection it answers ("Free, no card") |

Drop a section only with a reason. Add none that does not answer an
objection.

## Step 3: variants per audience

Same product, different job to be done. Change the headline, proof, outcomes
and primary action; keep the design system and pricing logic.

ExampleApp example (full copy in `references/section-templates.md`):

| | Fans | Teams and analysts | Bookmakers and media |
|---|------|--------------------|----------------------|
| Job | Know what will happen, enjoy the weekend more | Faster, checkable pace and strategy reads | Reliable probabilities to publish or price against |
| Hero | "Know the result before the lights go out" | "Race predictions your engineers can audit" | "Calibrated F1 probabilities, delivered by API" |
| Proof | Public track record, app rating | Methodology page, per-session error, backtests | Calibration chart, uptime, latency, data licence |
| Primary action | Get the free prediction | Book a 20-minute walkthrough | Request API access |
| Pricing | Free and Pro | Team plan | Contact, volume based |

Rules for variants:

- Separate URLs per audience with their own title and meta description (so
  search and ads can land each audience on its page), shared components.
- Never claim to a B2B audience what the product cannot deliver today;
  "coming soon" goes into a waitlist action, not into the outcomes list.
- Regulated audiences get the compliance lines they expect (responsible
  gambling link, licensing questions answered in the FAQ).

## Step 4: copy rules

- Headlines state the outcome; sublines say who it is for and how.
- Numbers beat adjectives; every number has a date and a source link.
- Jargon only on the audience page that uses it (MAE is fine for analysts,
  not for fans); run all copy through the `ux-copy` skill, which also hands
  it to the `humanizer`.
- No em dashes, no emojis, no exclamation marks in headlines.
- Buttons name the result of the click.

## Step 5: A/B test plan

One hypothesis per test, written before building it:

```
Because <evidence>, we believe <change> for <audience>
will increase <primary metric>, without hurting <guardrail>.
```

- **Primary metric**: the page's goal event per unique visitor (for example
  `signup_success / landing_view`), not clicks.
- **Guardrails**: bounce rate, downstream activation (the free prediction
  actually viewed), refund or cancel rate for paid actions.
- **Sample size**: for a baseline rate p and the smallest lift worth
  detecting d (absolute), each arm needs about `16 * p * (1 - p) / d^2`
  visitors for 80% power at 5% significance. At p = 4% and d = 1 point that
  is about 6,100 per arm. If traffic cannot reach it in four weeks, test a
  bigger change or skip the test and ship the better-argued version.
- **Duration**: whole cycles. For a weekly-event product, at least two full
  event cycles, because race-weekend visitors behave unlike weekday ones.
- **No peeking**: decide the stopping point in advance; do not stop on the
  first significant day.
- **Order of tests**: hero headline, then proof block position, then primary
  action wording, then pricing presentation. Big levers first.

Events: `landing_view{variant, audience, source}`,
`landing_cta_tap{variant, position}`, `landing_section_view{section}`, and
the goal event with `variant` as a property. Assignment is sticky per
visitor and logged once as `experiment_exposure{experiment, variant}`.

## Output

1. Brief (the four answers).
2. The page per audience: every section with final copy, the visual for the
   hero, and notes for the designer.
3. Meta title and description per variant.
4. The A/B plan: hypotheses in priority order, metrics, sample size,
   duration.
5. **Open questions**: proof that does not exist yet, legal lines to confirm.

For the visual design of the page, hand the copy to the `redesign` flow in
the `ux-craft` plugin; the page is designed and approved before it is built.