Turn a winning ad into a launch-ready landing page in an afternoon. Turn your Meta ads account into a creative strategy dashboard in an hour. Every tool, every link, every prompt, and the full playbook files, all on this page.
Six tools, all free to start. Fifteen minutes of setup and both workflows run for any brand, forever.
The engine for everything on this page. The desktop app is the easiest path for a first build. Install, then log in.
claude.com/claude-code →Lets the landing page preview locally on your machine. Check with node -v in your terminal; any modern version works.
One click downloads every image on a product page at full resolution. More reliable than any scraper, because sites lazy-load and block bots.
Find it on the Chrome Web Store →Connects Claude straight to your ad account. No developer account, no tokens, no keys. Add as a custom connector by URL: https://mcp.facebook.com/ads
Free hosting for the pages and dashboards you build. Install the CLI with npm i -g vercel, then vercel login.
fal.ai generates on-brand lifestyle images the product page can't give you. Apify powers the Reddit persona-mining expansion at the bottom of this page.
fal.ai → apify.com →The premise: a winning ad already told you what to build. The landing page's only job is to pay off that ad. Here is the whole flow, from an empty folder to a live URL.
Make a folder with client-assets/ inside. Drop in: the winning ad screenshots (ads/), every product-page image via Imageye (images/), any videos (videos/), the entire product page pasted as text (pdp.md), and the brand's blog/science/FAQ pages pasted (education.md). A person with a browser does this better than any scraper. Same 20 minutes for any client.
Save the self-driving playbook below into the folder and say: "Read CONSULTING_LP_PLAYBOOK-v2.md and run it." It scrapes the brand's real fonts and colors from the live site, builds a brand brief, and does deep research on how real customers talk.
It asks: conversion page, listicle, or quiz (or hand it a reference page to mimic). Pick based on what the ad sold. Then it proposes the structure and two hero options and waits for your call. The headline must pay off the ad's exact promise.
One self-contained page using the brand's real design tokens and real photos, held to a 15-rule build-quality contract. Then it reviews itself across five lenses (message match, above-the-fold, offer, mobile, trust) and fixes what would cost you sales, re-scoring until 90+.
Deploy to Vercel with noindex on, test add-to-cart and checkout, then point the winning ad at the URL. The ad that made the page becomes the traffic for the page.
| Page shape | Use when the ad sold... | The page's job |
|---|---|---|
| Conversion page | a specific product | Close the click now |
| Listicle ("5 signs...") | curiosity or a frame, not a product | Teach, then close |
| Quiz ("what's your type?") | self-diagnosis | Personalize, then close. Results live in the URL so ads can deep-link to them |
Copy this, save it as CONSULTING_LP_PLAYBOOK-v2.md in your working folder, open Claude Code there, and say "read it and run it." It prompts you for the harvest, scrapes the design tokens itself, asks which page shape you want, builds, QCs, and ships. Includes the full anatomy for all three page shapes, the copy formulas, the 15-rule build contract, the five-lens QC loop, and the compliance rules.
Paste these into Claude Code in order, from inside your working folder (after the harvest). Each one is a checkpoint where you steer.
Point Claude at everything you harvested and make it study before it builds.
The harvest is what the brand says about itself. This goes wider: real customer language, competitors, objections.
Everything gathered becomes one BRAND-BRIEF note, with the exact hex colors and fonts pulled from the live site.
Mimic a reference page (drop it in client-assets/reference/ first):
Conversion page (the ad sold a specific product):
Listicle (the ad sold curiosity):
Quiz (the ad sold self-diagnosis):
Two hero options, you pick or push back. The most important checkpoint on the page.
The long one, on purpose. These rules are what separate an 84 page from a 90 page.
Five lenses, evidence required, ranked P0/P1/P2. This prompt is the seed of a review agent: run it a few times and you have built one.
Apply the fixes and re-score:
Deploy:
Freeze the conversation into your own reusable files:
One paste. No API keys, no developer account, no files. Claude pulls your top ads, looks at the actual creatives, tags every ad by persona, angle, motivation, and format, indexes everything to YOUR target KPI, and builds a dark-mode dashboard that tells you what to iterate, what to reskin, and what to rethink.
In Claude: Settings → Connectors → Add custom connector → paste https://mcp.facebook.com/ads. Claude Code CLI alternative: claude mcp add --transport http meta-ads https://mcp.facebook.com/ads then /mcp to authenticate. Log in with the Facebook profile that has access to your ad account (the one you use in Ads Manager). Approve the read permissions.
New session, type "list my ad accounts." Your accounts should come back in seconds. If it fails: wrong Facebook profile (reconnect with the right one), or Meta hasn't rolled the connector out to that specific ad account yet, or a flaky call (retry, the prompt is built to survive it).
It asks exactly three questions: which account, your north-star KPI and target, and your monthly spend. Everything else is autonomous, with one taxonomy checkpoint where your corrections win.
Before building, Claude shows its personas, angles, and 5 sample tagged ads. Push back: rename a persona into your customers' language, merge angles. This is the step that makes it your dashboard, not a generic one.
One self-contained HTML file: KPI tiles, persona and angle leaderboards, a persona-by-angle matrix colored around your target, a sub-persona depth table with under-indexed winners pinned to the top, and three buckets (iterate, reskin, rethink) that read like Monday's to-do list. It ends by telling you, in three sentences, what to do next week.
Copy everything. This single prompt is the entire spec: pull rules, tagging taxonomy, spend-weighted math, the checkpoint, the dashboard build, and the ship step.
Weekly refresh: "Re-pull the last 30 days into the same dashboard and tell me which personas and families moved vs the 90-day view."
Close the loop to creative: "Turn each ad in the RESKIN bucket into 3 static ad briefs: hook, visual direction, and which sub-persona it targets."
Close the loop to landing pages: "Take the top under-indexed sub-persona and write the landing page brief for it: persona, angle, emotion, hero copy, and the 4 moments." Then run Workflow 1 on it.
Your dashboard doubles as your "already tested" database. This prompt reads it, scrapes Reddit for how real people talk about your category, and surfaces the personas your account has never touched. A new persona paired with an angle that already beats your target is the cheapest possible test.
Free account at apify.com. Then Claude Settings → Connectors → Add custom connector → paste https://mcp.apify.com. CLI: claude mcp add --transport http apify https://mcp.apify.com. No Apify? Claude's own web search over reddit.com works for a light version; Apify gets you comment-level volume.
It maps the subreddits, tells you the scrape cost before running, mines threads for real people with verbatim quotes and permalinks, cross-references semantically against everything you have tested, and adds a "Net-New Personas" section to the bottom of your dashboard.
It closes with the 3 threads you should personally read before making a single ad. The tool finds the room; if you do not learn how the people in it talk to each other, you will never sound like an insider, and outsiders do not convert anyone.
Paste after the dashboard exists, in the same folder.
Both workflows have a human checkpoint on purpose: you approve the hero before the page is built, and you correct the taxonomy before the dashboard is built. The tools produce the volume; the judgment stays yours. And every prompt on this page is just a good conversation, frozen. Run the workflows once and you will have your own playbook, not someone else's.
If the ad says "Can't stop farting?", the page cannot open with "Advanced 24-strain synbiotic." The ad's promise is the headline.
Exact hex and fonts from the live site, the brand's own photography and reviews, and numbers that agree. That is what separates "AI-made" from "the brand made this."
A hardcoded breakeven calls a healthy subscription account a failure. Everything indexes to your KPI target, so over 100 always means winning.
Claude looks at the rendered ads, not just the ad names. An ad named "HumorHeadline" turned out to be an ingredient-transparency static.
The newsletter gets you every file on this page plus the production versions as they evolve. If you want it done with you or for you:
Hands-on AI creative and landing page systems for your team: workshops, buildouts, and the workflows on this page installed inside your company.
sartoriusmedia.com →The agency behind these systems. Paid social creative, landing pages, and growth for DTC brands, powered by the production versions of everything here.
selfmade.co →The full playbook files, the dashboard prompt, and the upgrades (video tagging, the API version, new page shapes) as they ship.
Subscribe free →