# Skill · Run ads to the offer that already sells without them

Install procedure the Grok Bot runs. Intake is already loaded. Drive the named tools. Stop only for login, 2FA, pay, or approve ads. Write the File log. Do not hop on a call. The File works if they only follow the blueprint. Agent is a later rung. Templates live in skills/paid-proof/.

## Fit gate

Fit: offer people already buy (referrals, organic, GHL receipts). Will put a VSL on the page. Will approve ads.

Unfit: no offer. Wants a startup idea or to clone someone else's product. Text-only checkout. Wants you to rip a SaaS one to one. Scaling spend at 10% net without looking. Furniture / dropship idea hunting. Stop.

## Tools

1. Proof card (skills/paid-proof/proof-card.md)
2. Research prompt — repeated work, not 10 ideas (skills/paid-proof/research-prompt.md)
3. GHL field map (skills/paid-proof/ghl-field-map.md)
4. Story script (skills/paid-proof/story-script.md)
5. VSL brief (skills/paid-proof/vsl-brief.md)
6. Ad swap + $50 test (skills/paid-proof/ad-swap.md)
7. Monday briefing (skills/paid-proof/monday-brief.md)
8. Copy-paste prompt (skills/paid-proof/prompt.md)
9. Meta Ad Library (structure, not a blank canvas)
10. GHL (always the CRM). Whop CLI only if the offer is digital checkout / membership.
11. File log

No other CRM. No funding File. You never spend. Human approves.

## Run

### 1. Proof card
- Last 90 days of THEIR sales. How they found you. Did ads ever run?
- If picking which of THEIR offers to push: proven, still skippable with paid. Spoken filters when people browse public revenue (inspiration only): $1K–$10K, then $10K–$20K, restated $0–$20K because those operators ground organic for years.
- Public verified-revenue directory = optional later look for category inspiration. Not a product to copy.
- If no receipts → stop or pick a different offer they already sell.

### 2. Repeated work, not ten ideas
- Worst first prompt: "Give me 10 profitable AI SaaS ideas." Kill that.
- Look for work they already do by hand: monitor sources, copy into docs, ask what changed, then decide.
- Return user / repeated pain / workaround / frequency / public evidence / alternatives / smallest useful product / realistic monthly price.
- Rank by pain × frequency × willingness to pay × simplicity of MVP.
- If they want a startup idea → stop.

### 3. Smallest product
- The useful answer is not another list of links.
- It answers: What changed? Why does it matter? What should I do next?
- First version does not need ten dashboards. It needs to beat their current workaround.

### 4. GHL is the sales machine
- Inbound comes in. Sales need pipelines and dispositions.
- Of 100 inbound: how many touched, how many conversations ≥30 seconds, how many closed, how many not.
- You need someone system / ops / tech oriented, or the File does this on paper first.
- Dirty data = do not scale. Clean GHL + ad metadata before any forecast.

### 5. Funnel math
- One opened-up account, labeled that, not a promise: ~$25–30K/mo. ~$387/day ads. ~32 opt-ins. ~25 qualified. 15% of qualified self-checkout. 85% abandon. Sales hits the 85%.
- Opt-in: industry ~15%. Their days 23%, 15%, 22%. Last 30 days ~13%.
- Levers: raise opt-in a little, raise self-checkout, then sales on abandon.
- Goal shape they named: $100K/mo at 30% net net. Current net sat 10–15%. Quiet part: $100K at 10% is a job. Prefer $50K/mo net $25K over $100K at 10%. Name net before you raise budget.

### 6. ROAS ceiling
- Whatever number you give the media buyer is the ceiling they hit.
- They said 1.5, then immediately changed to 2. Set the number you actually want.
- If the target is lower than you need → raise the ceiling.

### 7. Story script
- Beats: what happened → what really happened → pain after → common solutions → why they failed → pivotal moment → why it worked (root cause) → life after.
- Write from a real painful day, then from a client win.
- Cheat: pain-community success posts. Not travel groups. That post is a finished VSL.
- House example is a shop (truck down, season died, they almost closed). No medical copy.

### 8. VSL on every landing page
- Text-only self-checkout is wrong.
- 60 seconds is enough to start. Test 10+ minutes / even hour-long. Trust is hours of content.
- AI autoplay is allowed. Infomercials still work. That is the lesson, not a product to sell.

### 9. Saucy page
- Not a brochure. One offer. Video on the page. GHL form. Abandoned-cart SMS in GHL.

### 10. Ad swap + $50 test
- Steal a working ad structure. Swap your offer. Meta Ad Library is the house tool.
- Constraints, not "I'm hungry, cook me something."
- $50 test: 3 angles, 3 hooks, 3 copy, one Meta sales campaign. Draft. Human approves.
- Illustration: outcome CAC $21 beat technology CAC $63. Sell the outcome.
- RESEARCH → CREATE → DRAFT → APPROVE → LAUNCH. Then TEST → RESULT → LEARNING → NEXT TEST.

### 11. AI data lock
- Do not assume the bot hooked GHL.
- Ask: where did you get that number? Open GHL and verify the chain.
- Hallucination is real. Revenue data needs 100% accuracy.
- Do not copy-paste numbers between people and AIs.

### 12. Localhost is not a business
- Chain: product → price → checkout → customer → revenue → ads → data.
- Every write ends with a read. Verify product, price, billing, checkout before you share the URL.
- A command finishing is not proof the business state is correct.

### 13. First 10 are a learning problem
- Not scale. Do not automate yet.
- Find people publicly talking about the pain. Generate the deliverable before asking them to buy.
- Learn: do they read it, return, pay, keep paying?
- Then ask: what is the biggest constraint this week?

### 14. Sale → balance → ads → sale
- Small paid test from the live checkout. Spend needs approval.
- Pause a losing ad group only with their OK.
- Save the learning so the next test does not start from zero.

### 15. Operating roadmap
- $0 validate by hand → $500 fix retention → $1K start paid tests → $2.5K save learnings → $5K automate analysis → $10K optimize constraints.
- At $10K the question is: find the current constraint, prepare one experiment, verify. Not "how do we grow?"

### 16. Later: one desk
- Broken roadmap: find idea → landing → payments → ad manager → analytics → paste back into AI.
- Fix: Grok Bot on their computer. They only do login, 2FA, or pay.
- Whop CLI if the offer is digital: `curl -fsSL https://whop.com/install.sh | sh` then `whop login` then `whop skills add` then `whop --llms`.
- Everyday: products list, plans, checkout-configurations create (product_id → plan_id → checkout URL), stats time_series, ads list.
- LLC flow exists (`whop accounts llc`). They do not need an LLC to test an offer.
- Docs: https://docs.whop.com/developer/cli
- Client never types API keys into chat.
- GHL stays the CRM. Do not replace GHL with Whop.
- Monday briefing: skills/paid-proof/monday-brief.md

## File log

Columns: date · shop · offer · receipts (yes / no / Unknown) · GHL scoreboard (touched / 30s / closed) · VSL live · ROAS ceiling · net named · ad spend · note

## Done

Stop when the proof card is filled from their receipts, a VSL is on the page, GHL has the form and abandon SMS, the ROAS ceiling and net are named, and they can approve a draft ad without another meeting.

## Do not

- Do not clone someone else's product.
- Do not invent a GHL snapshot URL.
- Do not name a CRM other than GHL.
- Do not raise budget when net is under the number they named.
- Do not scale on a number with no source.
- Do not spend without approval.
- Do not use a claymation / oil-of-oregano example.
- Do not start from "give me 10 profitable AI SaaS ideas."