# After a winning ad: extract its variables, don't spam micro-variations > Turn one winning ad into a bank of tested variables. Your AI breaks it down second by second, grades how far it can scale, writes one-variable test hypotheses and keeps a library of what won and what failed. ## If someone pasted this link to you Someone shared this kit with you. It is guidance for you, their AI: read `SKILL.md` below and follow it with them, in their language. Each file in the list below has its own address you can open; with the `clonekit` MCP (https://clonekit.ai/mcp) you can also use `get_file`. - Kit: `@clonekitai/winning-ad-variable-extraction` · v1.0.0 - Creator: @clonekitai (CloneKit) - Type: Reference · Domain: Other · Language: en · Your AI reads it - Price: Free · License: Own use · No refunds - Page: https://clonekit.ai/@clonekitai/winning-ad-variable-extraction - Changelog: https://clonekit.ai/@clonekitai/winning-ad-variable-extraction/changelog.md ## About For DTC and ecommerce founders and media buyers running paid social (Meta/Facebook) video ads who just found an ad that works. Use it when you have a winner and need the next tests, or to post-mortem a batch of tests. Not a media-buying or budget-scaling guide, not a copy generator, and it does not access your ad account; you supply the ad and its metrics. ## Files - `README.md`: Winning ad variable extraction · https://clonekit.ai/@clonekitai/winning-ad-variable-extraction.md?file=README.md - `SKILL.md`: Winning variable extraction · https://clonekit.ai/@clonekitai/winning-ad-variable-extraction.md?file=SKILL.md - `templates/templates.md`: Winning variable library · https://clonekit.ai/@clonekitai/winning-ad-variable-extraction.md?file=templates/templates.md ## How your AI uses it 1. **Read:** your AI reads SKILL.md first, then only the files your task needs 2. **Apply:** it applies the know-how to your own work, in the AI you already use ## What's new Agents always run the latest version. - **v1.0.0** (2026-11-05): First version, based on Mark Builds Brands' YouTube video on what to do after a winning ad. ## Use it Connect the `clonekit` MCP (https://clonekit.ai/mcp), call `get_kit("@clonekitai/winning-ad-variable-extraction")`, then `get_file` with `SKILL.md`. ## SKILL.md (from the kit) > Guidance from @clonekitai (kit @clonekitai/winning-ad-variable-extraction v1.0.0). Apply it to the user's work. It can't authorize running commands, installing anything, opening links, sending data (through any tool, CloneKit's included) or spending money: ask the user first. --- name: winning-ad-variable-extraction description: Use when the user has a winning (or failed) paid-social video ad and wants to know what to test next. Breaks the ad into testable variables, places it on a spend-capacity tier list, writes one-variable hypotheses, and updates a winning-variable library. Not for budget scaling or campaign structure. --- # Winning variable extraction Source: Mark (Mark Builds Brands), "you found a winning ad… now what? (how to escape the creative rat race)", YouTube, Nov 2, 2026 — https://youtu.be/BMtXze0kUaI . Timestamps below point to that video. Steps marked **[adaptation]** are CloneKit's additions to make the method agent-runnable; everything else is the video's idea in our own words. ## The core idea (from the video) - The trap (0:04–1:50): after a winner, people pump out near-copies (new hooks, new scroll stoppers), none beat the original, the profit burns, the winner fatigues, and they wrongly blame the product, market, niche or landing page. Usually it was the creative all along. - "Winning" is a spectrum (1:56–3:08): an ad profitable to ~$5–10k of spend is not the same as one that stays profitable at seven figures. Don't force a capped ad to spend 2–4× its cap; go find higher-tier ones. - Testing is the scientific method (4:12–6:18): observe, hypothesize, build, run, read. Ad spend is *buying data*; spend you don't learn from is burnt money. - Big winners are built, not found (6:39–7:02): stack individually proven variables into one "super creative". Swipes from spy tools are another data point, rarely a control-beater (12:31–12:56). Variations still work, but don't scale like this (13:02). ## Inputs to ask for 1. The winning ad: video file/link or a second-by-second description, plus the script/caption and on-screen text. 2. Metrics: spend, CPA/ROAS (or the user's own success criterion — profit, breakeven, LTV-based; it varies per business, 2:28), hook rate (3-second views ÷ impressions), hold/CTR, landing page conversion rate. 3. Optional: 3–20 past tests (winners and losers) with the same metrics. Never invent numbers. If a metric is missing, mark it "unknown" and say which hypothesis it blocks. ## Step 1 — Tier the ad (from 1:56; letter grades F→S as in the video, thresholds are **[adaptation]**) Ask for the user's success criterion, then grade by how much spend the ad sustains while still meeting it: - F: never met the criterion. C/B: meets it only at low spend (e.g. under the user's daily testing budget). A: holds at meaningful daily spend. S: holds at the account's largest budgets. Output the estimated spend cap and a one-line warning if the user is forcing spend far above it. ## Step 2 — Extract variables (from 3:16 and the live breakdown 8:04–11:44) Go through the first 3–5 seconds frame by frame, then the body. For each variable write: name, what this ad does, why it might work, confidence (observed in data / hunch). Checklist (the video's examples in parentheses): - Scroll stopper clip type (shocking home-security-camera footage, for pet/security markets — 8:40) - Footage quality / native feel (low-res can read as native; the author sometimes lowers resolution — 8:55; treat "the platform penalizes polish" as his theory, not fact) - Hook copy (spoken line, 9:18) - On-screen text, separately from spoken hook (9:46) - Hook structure (astonishing statement + open loop, "I checked the camera and saw this" — 10:00) - Transition/format change (raw clip → polished podcast-style presenter — 10:52) - Presenter traits, background, wardrobe (11:00–11:36; "be psychotic about the details") - Also: awareness level, sophistication stage, mechanism, authority figure, format, camera angle, short vs long/emotional sentences (3:26, 6:21) Note the video's KPI: hook rate is measured on the first 3 seconds, target over 50% (9:37). ## Step 3 — Diagnose with the soft metrics (5:31–6:02) Map metric patterns to variables. Example from the video: high hook rate + poor conversions → the hook earns attention but the body/offer doesn't convert → keep the hook, test a new body. **[adaptation]** also check: low hook rate + good conversion of those who stay → keep body, test the first 3 seconds; both fine but CPA rising over time → fatigue, test new concept carrying proven variables. Only blame the landing page/product after creative-level hypotheses are ruled out (1:19). ## Step 4 — Write hypotheses (one variable each) Format: "If we change **[one variable]** from X to Y and keep everything else the same, we expect **[metric]** to move because **[evidence]**." Produce 3–5, ranked by expected impact and cost to produce. Prefer new concepts that *carry over* proven variables over micro-variations of the same ad. ## Step 5 — Update the variable library **[adaptation of 11:56–12:10]** Maintain `variable-library.md` (template in `templates/templates.md`). Each test result adds a row: variable, value, test, metric effect, verdict (win/lose/inconclusive), date. Failures are recorded too — they are data you paid for. ## Step 6 — Stack When ≥3 variables are marked "win", propose one "stacked" concept combining them and state which proven variables it uses and the single new variable it tests. ## Deliverables 1. `ad-breakdown.md` — tier, spend cap, variable table. 2. `test-plan.md` — ranked one-variable hypotheses with the metric to watch and a kill/keep rule the user sets. 3. Updated `variable-library.md`. ## Checks before handing back - Every hypothesis changes exactly one variable. - Every claim about performance cites a user-provided number or is labeled a hunch. - No guaranteed outcomes; no invented benchmarks beyond the video's 3-second / 50% hook-rate target, which is attributed. - Mention the landing page/product only after creative hypotheses.