Fifteen Playbooks

Creative is the targeting now. Meta's Andromeda retrieval step shortlists a few thousand ads per user from tens of millions by matching creative "vectors" to user vectors before any auction. If your ad is not shortlisted, bid and budget do not matter. So test genuinely different concepts (different reason to buy, pain point, persona, awareness stage), not colour, crop or hook-line variants. Dd4fzTjR9rk DfXrim6xnyH DdoxpuXBQNF DeFATG6umBR DdpfvdHCDp1 Dd83hWNjzaK Expect most creatives to lose. A study of 578,000+ creatives and $1.29B spend (read as Motion's) found only about 3.7% to 8.8% become "winners" (about 5% on average, so roughly 11 to 27 concepts per winner), and the winner rate is lowest under $10K/month spend. Test cheap, simple formats first, add production value only after a concept wins. DdoxpuXBQNF Clean the default settings before every launch: Audience Network, Threads and Messenger off; untick "Allow limited spending to excluded placements"; Related media off; creative Enhancements off (except maybe "reveal details over time" on video); multi-advertiser ads off for video; consider dynamic creative off. Re-check after publishing and after duplicating, because Meta re-ticks them. Then judge placements on cost per real result, not CPM. Dba8Pqvu-Y4 Db3L0XITUbY DecMToApLvz [demo, no result data]

Concept diversity beats variant volume. Meta clusters look-alike ads, so change the reason to buy (angle, desire, awareness stage, persona, job to be done), not the colour or one hook line. Expect about 1 winner in 11 to 27 concepts, fewer at small budgets. DdpfvdHCDp1 DfXrim6xnyH DdoxpuXBQNF Dd4fzTjR9rk Dd83hWNjzaK Build ads as modules: hook + meat + CTA. Write many hooks (Hormozi: 50 from 5 sources), 3 to 5 meats (demonstration, testimonial, educational, story, faceless), 1 to 3 CTAs, film in one session, recombine. DeFATG6umBR A full CTA tells five things: what to do, why it matters, when, how easy, and what happens next. DeFATG6umBR

Simplify the page around one action. Seven named before/after tests (Yuppiechef, Imaginary Landscape, Career Point, Humana, Ring, Highrise) all won by removing nav, cutting form fields, cutting copy, one image, one obvious button. Lifts of 2x to 15x. This is the best-evidenced lesson in the set. DeLHuWuPiwj In India, prepaid share is a lever you can move with the page plus money: a clean, visually uniform PDP plus a flat 10% prepaid discount gave 44% prepaid on day one (n=9 orders). A 15% prepaid discount still left about 80% COD for a low-ticket dropshipper. Discount alone is not enough; trust on the page matters. DdoNputBhG1 Dc1rL8hSzS7 Measure what actually lands: check RTO before you celebrate ROAS, know breakeven ROAS first, track delivered revenue. RTO claims range from 10% (Indian operators in comments) to 60% non-delivery (one low-ticket student). Dc1D7FWyPGT Dc1rL8hSzS7

Reply fast and log everything. One inbox for web, WhatsApp, Google, Instagram, phone and chatbot; auto-reply in 60 seconds; qualify budget and location; offer real slots; track revenue per source; re-contact old leads. DdpPxHkIsuN Dcjrk_HxEPR High-ticket recovery works with a human: call every abandoned cart right away and close on a Shopify draft order. Claimed 60% of one store's revenue. DcQ9S3EAaA_ Comment-keyword to DM is now everywhere and is also native in Meta ads ("Reply to keywords" in engagement-campaign ad settings, partial rollout). It turns reels, stories and ads into contact capture. DcbPON-zciJ DdtNdBOFGHJ Dd7gztHRzP9

Trial Reels are the main growth lever in these reels. Post a test to non-followers first, then put winners back out in a changed form: a new hook, a new edit, a new angle, a new format, or a full remake. Five creators describe versions of this loop. The results come from their own screenshots. Nobody has controlled data, and commenters report duplicate flags and reach drops. Dco8DB2oRRt DeSJg4Mkg42 DdwPZAku3Jx DeFFzgiCQOe DeGVbSJtBfn De-HXzJtpku Research before you post. List the top 10 videos in your niche. Track competitors for outlier topics. Mine winning hooks from your own posts, other creators and the longest-running ads. Then remake the idea in your own voice. Dd4QXwYust3 DeUqvO8zKcP DdwPZAku3Jx DeFATG6umBR Ddt4Tq0BGS_ Put the value in the video, not the caption. The "3-second reel plus long caption" trick from 2024 now annoys most viewers. Build for average view duration (put the payoff at the end) and for shares (to a friend, or to the viewer's own profile). Dd4oJ7gjqGX Dd4QXwYust3

A hook is a promise. Brock's definition: a promise, a question you will answer, a topic you will cover, or a problem you will solve. Visual gimmicks get the first glance, but the promise keeps people watching. Best practice in these reels is both: a visual pattern interrupt plus a real promise within 3 seconds. Dd8kCcCCGG3 DeSKm5HB3MF Open a loop and explain nothing at the start. Barnum's brick man, the "egress" sign, an unexplained 25 lb mug, a "quest" with no known ending. People move to find out. Close the loop at the end. DeXr6rKlB4G DeSKm5HB3MF Dd7_w1wsdKf Use a script skeleton with a deliberate middle. Hook (0-3s), Explain (3-10s, do not pop the climax), Illustrate (10-30s: example, story, maths, quote, diagram, analogy), Apply (takeaways). Most people skip Illustrate. DdjZgv9BFvf

AI video editing in Claude Code is now a real, repeatable stack: Claude Code as the editor agent, Remotion for motion graphics, FFmpeg for cuts and colour, Whisper (or AssemblyAI / ElevenLabs) for word-timed captions, yt-dlp to pull reference videos, a person segmentation model for text-behind-head, and HyperFrames for motion and shot-level review. Seven reels show the same stack. De8SPjEAuV0 De2SEI7JP8r DevtyU1kmHC DeUUFJQu8cw DeZKCA3PNQ7 Dd1_xOzBHv1 Dep0wuhvizo The one rule every serious workflow shares: make Claude write a storyboard (spoken line, what is on screen, which tool makes it) and approve it before it renders. De8SPjEAuV0 DeUUFJQu8cw De2SEI7JP8r Default AI output looks the same for everyone ("Claude flop", "slop", "it looks like a claude edit"). The fix is art direction: references, a brand sheet or design system, a fixed palette and font, and banned effects. DemBKpbxSyd DeZKCA3PNQ7 DevtyU1kmHC De2SEI7JP8r De1xXtICvqs DeUUFJQu8cw

Followers are not money. Customers are. Build the brand around one problem you have actually solved, repeat one message, show your face, mistakes and results, then sell an owned product (course, community, mentorship, service). "If you don't sell something, you can't make money." DeERlGvM9LK A follower-maximising formula exists (be the underdog, be credible for your exact topic, go on a visible quest with an open loop), but its author says it brings followers, not money or fulfilment. Pair any growth tactic with a conversion path. Dd7_w1wsdKf [data, one outlier] The profile is the sales page: smiling photo, name field with a search keyword, "I help X achieve Y without Z", proof, a DM keyword, one application link, three pinned posts (best client result, your story, deep-value post). DdrSAMWq-TP Dd6ZIHnBBwC

Respond to inbound leads in seconds, not hours. A $1.4B remodeler called a test lead back in about 5 seconds; the 60-second rule is the benchmark to beat. One commenter says instant calls hurt his trust, so test timing per business. Dcjrk_HxEPR DdpPxHkIsuN Fix lead quality at the source, not with more leads: feed Meta qualified-lead events (Lead, MQL, SQL, Purchase) through CAPI, cut junk placements (Audience Network, Messenger, Threads) after checking the Placement breakdown, and exclude areas that never qualify. DcsX3MryYdy Db3L0XITUbY DdoiULSzAw_ For your own agency, a direct-hook Meta ad straight to a Calendly page (pixel installed on Calendly, multiple countries tested) is claimed to close Rs 5.5 lakh of international clients in 3 months with zero cold outreach. DdZgna8Tzoh

Know how saturated your market is. Five levels: only seller (claim anything), bold claim plus unique mechanism, proof, guarantee plus extra offers, then brand (be remembered). Build every offer to compete at level 5. DaqtApmqK28 Put a decoy next to your best seller. One extreme, rarely bought option gets attention and makes the normal option look like the smart choice (Tesla Rocket / Thrust $250,000 / Performance $50,000). DekuqC0IpMG Sell big tickets through a smaller paid first step, not a discount: a $3,000 strategic assessment before a $50,000 project lets the buyer experience your thinking and lowers risk. DfAKJ9LCxql

Process beats model. Same AI, different outcomes depending on the checklist, context and tests you give it: a 20-item security checklist made the difference between Claude and Codex sites; CI/CD with unit, integration, E2E and performance tests is what turns vibe coding into engineering; skipping evals brings hallucinations back. DeM46Iay29L DfdtXLfAim5 DfJXMRLojO1 An agent cannot reliably check its own homework. Use a separate auditor (Cloudflare's open-source security-audit-skill: map the app, hunter agents per input, a fresh verifier tries to disprove each finding) and run it more than once, before every launch, because one run finds about half. De9bqWtoARc DfH6Z5NzECg Best AI uses for creators and marketers: research and data/pattern analysis (S tier); scripting, ideation, funnels, design (A-B); admin (C); editing (D-C); replying to DMs (D-F, keep human). Ddt4Tq0BGS_

Early on, the product is the brand. One ranking puts product experience 1, positioning 2, founder story 3, brand codes 4, consistency 5, name 6, guidelines 7, logo 8, purpose 9. Opinion only, heavily disputed in comments. DdjrOhAx9yE A logo handover is not finished without: a Safe Space page, a dedicated inverted (dark background) version, a tiny icon version for favicons and profile pictures, a minimum-size guide, and a "Logo Misuse" page. Name one person as "brand protector". DbCzQQwRpgZ In a crowded market, claims and features stop working. The ladder goes claim, unique mechanism, proof, guarantee, then brand (being remembered). Fine jewellery and Indian supplements are near the top of that ladder. DaqtApmqK28

Do not copy a commodity product from a viral "buy at Rs 40, sell at Rs 250" reel. The incumbent brand outspends you and owns trust. Check competitors in the Meta Ads Library and do the real per-order margin (price minus COGS, ads, RTO, returns, packaging, GST) first. DXd5NR2DvUZ Set goals as numbers and back-solve them from conversion rates (60 customers a day at a 20% close rate means 300 conversations). Dd7E0vbAWGQ Scaling ad spend has diminishing returns: the platform serves the likeliest buyers first, so each extra rupee buys a harder sale. Judge scale on marginal CAC, raise budgets in steps (about 20%), consolidate campaigns, and scale with creative, landing page, offer and product levers. DeRzDazODdV [data, illustrative] DdtatLDiS9g

For Indian agencies receiving foreign payments, compare fees, forex markup and FIRA/FIRC automation. One ranking: Skydo S, Wise A, Stripe / bank wire / Razorpay B, PayPal C. Possibly sponsored; check current fee pages. DdtoCqICWgU Protect Meta assets with separation: a host Business portfolio owns the Facebook page, Instagram account and pixel; an operating portfolio holds ad accounts with partner access; a backup portfolio is ready. Multiple real admins. Commenters warn that sloppy multi-BM setups trigger restrictions. DdwDFt8Bsoj India COD reality check: a US mentor claims 80% COD and only 40% delivered for a dropshipping student despite a 15% prepaid discount; Indian operators say RTO is closer to 10 to 20%. Measure your own delivered rate by courier and pincode, and report "delivered ROAS". Dc1rL8hSzS7

To get cited by ChatGPT: build a page whose headline is the prospect's exact question, make sure answers are in the server-rendered HTML (Ctrl+U then Ctrl+F test), and state an opinion. A 1-month-old agency site claims 16 ChatGPT leads this way while not on Google page 1. Ddl1taBAppw AI answers are easy to steer for low-competition queries: a $6 press release with the phrase in the title and first sentence changed Google AI Overview within hours. Monitor what AI says about your brand. Do not fake events. DdsLYSUPwGQ Google's guidance (updated 1 Oct per the reel): manually fact-check and review all AI-generated content, including titles, meta descriptions and image alt text, before publishing. DeBJExFPKUF