A 20-asset RSA starter pool (5 headline themes × 4 description angles) explodes to 1,000 strategically-curated combinations. Gemini does the combinatorial math and pre-flight validation, Google Ads Editor ships the bulk upload. A workflow that lifted one ecom account's CTR by 27%.
A real funnel that turned 2,400 candidate keywords into 87 ranked targets — 22 of them top-10 within four months. The five filters, the Gemini classification prompt with explicit JSON output, the funnel math (2,400 → 480 → 280 → 140 → 100 → 87), and why trusting KD score alone is the most common way to waste six months.
A 1-hour batch workflow to ship a 3-bullet TL;DR box to 50 old posts using sitemap scraping, a strict Claude rubric, and a JSON paste-in. The highest-leverage content refresh you can ship this week.
A 30-day blind test: 800 meta titles + 800 product descriptions rewritten by both Qwen 2.5 14B (self-hosted on a refurbished workstation) and Claude Sonnet 4, rated by 3 SEO contractors. The result is not a clean win for open weights — it's a split. Where Qwen breaks even, where it collapses, and the actual cost math behind self-hosting for repeat-pattern SEO work.
A production 3-model content pipeline where each AI does only what it is actually best at: ChatGPT drafts, Claude reviews, GPT-Image generates visuals. The actual hand-off prompts, the JSON contract between stages, the $0.41 / 18-minute cost and timing comparison vs single-model, and the model-identity-confusion failure mode that cost me a published post.
A one-page Google Sheets dashboard with 8 marketing KPIs and a Gemini =AI() cell that writes the week-over-week narrative for each metric. Built in 90 minutes, fits in a Slack screenshot, kills the 20-minute Monday meeting.
An n8n + Claude Haiku 4.5 agent that watches 9 subreddits, scores every post on a 3-axis rubric (intent, urgency, fit), and posts a top-5 daily digest to Slack — with the verbatim Claude prompt, the 9-sub shortlist, the cost math, and the week-2 finding that pruned it down to 4 subs for a +120% signal lift.
A 24-message winback matrix — 4 segments × 6 emails — for buyers who haven't purchased in 60-90+ days. Includes the Klaviyo segment conditions, the cadence, the Claude copy prompt, and the $48,200 case study.
A test-matrix methodology for cross-channel paid media — 5 channels × 4 audiences × 2-3 creative concepts, structured in a spreadsheet, with minimum spend math, Day 7 vs Day 14 kill rules, and a $90k B2B SaaS case study where a founder LinkedIn video beat every paid creative on cold prospect.
Most 'translated' content fails in China not because of vocabulary — it fails because the examples don't translate. A 4-step EN-to-zh pipeline that turns an English pillar post into a zh version that reads like it was written there, with the actual prompt, a swaps table, and a before/after.
A 4-step workflow that turns a B2B SaaS sitemap into a hub map, pillar/spoke gap list, and an internal-link scoring matrix — using Claude's long context. The actual prompts, the actual JSON, the actual 60-day result.
Self-hosting a 70B model sounds reckless for a marketing team. For 90% of teams it is. But there are 4 specific jobs — bulk ticket classification, private competitive intel, overnight SEO meta-generation, PII-redacted list cleaning — where the math flips and a single A100 + Ollama pays for itself in 4-7 months. Hardware reality, Docker compose, real throughput, and the 4 prompts.
