In May 2026, I designed and deployed a complete AI-powered content system for A1 Employment — an Australian recruitment and migration consultancy serving Vietnamese students and workers across Australia and New Zealand. Using Claude AI as the operational core, I built six custom Skills covering automated news monitoring, GEO blog content, Facebook banners, and Elementor pillar pages — all within a single Claude Project, with no Make.com, no additional tools, and no external agency.
The Challenge
A1 Employment operates in a highly policy-sensitive market. Their audience — Vietnamese students and workers navigating Australia's visa system — reacts directly to announcements from Home Affairs, the Fair Work Ombudsman, and Immigration NZ. Missing a key policy update means missing a high-engagement post and potentially losing a lead to a faster competitor.
Before May, the content operation lacked three critical components: (1) a systematic news monitoring workflow rather than manual searching; (2) a clear SEO/GEO standard for WordPress blog articles; and (3) a repeatable banner production process consistent with A1's brand guidelines. Make.com had been evaluated and rejected — too slow to load, too prone to breaking on edge cases.
The Approach: Claude Projects as the Operating System
Rather than building isolated tools, I designed the entire system around Claude Projects + Claude Skills. The architecture is simple: the Project is the permanent client workspace — brand identity, ethics, tone, team bios. Each Skill is a specialist module loaded on demand. This keeps context lean and outputs consistent without bloating the conversation.
1. Map the workflow before building anything
I started by mapping A1's entire content operation into four stages: News Monitoring → Content Creation → Visual Production → Web Publishing. Each stage has defined inputs and outputs. This meant each Skill could be designed independently but assembled seamlessly when the workflow called for it.
2. GEO-first, not SEO-first
Every blog article written for A1 is optimised against GEO criteria: fact-dense sentences with at least one specific data point per paragraph, full entity anchoring on first mention (e.g. "Visa 482 (Temporary Skill Shortage visa)"), and FAQ-ready H3 headings phrased exactly how users ask Perplexity or ChatGPT. The goal is to be cited, not just ranked.
3. Playwright instead of Canva API for banners
Instead of relying on Canva's API (paid, inflexible on layout), I built the a1-banner skill to generate pure HTML/CSS, then use Playwright to screenshot and export a 1080×1080px JPEG. Each banner is produced in under 90 seconds, pixel-perfect to brand, with no manual design work required.
What Was Built in May
Six complete systems were deployed across the 30 days:
.a1-blog.Execution Timeline
- Week 1 (May 1–9): Built the
a1-bannerskill and produced the first Facebook post — a Sales & Marketing Executive hiring banner. Confirmed the Playwright export workflow was stable inside Claude's container environment. - Week 2 (May 10–16): Built
a1-geo-contentv1 and v2. Wrote the first 5 pillar articles (Visa 482, international student job search, regional hospitality, Australian CV standards, Fair Work rights) and compiled them into a CSV batch import for Google Sheets. - Weeks 2–3 (May 13–22): Upgraded all skills to English-only output. Redesigned
a1-geo-contentto v3 — sidebar-aware layout, white background,.a1-blogCSS scope. Builta1-page-builderfrom scratch; delivered the A1 Blog Hero + CTA page. - Weeks 3–4 (May 20–30): Ran the weekly news brief 4 times (May 20, 25, 27, 30). Each brief covered 4–6 major stories across 10 sources with recommended post copy directly below each headline. Built the Healthcare Migration Pillar Page (~2,000 words, full schema).
- Bonus — Visa 189 Article: Converted a news flash from the brief into a full GEO blog post to the A1 Standard v3 spec — from brief to publish-ready HTML in approximately 15 minutes.
Results & System Impact
The most significant outcome is the change in operating tempo. Before May, creating one Facebook post took 20–30 minutes: research the news, write, translate if needed, format. After the system was live, a complete news brief — 4–6 stories with copy-paste post copy — takes 3–5 minutes. A full GEO blog post with schema, fact cards, timeline, and FAQ takes 10–15 minutes instead of several hours.
- News monitoring: 0 → 10 sources tracked in parallel every week, with no major policy announcement missed (482 salary threshold rise, 2026–27 Migration Program finalisation, Fair Work annual wage review)
- Content volume: [METRIC NEEDED: total Facebook posts in May] Facebook posts + 6 GEO blog articles + 2 full pillar pages delivered
- Banner production: From 0 (no system) → branded JPEG export in under 90 seconds per banner
- Skill reusability: All 6 skills are fully documented and architecture-agnostic — cloneable to a new client in under 2 hours
- Additional tech cost: $0 — the entire stack runs on Claude Projects + Python stdlib + Playwright (free tier)
Want to build a similar AI content system for your business?
I take a maximum of 3 new consulting projects per month — specialising in Education, Migration, Events, and Professional Services.