How I Use AI to Run a Digital Marketing Agency: My Actual Tool Stack and Workflows

Every AI productivity article makes the same promise: "Here's how AI can transform your workflow." Then it lists a dozen tools you've never heard of and describes hypothetical use cases that sound impressive but don't match how anyone actually works.

This post is different. It's my real workflow — the specific tools I use every day, the exact tasks I've handed to AI, what I still do manually, and how much time I estimate AI saves me each week. No hypotheticals, no sponsored recommendations, no tools I tried once for a blog post and never opened again.

I run Codeble — a digital marketing and web development agency based in Sydney. My work spans content strategy, SEO, AEO (Answer Engine Optimisation), Google Ads management, marketing automation with Dynamics 365, web development in Next.js, and analytics with GA4.

Here's how AI fits into each of those.

My core AI stack

That's it. No $500/month enterprise platforms. No twelve overlapping subscriptions. Two paid AI tools plus free tiers of supporting tools. The ROI on that $62/month is significant enough that it's the last expense I'd ever cut.

Content creation: where AI saves me the most time

Content is the biggest part of my workflow and where AI has the most impact. I produce blog posts, email campaigns, social media content, proposals, and strategy documents regularly.

Blog writing workflow

Before AI: Research (2 hours) → Outline (30 min) → First draft (3–4 hours) → Edit (1 hour) → Format (30 min) = ~7–8 hours per long-form post.

With AI: Research with Perplexity (20 min) → Outline in Claude (15 min) → First draft in Claude (45 min) → My editing pass — adding real examples, opinions, voice (1.5 hours) → Format (15 min) = ~3 hours per post.

Time saved: ~4–5 hours per blog post.

The key is how I use Claude for the first draft. I don't just say "write a blog post about X." I give it my voice profile, a detailed outline, reference material from research, and a "never do" list (no buzzwords, no filler, no generic openings).

Email campaign copy

For marketing automation using Dynamics 365, I draft email sequences in Claude. Claude produces the sequence in 2–3 minutes. I review, adjust the tone, and add specific details. What used to take a full afternoon now takes about 45 minutes.

Social media content

I use ChatGPT for social media brainstorming and Claude for the actual writing (using my saved style profile). Weekly social media time is now 30 minutes, down from ~2 hours.

SEO and AEO: AI as research assistant

Keyword and content gap analysis

I export data from Google Search Console and SEMrush, then upload to ChatGPT. It finds patterns in the data much faster than I can manually. It doesn't replace my judgment, but it surfaces opportunities I might have missed in a spreadsheet of 500 rows.

Schema markup generation

For structured data (Article, FAQPage, LocalBusiness schema), I use Claude to generate valid JSON-LD I can paste directly into the page head. What used to take minutes takes 30 seconds.

AEO auditing

For AI search visibility checks, I manually test queries across ChatGPT, Perplexity, and Google AI Overviews. This is one area where AI doesn't help much — the audit itself requires human judgment about whether citations are accurate and how competitors are being cited.

Google Ads analysis: AI as data analyst

I manage Google Ads campaigns where performance data needs regular analysis. This is where ChatGPT's data analysis capability earns its subscription cost.

Weekly performance review

I export campaign data from Google Ads and upload the CSV to ChatGPT. I use specific prompts to compare performance to the previous 4-week average, flag campaigns where CPA increased, and identify search terms with zero conversions.

Performance Max asset review

For PMax campaigns, I export asset reports and ask ChatGPT to identify which headlines and images are performing best — then suggest replacements for underperformers.

Web development: AI as coding partner

I build websites in Next.js with Tailwind CSS, and AI has fundamentally changed my developmenet workflow.

Claude Code for architecture and features

For complex development work — building components, implementing features, debugging — Claude Code is my primary tool. It handles component scaffolding, debugging, refactoring, and documentation well.

Cursor for daily coding

For everyday development — smaller changes, quick fixes, inline suggestions — I use Cursor's AI-assisted IDE. It saves about 30–40% on coding tasks by reducing typing and speeding up debugging.

Client communication: AI as drafting assistant

Proposals and quotes

For client proposals, I use Claude with a template structure. The AI handles 40% of the mechanical drafting; I handle 60% of the custom strategy and investment details.

Meeting follow-ups

After client meetings, I draft follow-up emails in Claude using my meeting notes. This takes 2 minutes instead of 15. The email is accurate, well-structured, and professional.

What I deliberately don't automate

Not everything should involve AI. Here's what I keep manual:

• Strategy and recommendations. Deciding what a client should actually do requires business judgment and experience AI doesn't have.
• Client relationships. Difficult conversations and building trust are fundamentally human.
• Creative direction. The direction and point of view must come from my understanding of the market.
• Quality assurance. Every piece of AI-assisted work gets a human review before it reaches a client.
• Ethical judgment. Deciding what data is appropriate to upload and what claims are defensible requires professional responsibility.

The real time savings

~17–18 hours saved per week. That reclaimed time goes back into strategy, client relationships, and business development.

Key takeaways

• My entire AI stack costs $62 AUD/month — Claude Pro + ChatGPT Plus.
• AI saves me roughly 17–18 hours per week across agency workflows.
• Claude handles 70% of my AI work — it is strongest for writing and coding.
• ChatGPT handles the other 30% — image generation and data analysis.
• AI output is always a draft, not a deliverable.
• The time AI saves goes into the work AI can't do: strategy and relationships.