The 1-Hour AI Marketing Workflow

Before you start the clock

Do not start the hour without these three. The first time you run this, gathering them is the work; after that they live in the Campaign Brain and the clock can start on time.

If you don’t already have the voice, offers and past winners loaded somewhere permanent, build that first — the 45-minute Campaign Brain setup is the version we run, so this hour doesn’t start with a paste. How the team actually uses AI across drafting, visuals, ads and analysis is in How we’re using AI in our marketing.

Minutes 0–12: draft the pillar piece

Draft the pillar in Claude — 1,200 to 1,800 words. That’s the asset everything else is cut from, so it has to exist before social, email or ads. Cowork can sit in a shared folder if the sources are already there. ChatGPT or Gemini will also do this job; Claude’s edge is prose and long source documents, which is why it’s our default for the pillar.

Twelve minutes is a first draft, not a publish. You are aiming for structure, the argument, and the offer in the right place. Polish comes after Gate 1, not before. If prompting itself is still the shaky part, the five ingredients and 13 starter prompts will get a cleaner first pass than “write a blog about this.”

Minutes 12–22: QC Gate 1 — fact-check

This gate is not optional and it is not a skim. Every number, name, date, quote and citation gets checked against a source you can open. Anything that can’t be verified comes out, even if it sounded good.

Then add one human-only specific the model could not have invented: a guest quote, a number from your own dashboard, a named competitor the brief actually named. That single addition is usually the difference between a usable draft and interchangeable slop. Treat confident-sounding output as unverified until someone has actually checked it — the same rule as the 90-second pass in our AI guardrail playbook.

Minutes 22–30: cut into social

Cut the pillar, don’t rewrite it from scratch. Four shapes from the same argument:

• LinkedIn — the professional cut, one idea, a reason to comment or save.
• Instagram / Facebook caption — shorter, one image-led thought, one action.
• Short-form hook or script — the first three seconds, then the same point in spoken rhythm.
• X thread — the argument broken into posts, not a caption with line breaks.

Visuals run in parallel via Canva Magic Media, not after. A designer still finishes anything that will run paid or on a brand account; the model gets concepts and first frames moving so the eight minutes aren’t spent staring at a blank artboard.

Minutes 30–38: email

One nurture or newsletter from the same brief. Two or three subject lines. One CTA — the same behaviour the brief named, not a new one. Then a prospect version and an existing-customer version: same argument, different knowledge of the reader. Eight minutes is enough for that split if you don’t let the model invent a second offer.

Minutes 38–48: ad variants

Three to five headline / description / primary-text variants for Google and Meta, cut from the pillar and the offer. Claude drafts; Google AI Max / Performance Max and Meta Advantage+ test. Meta reports an average 22% ROAS lift on Advantage+ Sales versus manual — that’s an optimisation claim, not a reason to skip the draft or the brand check.

2026 rule, don’t skip it: AI-generated ad elements carry an automated-content label. Disclose what the platform requires. Do not spend against variants nobody has read. Spend approval is not part of this ten minutes — it sits in the list of things we don’t automate.

Minutes 48–55: QC Gate 2 — voice and visual

Read the pillar, the social cuts, the email and the ads side by side against the brand voice one-pager. Then the visuals in the same pass. You are looking for one campaign, not four channels that happen to share a keyword. If a caption has drifted into corporate-neutral, or an image could belong to a competitor, fix it now. Seven minutes is enough to catch that. It is not enough to rewrite the hour — which is why Gate 1 exists.

Minutes 55–60: QC Gate 3 — approval and publish

A named person approves and publishes, or parks anything that still needs a claim check or a spend sign-off. The hour ends with a decision, not a folder of drafts. “Someone will post it later” is how the hour leaks into Thursday.

What we don’t automate

The hour is a drafting machine with three human checkpoints. It is not an autopilot. These six stay human on purpose — they are the line in Create, Understand, Act that we don’t cross, applied to this clock.

The tool stack

Claude is the default for prose. The rest of the stack is whatever already sits in the channel. Swap the left-hand column if you have to; do not swap out the gates.

On a hotel client, this sequence took a content timeline from two weeks down to three days once the brand context was loaded. We’ve trained 5,000+ people across the Philippines and Australia since 2017, at 98% satisfaction, and in one 4,000-person rollout 72% were using AI weekly within 30 days. The pattern that sticks is not a new tool. It’s a clock with gates.

Want it run live on your own accounts, with your briefs on the table? That’s the second half of Claude for Marketing. Team-wide, it’s AI training in the Philippines. If you’d rather someone else own the operating rhythm, that’s AI consulting.

Keep going: the AI guides hub has the rest of the marketing series, and the Insights archive carries everything else the team publishes.