What we set out to test
Vibe marketing — using natural language prompts to generate complete marketing assets rather than building them manually — has been discussed in tech circles for a while. In May 2026 it crossed into mainstream business coverage, with multiple marketing publications treating it as a default working style for small teams and solo operators.
We wanted to know whether it actually works as a full content production workflow. Not a single blog post or a batch of social captions. A real 30-day content operation: weekly blog publishing, daily social posts across two platforms, newsletter copy, and engagement targets. All driven primarily by conversational prompting with an AI assistant.
📋 Experiment Setup
Duration: 30 days. Output targets: 3 blog articles/week · 2 social posts/day/platform · 1 weekly newsletter. AI tool: Claude (claude.ai). Human review time: 15–25 minutes per article, 5 minutes per social batch.
The numbers
A single blog article from brief to published HTML took approximately 60–75 minutes total — of which roughly 20 minutes was human, 40–55 minutes was AI generating while other tasks were handled. That’s the number that matters.
What worked better than expected
Consistency was the biggest win. Maintaining a consistent brand voice across 12 articles written in separate AI sessions was better than expected. With a clear voice document and specific instructions, the output stayed recognisably BuzzRiding across the month.
Repurposing was nearly instant. Taking a finished article and prompting for social posts, newsletter teaser, and engagement comments took under 10 minutes. Manually, that work takes 45–60 minutes. For a content operation running 3 articles per week, this alone saves roughly 2.5 hours per week.
SEO structure was reliable. H2 hierarchy, FAQ sections, meta descriptions, internal linking — all produced correctly on first pass. No SEO specialist required.
What failed or underperformed
Article openings required consistent rework. Every single intro needed a human rewrite. AI-generated openings are structurally correct but recognisably formulaic.
Social post personality was the hardest problem. The BuzzRiding voice on X and Bluesky requires a specific kind of practitioner bluntness. AI output trends toward agreeable and slightly generic. We rewrote approximately 30% of queued posts before scheduling.
Data fabrication required vigilance. Every article required at least one round of live search to verify or replace data points. In a high-volume workflow this becomes a consistent time cost that the “just prompt and publish” framing ignores.
The one thing we’d change
Invest more in the brief, not the output. The gap between a generic prompt and a detailed brief is enormous. Articles produced from a 10-minute brief took 5 minutes to review. Articles produced from a 2-minute prompt took 25 minutes to review and often needed structural rewrites. The counterintuitive lesson: the faster you want the output, the more time you should spend on the input.
Related reads: I Used AI to Write a Month of Blog Posts · How to Use ChatGPT for Content Marketing