For Pixovo — my AI product photography SaaS — I built what I thought was a growth machine: a complete content marketing pipeline where AI did everything except decide to exist. It published 117 posts across four platforms. It brought zero users.
This is the kind of case I'd rather read than live through, so here's the honest version.
The system
The pipeline had three layers, all living inside Pixovo's admin panel:
Content planning. Claude Code generated a 30–90 day content plan as a JSON file: publish dates, topics, and target platform for every slot. Instagram got carousels twice a week; Threads, X and Facebook got text-plus-image posts once or twice a day.
Generation. The admin panel ingested the JSON and produced ready-to-publish posts. Texts came from different LLMs per platform — I used Grok for X because it wrote in the platform's register better than anything else I tried. Images came from Google's image model, good enough that individual cards and carousels genuinely looked professional.
Publishing. A scheduler laid posts out on a calendar and cron jobs published them automatically — separate OAuth integrations for X, Instagram and Threads (Threads runs a completely separate Graph API from Instagram, which cost me a day to discover). Another cron synced engagement metrics back into an analytics dashboard.
I also built an engagement tool on top: it searched X and Threads for relevant conversations, drafted replies, and published them in a couple of clicks.
The result
Over the campaign: 117 published posts, three connected platforms, and effectively zero organic traffic. No signups attributable to social. Views were negligible.
The whole suite still works — Pixovo is closed as a business, but I preserved it as a live demo, including the admin panel with the real posting history and analytics. You can click through the actual system that produced this failure.
What I learned
Posting is not marketing. Distribution without engagement is a broadcast into a void. The accounts that grow are in the replies, in the threads, reacting fast to what's happening — not shipping perfectly formatted content on a calendar.
People smell AI content and skip it. Even with genuinely good images and platform-tuned copy. In 2026 the reflex is instant: pattern-matched as AI, scrolled past. Polish doesn't fix it; provenance is the problem.
The unit economics of engagement tooling are brutal. The X API alone is $200/month, which kills scaling a reply-automation tool for a product doing zero revenue.
Organic first, paid second. If organic traction exists, paid will amplify it. The reverse doesn't hold — paid traffic doesn't teach you how to earn attention. I automated the appearance of a content strategy before proving anyone wanted to hear from us at all.
The most expensive part wasn't the build — it was the weeks of believing the metrics dashboard would eventually turn green on its own.
Originally discussed (in Russian) on my Telegram channel. The full system — admin panel, scheduler, analytics — is clickable in the Pixovo live demo.