Back to All Case Studies
marketingFull Access Unlocked

An AI Content Engine That Cannot Publish Without a Human

Every AI writing tool promises you stay in control. We built one where that is a missing function rather than a marketing line — the publish tool does not exist in the automation surface at all.

In-house product · AI SEO content tooling

marketing
12
publish Destinations
35
locales
0
auto Publish Tools

!The Need

AI content tools compete on how much of the pipeline they can automate, and publishing is the easiest step to automate away. The result is unreviewed machine text going live on a company's own domain at volume, where a single fabricated statistic or misattributed source is a reputational problem that outlives whatever traffic it earned. Every tool in the category ships the phrase "you stay in control" as copy. Almost none of them can show you a mechanism, and a setting that can be switched off is not a mechanism — it is a default.

The Approach

We inverted the usual design: the approval gate is the fixed point, and the rest of the product is built around it. The automation surface exposes no publish tool at all — not a disabled one, an absent one — so an agent driving the product cannot publish even if instructed to. Automatic generation has no publishing counterpart. Publishing defaults to draft status, and the gate lives inside the publish step itself, so scheduled jobs and bulk actions cannot route around it. Around that constraint runs a full pipeline: keyword and search-results research, a nine-check domain audit, then research, draft, score and fact-check for each article, with per-workspace voice profiles. The whole product is open source, which is the point — the claim is not "trust us", it is that you can search the codebase for a publish tool and fail to find one.

Technologies Used

ClaudeModel Context ProtocolDataForSEOSupabaseNext.jsTypeScript

The Output

An automation surface with zero publish tools — the capability is absent, not merely switched off
Publishing defaults to draft, with the approval gate inside the publish step so bulk and scheduled actions cannot bypass it
A four-step article pipeline: research, draft, score, fact-check
Twelve publishing destinations, ten of them content management systems, plus git and webhook targets
Thirty-five locales, per-workspace brand voice profiles, and a nine-check domain readiness audit
Rank tracking, search-console and webmaster analytics, and a check on whether AI answers name you or name someone else
The entire product open source — no feature-gated tier, no closed enterprise directory

The Impact

The control is checkable rather than promised: search the codebase for a publish tool and you will not find one
An agent operating the product autonomously still cannot put a word on a live domain
Generation scales without the review step scaling with it — volume is decoupled from risk
This is a product still before launch, with no customers and no growth figures attached, and that is stated deliberately rather than dressed up
The design constraint is the differentiator: competitors ship the same sentence as copy, and copy is not verifiable

Ready to achieve similar results?

Contact Us
Supalabs AI solutions