How it works

One loop, end to end — and a human holds the gate.

Seven stages turn a keyword gap into a published, measured, attributed article. The system runs the work; you approve what ships. Then it reads the results and decides what to write next.

STAGE 01 · MINE

Find the gaps worth writing

Rank Tuner mines competitor keyword gaps and the queries you already get impressions for but no clicks. It scores and clusters them into a backlog of ~150–200 topics, each with a brief — no blank-page guessing.

/keywords
STAGE 02 · RESEARCH

Grounded in live sources

For each topic, the model researches the live web with the search tool, capping searches and preferring trusted domains. Every source is logged to the task’s provenance row, so you can see exactly what informed the draft.

/workspace
STAGE 03 · WRITE

Your team’s voice, not a template

Drafting uses a persona built from real posts and a voice spec, with the static company context prompt-cached so every article is cheap and on-brand. Output is structured Portable Text, ready for the CMS.

/workspace
STAGE 04 · THE HUMAN GATE

A feature, not a limitation

A claim-validation pass extracts every factual and client claim and grounds it — allow-list match for client claims, web search for facts. It only ever flags. You approve, edit, or reject. Nothing reaches the CMS without a person signing off. On a real domain, this is what keeps you on the right side of the scaled-content policy.

/review
STAGE 05 · PUBLISH

Verified, then syndicated

Approved articles publish to the CMS on a schedule. The system re-fetches the live URL and confirms the revision, canonical, and a 200 before marking it published — never fire-and-forget. Then it pings IndexNow, submits to Search Console, and syndicates with the canonical pointing home.

/plan
STAGE 06 · MEASURE

Every publish, against the market

A daily pull aggregates rankings, traffic, and AI-answer citations onto one timeline, with publish events and Google core-update markers overlaid. Late or missing days render as gaps, never as zeros — so the analysis is honest.

/analytics
STAGE 07 · STEER · THE BRAIN

The loop closes

A daily job reads those results and re-scores the backlog: double down, refresh, or drop. Decay is statistical — period-over-period, slope, CTR-vs-impression divergence — never a hardcoded number. The output of measuring is the input to mining. That is the loop.

/recommendations

It researches, writes, and steers. You decide what ships.