Most SEO tools tell you what to change. Rankwell ships the change, watches the metric, and earns trust one verdict at a time — confirmed wins amplify, refuted ideas step back.
Closed beta. Designed first for stayFritz; opening to peer operators through 2026.
Each change is recorded with a target metric, an expected delta, and a measurement window. The loop never short-circuits — wins are wins because the data says so.
The Cockpit shows what the system did this week. The Hypothesis Card shows each change as a falsifiable bet. Measurement closes the loop with real data.
Cockpit. Real GA4 booking value, real GSC clicks/position, agent strip with last-run times derived from the action ledger. Source badges everywhere — you always see where each number came from.
Hypothesis Card. Statement, before/after diff, brand-voice gate, expected-vs-actual chart, what the system learned. Every approval is an experiment, not a vibe.
Measurement. Verdicts over time. Confirmed wins lift the reinforcement weight on that action type; refuted ones step it back. The next backlog is a function of the last one.
Specialised agents own one slice of the loop each. They share a schema, swap reinforcement signals, and run on cadence — not on a human's free Tuesday.
Crawls every page on a cadence, flags structural and technical issues, records what the site looks like today.
Scores the backlog by real keyword demand (DataForSEO) × inbound authority × reinforcement weights from past wins.
Drafts meta, schema, and copy on-brand. Every draft is gated against your banned terms — copy never ships if it trips a rule.
Repairs internal authority: weak hubs, orphans, broken canonical links. Auto-applies the changes you've pre-approved.
Reads GSC and GA4 once the window closes. Confirmed wins amplify; refuted ideas step back. The loop is what makes it learn.
Sauna and wellness framing on the destination meta. Expected +0.7pp. Actual +0.9pp. Booking-value delta +€2,400. The reinforcement weight for meta-rewrite ticked up to 1.10, and the next rescore favoured sauna queries on neighbouring clusters. This is what one closed loop looks like.
Booking value from GA4 organic sessions. Click + position deltas from GSC. Keyword volume from DataForSEO. No vanity scores invented to make the dashboard green.
Each action carries a target metric, an expected delta, and a measurement window. After 4-12 weeks the verdict lands: confirmed, refuted, or inconclusive.
Your banned-words list is enforced at the publish boundary. The Content agent refuses to emit copy that trips a rule. You set the voice once; the system can't drift.
Multi-tenant from row one. Every query carries a tenant_id, every credential is encrypted per workspace. Your data never crosses a customer boundary.
Verdicts land 4-12 weeks after a change. We don't pretend otherwise — we make the wait visible, attributable, and worth it.