First-party implementation note · SEO and AEO
NorthSignal internal build
One governed loop
Search evidence to verified action
NorthSignal had Search Console, Ahrefs, ZeroRank, content, and analytics processes, but the scheduled system was failing before it could produce a daily decision brief. This implementation separates health, collection, diagnosis, approval, execution, and measurement so a broken source cannot create unsupported work.
Evidence record
What is verified now, and what still needs time.
611 impressions and 2 clicks in the latest complete 28-day period
Search baseline
0.33% before the roadmap implementation
Click-through baseline
79.1% of visible-query impressions at position 21 or worse
Ranking-mix baseline
0 NorthSignal mentions in the tracked 30-day ZeroRank set
Answer visibility baseline
Human approval remains required before publication or external action
Control implemented
Measurement begins after production release. No ranking lift is claimed yet.
Outcome state
The Problem
More impressions arrived. Fewer clicks followed.
For the latest complete 28-day Search Console period ending August 30, 2026, NorthSignal recorded 611 impressions, 2 clicks, a 0.33% click-through rate, and an average position of 30.8. Visible query detail placed 79.1% of impressions at position 21 or worse. ZeroRank recorded no NorthSignal mentions in its tracked 30-day prompt set. The automation that should have diagnosed those changes was configured but its recent run failed across several upstream dependencies.
The Architecture
A fail-closed operating loop with an evidence receipt at every stage.
The implementation keeps deterministic preflight, source snapshots, freshness checks, change detection, strategy diagnosis, one bounded draft, human approval, production verification, and 7, 28, 56, and 90-day measurement as separate gates. A failed source marks dependent work blocked but still produces a health brief. Public changes cannot be inferred from a generated draft or a scheduler entry.
Under the Hood
How the agent works.
01
Preflight
Check dependencies, source access, time zone, and prior receipts before collection.
02
Collect
Write timestamped GSC, Ahrefs, ZeroRank, and analytics snapshots with freshness.
03
Diagnose
Classify the opportunity as rank, intent, click-through, authority, technical, or measurement related.
04
Approve
Present one bounded recommendation with evidence, expected signal, risk, and review date.
05
Verify
Confirm the exact production URL, canonical, structured data, links, analytics, and receipt.
06
Measure
Compare the pre-change baseline after 7, 28, 56, and 90 days.
Design Principles
What this architecture is built on.
01
A scheduler entry is configuration. A healthy receipt and verified output are operating evidence.
02
A failed source should block dependent decisions without silencing the daily health report.
03
The content queue should respond to qualified demand, ranking feasibility, and business fit instead of article volume.
04
Clicks, qualified actions, relevant links, and AI citations determine whether the system is useful.
Next Step
Want an agent built for your business?
The free Agentic Audit shows which customer-growth leak your business is carrying and which agent closes it first.