AISEO ManagerThe SEO platform that explains, decides and ships.
Scans every page, diagnoses why it underperforms, drafts the fix, and ships it through an approval-gated queue, built for BOLD Precious Metals.
- Every AI fix is validated against the live page, never invented.
- Nothing ships until a human approves the task.
- Every AI call is logged with tokens and cost.
- Client / status
- BOLD Precious Metals · in production
- Business outcome
- Weeks of manual SEO review → validated AI scans and shippable fixes
- Risk control
- Approval-gated task queue · every AI call cost-logged
- What we delivered
- Architecture, build, integration, operate
SEO tooling tells you what. This tells you why, and does it.
Rank trackers show the number dropping; audits pile up in a spreadsheet nobody actions. AISEO finds the weak pages, rewrites them, explains why each one underperforms, and hands your team a task to approve.
Smart Content Helper
Scans site pages, extracts keywords, detects cannibalization between pages, and drafts AI content rewrites, validated against the real page so the model can't invent fixes.
Diagnosis, not dashboards
A specialist-agent pipeline scores content quality, trust, structure and links, then a Master agent reconciles the evidence and a Remediator writes the task brief.
Approval-gated execution
Every AI recommendation lands in a team task queue with owner, priority and status. Nothing ships without a human sign-off, and everything the AI does is logged with cost.
Every part of technical SEO, in one system
One connected system for organic growth: the dashboard, scans, rank tracking, competitor intelligence and the task queue all write to the same source of truth, so they never disagree.
Unified dashboard
Open tasks, scan history, high-priority content issues, keyword rankings and competitor movement in a single view.
SEO Diagnosis scans
Full-site or rule-based scans that surface why pages underperform, backed by a configurable backlinks module.
Task queue
Every recommendation becomes a categorized, prioritized task, tracked to verified done.
SERP tracker
Daily rank capture per URL, per keyword, per device, so you see when a position moves and why.
Competitor intelligence
Automated competitor page audits and side-by-side comparisons on average position, domain rating and page coverage.
Content ideas from real questions
Reddit and community threads are mined for the real questions your customers ask, so content answers demand.
The hard parts your team never has to touch
Serverless can't run scrapers
Vercel can't run Python or headless browsers. So the web app just enqueues work, and a separate Python runner claims jobs, retries failures and recovers crashes.
Competitors don't want to be scraped
Competitor pages sit behind bot protection. The scraper tries stealth Playwright first, falls back to a proxy when blocked, and checks its own selectors so silent failures get caught.
LLMs confidently invent audits
Early runs flagged "too long" titles that weren't. Every AI fix is now validated against the real page, with counts checked and false flags removed before a human sees it.
Data drift poisons diagnosis
One stale keyword field once had an English gold-bar page diagnosed against foreign-language retailers. Drift and pollution detectors now gate every diagnosis before it starts.
Runs on its own. Ships nothing without you.
The system runs your SEO operation unattended, but a person approves every change before it ships. A separate engine handles the scraping, data and AI work, with every call logged and costed.
| Component | What it does |
|---|---|
| Scan pipeline | Playwright scrape → URL-tree clustering → AI keyword extraction → cannibalization detection across clusters → competitor-aware content rewrites. |
| Diagnosis agents | Four specialist agents (content quality, trust & authority, structure & schema, links & media) score each page; a Master agent reconciles evidence into active blockers; a Remediator writes the prioritized task brief. |
| Daily rank tracker | Real SERP positions per URL × keyword × device, captured daily and idempotently, feeding the "why did this move?" diagnosis view. |
| Task queue | Every recommendation becomes a task with owner role, priority and status; resolutions from the auto-resolver queue for human review before deploy. |
Signals in, approved tasks out
Data comes in from every SEO source; approved, human-checked tasks come out the other end. The moving parts never touch each other directly, which is what keeps the whole system auditable, easy to recover, and cheap to run.
What it changed
Multi-agent AI with the guardrails built in
Claude does the heavy reasoning, with a Gemini provider path for comparison and fallback. Everything the models emit is structured, validated, versioned and priced.
The AI can't invent fixes
Every recommendation is validated against the real scraped page, so a whole class of confident-but-wrong "fix this" flags never reaches a reviewer.
Every call logged with its cost
Each model call records its prompt version, tokens in and out, and a per-call cost, so every diagnosis traces back to the data and the spend that produced it.
Provider-agnostic with fallback
Claude does the heavy reasoning with a Gemini path for comparison and fallback, so drafting survives a provider outage instead of stopping.
Want SEO that ships instead of reports?
AISEO Manager was designed, built and is operated by Realisier Labs for BOLD Precious Metals. If your organic channel runs on spreadsheets and guesswork, let's talk.
Talk to Sachin