seo-data-analystUse when querying Google Search Console, Google Analytics 4, or Microsoft Clarity data for SEO analysis. Covers API endpoints, authentication, rate limits, and agent workflow recipes for content decay detection, cannibalization analysis, funnel optimization, and behavioral insights.
Independent GitHub snapshot · Automated classification · Original documentation governs
Data-driven SEO requires querying three platforms: GSC (search performance), GA4 (user behavior), and Clarity (behavioral/UX signals). This skill covers what an AI agent can access programmatically, rate limits, and practical workflow recipes.
This skill is the READ side — pulling and analysing existing data. To DESIGN and RUN experiments against it (A/B tests, sample-size and statistical-validity guards, cache-safe variation delivery, readout rules), see ecommerce-cro-experimentation.
| Platform | What It Provides | Auth Method | Rate Limit | Cost |
|---|---|---|---|---|
| GSC | Keywords, impressions, clicks, CTR, position, indexing status | Service account | 1,200 QPM / 50K rows/day | Free |
| GA4 | Sessions, engagement, conversions, funnels, e-commerce | Service account | 200K tokens/day | Free |
| Clarity | Heatmaps, rage clicks, dead clicks, scroll depth, session recordings | Bearer token | 10 requests/day | Free |
All three use service accounts or tokens for headless access — no browser required.
GOOGLE_APPLICATION_CREDENTIALS=/path/to/key.jsonAuthorization: Bearer TOKEN header| Tool | Package | Platform |
|---|---|---|
| mcp-gsc | github.com/AminForou/mcp-gsc | GSC |
| Clarity MCP | @microsoft/clarity-mcp-server | Clarity |
| SEO MCP | seomcp.dev | GSC + GA4 + 39 tools |
Data source: GSC Search Analytics API
Method: Compare two 28-day periods, per-page metrics.
Period A: last 28 days → dimensions: ["page"], metrics: clicks, impressions, position
Period B: previous 28 days → same query
Alert thresholds:
Content published < 3 months ago is too fresh to diagnose.
Data source: GSC Search Analytics API
Dimensions: ["query", "page"] → paginate to 50K rows
Group by query → filter where count(distinct page) > 1
Flag: queries where different pages have similar positions (both competing)
Resolution: See seo-keyword-strategist skill for decision framework.
Data source: GSC Search Analytics API
Dimensions: ["query", "page"]
Filter: exclude branded queries (regex)
Post-filter: position >= 4 AND position <= 20 AND impressions > 50
Sort: impressions descending
These are your highest-opportunity keywords — close to page 1, worth optimizing.
Data source: GA4 Data API (runFunnelReport — v1alpha)
GA4 tracks the e-commerce funnel:
view_item → add_to_cart → view_cart → begin_checkout → add_shipping_info → add_payment_info → purchase
Calculate drop-offs:
Break down by deviceCategory and sessionSource to find device/source-specific friction.
Data source: GA4 Data API
Dimensions: ["landingPage"]
Filter: pagePath BEGINS_WITH "/blog/"
Metrics: conversions, purchaseRevenue, sessions, engagementRate
Identifies which blog posts drive the most revenue — prioritize these for refresh.
Data source: Clarity API or MCP server
Dimension: URL
Metrics: Dead Click Count, Rage Click Count, Quickback Click, Excessive Scroll
Cross-reference: Pages with high frustration signals + declining GSC metrics = highest-priority fixes.
Join key: URL path (strip domain from GSC URLs to match GA4 pagePath).
| Analysis | GA4 Data | GSC Data | Insight |
|---|---|---|---|
| Content gap | High bounce rate | Low impressions | Content doesn't match intent |
| SEO wins | High conversion pages | Low position | Improve ranking for converting pages |
| CTR optimization | High engagement | Low CTR | Meta descriptions need work |
| Keyword intent | Conversion rates | Search queries | Match content to buyer intent |
| Platform | Limitation | Workaround |
|---|---|---|
| GSC | 50K rows/day per site per search type | BigQuery bulk export (free, unlimited) |
| GSC | ~47% of queries anonymized | BigQuery reveals slightly more |
| GSC | 2-4 day data delay | Use dataState: "all" for preliminary data |
| GSC | 16 months retention | BigQuery export (indefinite retention) |
| GSC | No AI Overview/AI Mode data in API | Monitor Google announcements |
| GA4 | Sampling above ~10M events | BigQuery export (unsampled) |
| GA4 | Segments not available via API | Use audiences instead |
| GA4 | Path exploration not in API | BigQuery for sequence analysis |
| GA4 | Predictive metrics need 1K+ qualifying users | May not activate for small stores |
| GA4 | 20-60% data loss in EU without advanced consent mode | Enable Consent Mode v2 advanced |
| Clarity | 10 requests/day per project | CSV export for deeper analysis |
| Clarity | 1,000 row max, no pagination | Filter to specific segments |
| Clarity | No heatmap image API | Manual dashboard review |
| Clarity | Copilot AI insights not in API | Raw data only via API/MCP |
Both GSC and GA4 can export to BigQuery for unlimited, unsampled data access.
| Feature | GSC Export | GA4 Export |
|---|---|---|
| Data | Site/URL impression tables | Raw event-level data |
| Granularity | Aggregated (like API) | Every event with all parameters |
| Sampling | None | None |
| Cost (small store) | ~$0/month (free tier) | ~$0-5/month |
| Setup | GSC Settings → Bulk data export | GA4 Admin → BigQuery Links |
For small e-commerce: BigQuery free tier (10 GB storage, 1 TB queries/month) is likely sufficient.
No official stable API. Options:
Use Google Trends for: seasonal trend identification, geographic demand analysis, comparing keyword popularity, identifying rising topics for content planning.
| Don't | Why |
|---|---|
| Rely solely on search volume from tools | 54% overestimated; use GSC actual impressions as proxy |
| Ignore anonymized queries (~47%) | Nearly half of click data has hidden query strings |
| Pull Clarity data more than needed | 10 requests/day — be strategic |
| Skip BigQuery for large sites | API has 50K row cap; BQ is unlimited and free |
| Use GA4 data retention default (2 months) | Change to 14 months; API not affected but Explorations are |
| Diagnose decay on content < 3 months old | Too fresh — wait for data to stabilize |