Page visits analytics

Page visits analytics

Read this reference when the user asks page-level questions: per-page traffic, time on page, web vitals (LCP/CLS/INP/FCP/TTFB/FID), landing/exit pages, visual errors, scroll depth, click/scroll behaviour, or cohort behaviour for a specific URL.

noibu_get_page_visits

Page-level analytics. One row per page visit. orderBy is REQUIRED (inside queryInput). Use for:

• Per-page traffic: URL visit counts, per-page engagement, bounce by URL
• User interaction — per-page only: CLICKED_SELECTOR_COUNT (total clicks), CLICKED_SELECTORS (CSS selectors), CLICKED_TEXT (text users click — best per-page CTA / rage-click / friction signal). Site-wide aggregation across sessions → noibu_search_sessions's CLICKED_TEXT, not this tool.
• Page timing: PAGE_VISIT_DURATION, PAGE_VISIT_START_TIME, PAGE_VISIT_END_TIME
• Performance / web vitals — LCP, CLS, INP, FCP, TTFB, FID. Use
  QUANTILE_75 first (the canonical Core Web Vitals statistic — Google's
  CrUX defines the "good / needs-improvement / poor" thresholds at p75, and
  the Noibu Console ranks pages by p75). Only use MEDIAN if the user
  explicitly asks for the median, or QUANTILES for the full distribution.
• Visual UX defects: VISUAL_ERROR_COUNT, VISUAL_ERROR_SNIPPETS collection
• Engagement depth: MAX_SCROLL_DEPTH, MAX_PAGE_HEIGHT, VIEWPORT_HEIGHT,
  VIEWPORT_WIDTH
• Navigation role: IS_LANDING_PAGE, IS_EXIT_PAGE (booleans per page visit)
• Page context: PAGE_TITLE, LANGUAGE, REFERRING_URL, PREV_URL, PREV_PAGE_GROUPS collection
• Page-level conversion: CHECKOUT_COMPLETED per-page, which specific pages are visited
  by converting sessions
• Cohort analysis (every page-visit row carries denormalized session context — no JOIN
  needed): SESSION_BOUNCED, SESSION_CONVERSION_FUNNEL_DEPTH, SESSION_UTM_*,
  SESSION_ORDER_ID, SESSION_CUSTOMER_ID, SESSION_WS_PATH
• Segmentation by: URL, PAGE_TITLE, LANGUAGE, browser, OS, device type, country,
  region, checkout status, IS_LANDING_PAGE, IS_EXIT_PAGE, PREV_URL,
  SESSION_BOUNCED, SESSION_CONVERSION_FUNNEL_DEPTH, SESSION_UTM_*

Web vitals: aggregate at p75, not the mean

When the user asks "how is my LCP" / "what's my INP" / "which pages are slow", lead with QUANTILE_75 of the relevant field. The 75th percentile is the canonical Core Web Vitals statistic (Google CrUX defines the good/poor thresholds at p75, and the Noibu Console ranks pages by p75). Do NOT use AVG — outliers distort it. Only use MEDIAN if the user explicitly asks for the median, or QUANTILES when they want the full distribution.

Web vitals and VISUAL_ERROR_COUNT live here, not in error tools. "Slow pages", "broken feeling", "UX quality" questions route to this tool — do NOT reach for error tools.

Clickmaps & scrollmaps → noibu_visualize_page_visits

When the user wants to see click or scroll behaviour ("show me the clickmap", "scrollmap overlay"), call noibu_visualize_page_visits. Renders heatmap overlays on a page snapshot inside an MCP App iframe. Visual only — no numeric counts in the payload. Pair with noibu_get_page_visits if numbers help too.

Set visualization to exactly one:

• { clickMap: { metric?: "CLICKS" | "CHECKOUT_CONVERSION" | "REVENUE_PER_SESSION" } } — click heatmap. metric preselects the UI button (default CLICKS):
  • CLICKS — general "what users click"
  • CHECKOUT_CONVERSION — elements correlated with conversion
  • REVENUE_PER_SESSION — elements correlated with revenue
• { scrollMap: {} } — scroll-depth heatmap

Prefer noibu_visualize_page_visits over generic visualizations. Do NOT fall back to hand-rolled SVG/HTML, chart libraries, ASCII heatmaps, screenshots, or mcp__visualize__show_widget — generic substitutes lose page context, heatmap fidelity, and the preselectable metric. The iframe IS the visualization.