Bot Watch — Neo4j

This dashboard pulls live GA4 engagement data and ranks channels, countries, and pages by how human they look. Click Sign in to begin.

Suite-wide bot filter

When on, other dashboards (Marketing, Blog, Anomaly Watch) will display an indicator and refer to the engaged-session-based view by default. Toggle saved to your browser.

Off

Engagement health ·

Engagement rate over time

Engaged sessions ÷ total sessions, daily. A sudden drop usually means a bot surge has arrived.

Trend

Suspect vs healthy

Share of sessions classified as engaged vs unengaged in the window.

Split

Channels ranked by engagement

Channels with low engagement rates are statistically bot-heavy. Investigate anything <30%.

—

Countries ranked by engagement

Geographic concentration with low engagement is a strong bot signal. Outliers are flagged.

—

Suspect pages · three bot patterns detected

Pages showing bot-flavoured patterns

Three signals are checked per page: Scraper (views/session ≥ 4 — agent visits many pages per session), Bot pings (sessions far outnumber views — pings that don't render the page), and Low engagement (less than 20% engagement on 1,000+ sessions — bounce floods). Hover any badge for the exact threshold.

—

Recommended actions · based on what's been detected

How bot detection works here

The detection model, plainly

GA4 already filters out known crawlers — Googlebot, Bingbot, and a list of about a thousand others. What gets through are the bots that don't identify themselves: scrapers using headless Chrome, automated SEO tools, content harvesters, and the occasional rogue script. Those leave behavioural fingerprints, which is what this dashboard reads.

The signals we use

Engagement rate — the share of sessions GA4 classifies as "engaged" (≥10 seconds active, or a conversion, or 2+ pageviews). Bots rarely qualify, so a slice of traffic with engagement rate below 30% is statistically bot-heavy. The site-wide average for genuine traffic typically sits between 55% and 70%.

Views per session — humans typically view 1–2 pages per session. Scrapers can hit 20 or 50, often in seconds. Anything above 4 is worth investigating.

Geographic concentration — a sudden spike from a single country that's not normally a major contributor is a classic VPN/proxy/datacenter pattern.

Channel outliers — bots often arrive as Direct/(none) (no UTM, no referrer) or via specific paid networks. Channels with engagement well below the property average are suspect.

What this doesn't catch

Sophisticated bots that simulate human behaviour — slow scrolling, randomised dwell times, full page parsing — can defeat behavioural detection. For those, you need either explicit user-agent allowlisting (which GA4 doesn't fully expose) or IP-level filtering (which requires server-side analytics). This is a strong heuristic, not a forensic tool.

Also: a low engagement rate isn't always bots. An ad campaign sending people to a landing page they immediately leave looks the same. Use this alongside Channel Mix and the per-page view to distinguish.

What "filter on" actually does

When the suite-wide filter is enabled, other dashboards will preferentially show engagedSessions instead of sessions as the headline metric, and display a small 🤖 filtered pill in the header so it's clear which view is active. The underlying GA4 data isn't modified — the filter just selects a different lens.