Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Marketing Analytics › Troubleshooting

Sudden Traffic Spike in GA4

How to diagnose a sudden traffic spike in Google Analytics: bots and crawlers, referral and event spam, test traffic, duplicate tags or genuine demand.
Troubleshootingintermediate6 min readUpdated

First, read the spike's shape

Build a free-form exploration covering the spike and the weeks before it, with date and hour as rows and sessions, engaged sessions, engagement rate and key events as values. The explorations tutorial shows the technique; what matters here is the comparison. Then swap the rows for country and city, then landing page, then session source and medium, and finally hostname, each time comparing the spike days with the baseline. Check the sampling indicator before trusting any figure.

Five readings separate the causes. A tight hour pattern, one or two locations and near-zero engagement on a handful of pages is automation. Hostnames that are not yours, or referral sources that are unknown domains, are spam. A cluster from an office city around a release date is internal or test traffic. Events and page views rising far faster than users and sessions point at counting rather than visitors. A rise spread across your usual locations and sources, with normal engagement and some conversions, is demand. If the excess sits in the Direct channel on deep pages instead, the sibling page on direct traffic that is too high may fit better.

Diagnosis table

Diagnosis

SymptomCheckLikely causeFix
Sessions surge from few locations at fixed hours, with near-zero engagement on a few pagesExploration by hour, city, landing page and engagement rateAutomated traffic not caught by known-bot exclusionBlock at the site or network edge; exclude in analysis
Unknown referral domains or hostnames that are not yours; events with no page viewsHostname and session source in an explorationMeasurement Protocol or referral spam with fake eventsExclude by hostname and source in analysis; protect any secret used for server events
Spike on release days from the office location or test devicesInternal traffic rules, filter states and the location of the sessionsInternal, QA or load-test traffic not filteredDefine rules for every internal range and activate the filter
Events and views per session jump; users barely moveTag Assistant on a live page; page views per session by dayA second tag or duplicated events inflating countsRemove the duplicate installation or event
Rise across usual sources and places, with normal engagement and conversionsSession source, campaign and landing page against known press, social or campaign activityGenuine demand from press, social or a campaignKeep it; annotate the cause and tag future coverage

Fix, verify and prevent each cause

Bots that got past the exclusion

The automatic exclusion covers known bots and spiders, not every scraper, monitoring script or headless browser. Unrecognised automation arrives as sessions with almost no engagement, often as direct, on a narrow set of pages and from a narrow set of locations.

Fix: block the offending traffic where it enters, at the site's firewall or hosting layer, and keep it out of analysis with a saved comparison or segment; the marketing data quality guide lists bot traffic in its failure catalogue for this reason.

Verify: the cluster stops appearing in new data and engagement rate for the affected pages returns to normal.

Prevent: keep an engagement-rate line next to session volume on the main dashboard, since automation shows as volume rising while engagement falls.

Spam that never visited

Fake events can be sent straight to the collection endpoint with a property's measurement ID, and referral spam plants unknown domains in the source report. Neither involves a visit to your site, which is how to recognise them: hostnames you do not own, sources nobody recognises, events that arrive without page views.

Fix: exclude the spam in analysis with hostname and source conditions, and if the property accepts server-side events, keep that secret private and rotate it if exposed, following the catalogue in the marketing data quality guide.

Verify: the hostname breakdown of the cleaned view shows only your domains.

Prevent: include a hostname check in the weekly routine; a new hostname is the earliest sign.

Your own team, counted as visitors

Developers checking a release, testers walking checkout, a load test pointed at production, or a staging site sending to the live property all look like a burst of keen visitors from one place.

Fix: define internal traffic rules for every office, VPN and agency range, set the internal and developer filters to active once tested, and send staging to its own property, as the GA4 setup checklist describes.

Verify: a visit from the office is labelled internal and excluded once the filter is active.

Prevent: add load tests and release testing to the change calendar and check the filter states quarterly; filters are not retroactive, so annotate past spikes.

Counting twice

A theme update or plugin that adds its own Google tag, or a second configuration tag in the container, sends every page view twice. Users and sessions rise a little or not at all; views and events per session leap.

Fix: confirm with Tag Assistant and remove one installation, as the section on two installations in common GA4 mistakes explains.

Verify: page views per session on the next closed day return to their earlier level.

Prevent: keep a record of which route owns the analytics tag, and check it whenever a theme or plugin updates.

Real people, real interest

Press coverage, a social post that travels, a campaign launch or a seasonal moment produce spikes that look like your ordinary audience, only larger.

Fix: nothing to remove; find the cause with session source, campaign and landing page, and with the referral list for the coverage itself, using the explorations tutorial to compare the new visitors with the old.

Verify: engagement and key events behave plausibly and the source matches a known event.

Prevent: give the press and social teams tagged links in advance, using UTM values from the convention, so the next spike is labelled on arrival.

Prevention: anomaly alerts and filters

The cost of a spike is the decision someone makes before it is explained. Set custom insights or alerts in the property for unusual rises in sessions and falls in engagement, so the analyst hears first. Keep internal traffic rules current and the filters active. Review hostnames and unknown referrers weekly. Hold a release and campaign calendar the analyst can read, so known events explain known spikes. Keep a saved exploration with the shape breakdowns above, ready to run. Agree that no budget or headline target (KPI) is changed on a spike until its cause is written down. And annotate every explained spike, human or not, so year-on-year comparisons stay honest. The reports and dimensions guide explains the metric definitions worth checking when engagement figures look strange.

Illustrative example — not a real result

A spike from one city at four in the morning

A recipe site sees sessions rise 180% over 3 days. An exploration by hour shows most of the excess between 02:00 and 05:00, from 2 cities in a country where the site has few readers, landing on 6 old recipe pages, with an engagement rate of 3% against the site's usual 61%. Session source is direct, hostnames are the site's own, and key events are flat. The pattern says scraper, not readers. The hosting team blocks the address ranges at the firewall, the analyst builds a comparison excluding those cities for the 3 days, and the dashboard gains an engagement-rate line. The next week's traffic is back to its usual level and the spike is annotated in the reporting log.

Common mistakes

Celebrating a spike before reading its engagement. Filtering out genuine press traffic because it looked unusual. Leaving the internal filter in testing. Treating a duplicate tag as a traffic problem. Forgetting that new settings do not clean past data.

Frequently asked questions

Does Google Analytics filter bots automatically?

The help centre documents that traffic from known bots and spiders is excluded automatically. That covers crawlers on the recognised lists, not every script, scraper or headless browser, so a spike can still be automated even though the exclusion is working. Read the spike's shape before assuming the visits are human.

Can I remove spam sessions from reports that already include them?

Stored data is not deleted retroactively by new settings. You can keep the spam out of analysis with a segment or comparison that excludes it, by hostname, source or location, and build recurring reports on that view. Annotate the affected dates so the spike is not read as demand later.

How do I tell a duplicate tag from real growth?

A second tag inflates events per session and page views per user, while users and sessions move much less. Real growth moves users and sessions, with engagement and conversion behaving plausibly. Tag Assistant on a live page shows how many analytics tags fire for one page view, which settles it in minutes.

Terms used on this page

Where this page is used

Also related

Sources

  1. Google Analytics Help — Google Analytics Help
  2. [GA4] Get started with Explorations — Google Analytics Help

Related