Reconcile your edge requests with your audience analytics
Trace one URL through edge logs, origin responses and browser events before treating mismatched traffic totals as lost readers.

Your edge dashboard shows a traffic spike while your audience report stays almost flat. Start by following one URL through the systems that count it. Each system can observe a different part of the same request path, so the comparison needs a shared interval and a stated unit.
Bot reports explain why automated access deserves attention. Imperva's 2026 report overview describes full-year 2025 activity. Applying that network observation to a local discrepancy requires local evidence. The investigation here concerns where a request becomes visible in your own reporting.
Pick a URL you can trace
Choose a public page with enough activity to investigate and a stable route through the application. Record the exact hostname and path. If redirects send several URLs to that page, retain the original request URL alongside the final destination. Otherwise, two reports may appear to disagree while counting different stages of the redirect.
Set the same UTC interval in each export. Check whether a dashboard groups activity by the visitor's local day. A request near midnight can move between reporting days when time zones differ. This is a concrete explanation to rule out before attributing the difference to a bot classifier.
Write down the unit in each export as well. One browser session can request the page repeatedly. An edge counter can include each attempt. A chart labelled users may apply its own identity and aggregation rules. Compare equivalent events wherever the tools expose them, then explain the remaining difference.
Cloudflare's Radar glossary illustrates the need to inspect scope: its default bot-versus-human view concerns HTML requests. Your own edge report may include additional resource types. Retain the chosen content-type filter so a later export can reproduce the comparison.
Identify what reached the origin
An edge service can answer a cached request without asking the origin for the page again. It can also block a request before the application runs. Both events can appear at the edge while leaving the origin without a corresponding page response. Inspect the cache outcome and enforcement action where your service records them.
Choose a sample of requests from the spike and trace the available request identifiers. Match the records through your own systems using the identifiers they actually propagate. Document a missing join explicitly. Combining records solely because their timestamps are close can attach the wrong origin response to an edge request.
Inspect response status as part of that trace. A successful document response has a different path from a redirect or a rejected request. Preserve the distribution of those outcomes in the working comparison. A drop in successful documents may deserve investigation even when the total edge count looks healthy.
Examine the browser event path
Your browser analytics script observes clients that execute it and successfully send events. Check the page's instrumentation using a browser session you control. Confirm which event is sent for the chosen URL and whether client-side navigation creates another event. This establishes the behaviour of your implementation under the conditions tested.
Google Analytics automatically excludes known bots and spiders. Its audience totals can therefore diverge from edge counts by design. Keep that documented filter in the explanation. It supplies one reason for a difference while other collection conditions still need inspection.
A consent setting can also change which events your own site sends. Review the actual implementation and the visitor state used for the test. Keep the investigation focused on observed collection behaviour. The goal is to understand the count while preserving the site's chosen consent behaviour.
Account for legitimate machine access
Check your own scheduled jobs before assigning unexplained traffic to an external actor. A health monitor or preview fetcher may access a public page repeatedly. Identify the job owner and compare its expected schedule with the observed requests. A known source gives the investigation a testable explanation for part of the volume.
Keep unidentified activity in a separate working group. Inspect its requested paths and the cost those requests impose. Repeated access to cached articles can have a different effect from repeated application searches. An endpoint-specific cost estimate helps decide whether a policy change is necessary.
If you test a filtering rule, record its start time and the traffic it affects. Preserve a way to reverse it if legitimate visitors lose access. Examine successful reader journeys after the change alongside the request reduction. A lower counter alone gives an incomplete account of the outcome.
For example, a scheduled monitor might request the article every minute while an editor checks the page twice. Identify the monitor through its configured schedule and the service's own records. Compare that expected workload with the observed interval. The example illustrates a testable explanation for a count difference, leaving any remaining requests to be investigated separately. If the expected count and the log diverge, inspect failed attempts and repeated requests before adjusting the monitoring schedule. Preserve the monitor's purpose when deciding whether to reduce its frequency.
Leave a reproducible explanation
The resulting report should show how the counts relate for the selected URL and interval. Explain which requests ended at the edge and which reached the application. Describe the browser events you could observe and the collection conditions that affect them. Retain the exports behind the calculation.
Repeat the same comparison after the next unexplained spike. A documented path through the reporting systems makes that investigation faster and exposes genuine changes in behaviour. Your traffic totals become easier to use when everyone can see where each count begins.