The client asks how many visits the website received. GA4 gives one answer, Cloudflare gives a larger one, and a cookieless counter lands somewhere between them. Picking the middle number feels reasonable. It is also a way to invent a metric that none of the systems actually measured.
Choose the question before choosing the counter. A reach report, a conversion analysis and an infrastructure report do not need one universal traffic total. They need clear definitions and a stable basis for comparison.
First, name the Cloudflare report
Cloudflare is not a single counting method. Its Web Analytics documentation distinguishes a browser JavaScript beacon from edge analytics collected for proxied traffic. The beacon can be blocked. A request observed at the edge is a different measurement from a browser sending an analytics event.
Cloudflare's explanation of apparent discrepancies includes bots, plugins and API activity that may not load a complete page. It also distinguishes an Overview containing subrequests from a Traffic view excluding those subrequests. That filter is not a certificate that every request represents a person.
Before explaining a large number away as bots, write down the exact product, dashboard and metric. Then check the hostname, reporting window and filters. Calling everything Cloudflare traffic obscures the very distinction the comparison needs.
GA4 is a defensible baseline, not a census
Keeping GA4 as the headline source can be sensible. A client may already use it to connect acquisition with conversions, and changing the reporting source every month destroys comparability. A consistently defined number is more useful than a different flattering number at every review.
But describe what the report contains. Do not label GA4 the real-people count and dismiss every additional visit elsewhere as noise.
Google documents behavioral modeling for Consent Mode , which can combine observed and estimated behavior in eligible reports. It requires the appropriate implementation and enough suitable data; meeting the published prerequisites alone does not guarantee eligibility. The reporting identity and data-quality indicator matter. GA4 does not universally mean only visits that accepted cookies, nor does every property receive estimates for missing activity.
That qualification changes the reporting conversation. State the chosen metric, source and modeling status. If the collection setup changes, annotate the break instead of presenting the new total as marketing growth.
A cookieless count still needs a definition
A separate cookieless counter can be a useful reach measure, especially when the team wants to examine activity outside its GA4 reporting boundary. It does not need to replace the conversion report to earn a place beside it.
Nor does cookieless mean a verified headcount. Umami's current metric documentation describes an anonymous identifier built from inputs including the website, IP address, browser information and a rotating salt. Identifiers are not automatically linked across rotation periods, and configuration can alter the behavior. Check the installed version and settings rather than applying today's documentation blindly to an older report.
Use the counter's own visits or pageviews label. Do not silently rename either one people. Keep conversion analysis tied to a clearly stated measurement population rather than dividing one system's conversions by another system's visits and treating the result as directly comparable.
Explain the gap without pretending to solve it
The dashboards alone cannot establish how much of a discrepancy comes from consent, blocking, automated activity or different counting rules. Start with a controlled check: repeat the same small journey with consent granted and declined, then inspect which requests and events actually arrive. Test a relevant browser with blocking enabled as a separate condition.
Match the date range and time zone before comparing totals. Search clicks, sessions and pageviews should retain their own labels; putting them in adjacent columns does not make them interchangeable. Do not subtract two totals and call the remainder lost humans or bots without establishing how the datasets relate.
For the next client report, choose one named headline metric for the agreed business question. Keep its source and definition stable. Put the other measures alongside it only when they answer a useful additional question, with the differences explained.
A range can show that systems disagree. It cannot turn incompatible measures into a confidence interval. The honest answer is a defined count with a known boundary, not a compromise number that makes the slide easier to present.