Your international product content is attracting more automated requests. The infrastructure chart rises sharply while the analytics session chart barely moves. Subtracting one from the other will not establish how many potential customers actually visited.
For overseas brands managing content, suppliers and sales across China and other markets, the practical question is where to investigate next. A measurement problem, an infrastructure problem and a content problem call for different work.
Start with what each system measures
AWS’s AI traffic dashboard article, published May 5, 2026 and updated October 13, describes CloudWatch metrics collected as WAF evaluates requests, including path-level request counts. It describes AWS’s observation layer, not your brand’s installed configuration or traffic mix.
Google’s current Known bot-traffic exclusion documentation says Analytics properties automatically exclude known bots and spiders. That exclusion cannot currently be disabled, and the excluded quantity is not available to view. This is not a guarantee that every automated visit is recognized or every remaining record is human.
ONCE’s inference is straightforward: reconcile definitions before interpreting a gap. This is an evergreen reporting method, not a claim that these established tools launched today. No vendor’s industry bot percentage is applied to your website.
A request is not a session, and a session is not a customer
Opening a page can involve requests for the document, images and other resources. An automated process can also request the same path repeatedly. Request volume can help explain load and access patterns without being a count of people.
A GA4 session belongs to its analytics collection and processing model. It is not a server counter incremented once per HTTP request. Without compatible definitions and traceable records, do not assume every fetch should produce a session or every unmatched request represents a lost visitor.
Qualified enquiries need another definition. Submissions may be duplicated, irrelevant or preliminary. Business review establishes whether an enquiry meets your criteria; changing the label on an analytics column does not turn it into a customer count.
ONCE’s three-part reconciliation card
For each chart used in a content-budget decision, document the unit, collection point and scope. This proposed review practice is not a deployed ONCE analytics product.
Keep the unit visible
Preserve the actual metric name: requests, events, sessions or deduplicated qualified enquiries. A ratio requires compatible units. Identified AI requests divided by all requests in the same defined log scope can describe that request mix. It does not establish the proportion of visitors who are bots.
Name the collection point
Specify CDN, WAF, origin server, web analytics or business records. An edge-served request and an origin request are observed at different places. Browser-side analytics is another collection process. Do not force all their totals to match and call the difference missing customers.
Record scope and filtering
Align time zones, dates, hostnames, page paths and language-site coverage. Note whether static resources, unsuccessful responses, internal tests and recognized automation are included. If rules or identification coverage change, flag that before presenting a period-over-period comparison.
When these conditions are not aligned, keep the charts separate. An unknown definition should remain unknown rather than becoming an estimated human-traffic percentage.
Work through a hypothetical mismatch
Suppose a log contains 10,000 server requests, of which its rules classify 8,000 as AI. A separate report contains 1,000 GA4 sessions. These are illustrative figures, not ONCE or client data.
Subtracting 8,000 from 1,000 mixes requests with sessions and ignores different collection and filtering processes. The gap between 10,000 and 1,000 also does not reveal the number Google excluded: its documentation does not provide that count for such a reconciliation.
After verifying the log’s scope and classification, you could report that its rules classified 80% of those requests as AI, with identification limits stated. Do not rewrite that as “80% of our audience is bots.” Report the 1,000 analytics sessions independently; adding the AI requests would not create a larger reach figure.
Route the investigation before commissioning a fix
If repeated automated requests concentrate on large assets or a few paths, ask the technical owner to examine load, caching, responses and identification. Any access restriction is a separate business decision. A bot classification alone is not a reason to close every discovery route.
If analytics changes abruptly after a redesign, review collection, domain coverage, rules and timing before attributing it to creative quality. A new product video should not be reshot merely to fix a broken measurement setup. Equally, more crawling does not demonstrate that the new concept caused more demand.
If traffic records remain stable while qualified enquiries fall, inspect actual enquiries, page promises, product availability and the purchase path. The reports can narrow an investigation without establishing causality on their own.
Commission a report that supports a decision
Ask a content-operations partner for the reconciliation cards, reproducible summaries and a note of changed definitions. Routine reporting does not require copying private conversations, customer records or unrestricted raw logs. Use appropriately scoped samples and access controls where verification needs them.
ONCE’s earlier article on AI content-use reporting (Chinese) separates retrieval, context use and citation. The question here is different: whether counts from separate measurement systems can legitimately be compared or combined.
For a team commissioning product imagery in China while marketing internationally, keep production and analytics responsibilities explicit. Content needs an approved version and destination; measurement needs documented collection conditions. Neither supplier should promise control over the other’s entire system.
For ONCE’s ongoing content planning, the durable record connects content changes, measurement definitions and subsequent business decisions. That can help distinguish a content update from a measurement repair or infrastructure intervention. It is a proposed service practice, not measured lead growth or margin improvement.
When the crawler and session curves diverge, put the three definitions beside the charts. They can inform one decision without being forced into one invented count of human traffic.
Sources checked October 29, 2026: the AWS article dated May 5 and updated October 13, and the undated Google Analytics help page linked above. Method and example are ONCE editorial analysis. The original AI cover is a conceptual business-analysis scene, not an actual dashboard or client case.
