The machine layer
Server and CDN logs answer questions about automated consumption: which agents requested the site, which paths they preferred, how often they returned, and whether the origin served successful responses. This layer is valuable for infrastructure, access policy, content architecture, and licensing analysis. It does not tell you whether a human ever saw a recommendation based on that content.
A crawler-heavy site may have high machine visibility but little human value. Conversely, a site can receive valuable AI referrals through search indexes, browser integrations, partner data, or previously cached information without a neat one-to-one relationship with current crawler requests.
The human referral layer
Human referral measurement starts when a person arrives. Analytics can classify the referring host, campaign parameter, landing page, engaged session, signup, purchase, or other conversion. OpenAI's publisher guidance notes that ChatGPT referral links can include a utm_source=chatgpt.com parameter, providing one concrete way to distinguish that traffic when it is present.
The valuable metrics are not just sessions. Compare bounce or engagement, pages per session, conversion rate, revenue per visitor, and the content categories that receive the referrals. An AI-origin visitor who converts at three times the site average may justify significant investment even if total volume is initially small.
Connect the layers without claiming causality too early
A useful dashboard can place crawler trends and referral trends on the same timeline while still labeling them as different evidence. If OAI-SearchBot access is enabled on Monday and ChatGPT referrals increase over the next month, the operator has a correlation worth investigating. It is not proof that a specific crawl caused each referral.
The same discipline applies to content changes. If a guide receives frequent machine requests and later becomes a strong AI referral landing page, that is a commercially interesting pattern. The analytics system should make the relationship visible without turning it into an unsupported causal claim.
The commercial funnel
The strongest machine-web funnel is: discovery controls → observed automated access → human AI referral → on-site engagement → conversion → revenue. Not every business needs every stage, and paid machine access can create a second funnel that bypasses the human referral step entirely. What matters is that each stage has its own definition and evidence source.
Once those definitions are stable, a publisher can finally answer the question that crawler-volume charts cannot: which forms of AI access actually create value for the business?