At some point someone in a meeting asks what all of this is worth. The honest answer has two parts: the clicks you can count, which are real and growing but small, and the influence you cannot count directly, because much of AI search influence arrives with no click at all. Someone researches in ChatGPT, decides, and shows up typing your URL or searching your brand name. This chapter covers what can be measured, the proxies for what cannot, and how to present both credibly.
Capture the referral traffic properly
AI referral traffic identifies itself by referrer hostname: chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, and the rest. GA4 does not break these out by default, so build a custom channel group: define an "AI Assistants" channel matching referrer patterns for the assistants you care about, and reorder the channel list so it sits above Referral, or the default grouping swallows it.Google Analytics Help, Custom channel groups, which walks through the regex-based channel definition and warns that channel order decides which rule captures the traffic. The pattern list needs updating as assistants launch and rename.
Then treat the number as a floor, because the identification has known blind spots. Traffic from native apps often carries no referrer header at all, so it lands in Direct: Cloudflare notes this for Claude's app and believes the same holds for other providers' apps.Cloudflare, the crawl-to-refer analysis on Radar, including the caveat that app-originated visits without a Referer header are invisible to referral counting. Once captured, read the segment like any acquisition channel: which pages AI visitors land on, and how they convert. Landing-page mix is diagnostic here, because it shows which of your pages the answers are actually citing.
The dark funnel is the larger share
The click-shaped evidence understates the channel, and the best public data says so directly. Pew Research tracked the browsing of 900 US adults: on Google search pages with an AI summary, users clicked a traditional result link in 8% of visits, against 15% without a summary, and clicked a source cited inside the summary in 1% of visits. Sessions also ended outright more often after a summary page, 26% against 16%.Pew Research Center, Google users are less likely to click on links when an AI summary appears: browsing data from March 2025, 900 US adults. The reading matters: the influence happened, the click did not.
The platform-level economics point the same way. Cloudflare's crawl-to-refer ratios compare how many HTML pages a platform's crawlers request against how many visits it refers back; in one mid-2025 week, Anthropic's crawlers requested nearly 71,000 pages per referral, and the stated trend across AI platforms is more crawling per referral over time.The same Cloudflare Radar analysis; ratios for the June 19-26, 2025 period, ranging from Anthropic's 70,900:1 down to Mistral's 0.1:1, with the app-referrer caveat meaning true ratios are somewhat lower. Your content does work inside answers that never send you a session.
Proxies for the invisible part
Three proxies, none sufficient alone, together triangulate the dark funnel.
Branded search lift: users who decide in an AI conversation often arrive by searching your name, so plot branded query volume from Search Console against your visibility rates from Building a Visibility Tracking System and look for correlated movement. Direct-traffic cohorts: a rise in new direct visitors to deep pages, rather than the homepage, is the app-traffic blind spot showing up under another name. Self-reported attribution: a "how did you hear about us" field with an explicit AI option on signup and checkout forms. Self-report skews toward the memorable touchpoint, so treat it as directional, and as one of the few signals that names the platform.
Model the channel's value as a range built from those independent signals, stated with its assumptions. A range built from three imperfect signals is more defensible than a single number built from one, and executives who lived through SEO's "(not provided)" era have seen this movie: the channel was real, the tracking was partial, and the teams that presented honest ranges kept their budgets longer than the teams that presented false precision.
The board slide
Present it in three layers, labeled as what they are. Measured: AI-referred sessions and their conversion rate, from the channel group. Correlated: visibility rates alongside branded search and direct cohorts, with the version-marked time series from your tracking system. Estimated: the range for total influence, with the dark-funnel evidence above as the reason a range is the honest shape. Close the loop by connecting spend to the layer it moves: content and technical work to visibility rates, visibility to the measured and correlated layers. That chain is an assessment, not a proof, and saying so plainly is what makes the rest of the slide believable.
What this chapter cannot tell you is why visibility itself rose or fell. That diagnosis is Auditing and Diagnosing Visibility Problems.