Last updated
AI referral traffic
Example referral log. Not this site's traffic. The charts below are invented. The “this tab” box reads only this browser.
Why you'd want to
The charts are a seeded example of clicks and crawlers, not this site's traffic. The this-tab box classifies this browser's referrer, utm_source, and User-Agent, and it does not send them.
How to use it
A human click shows up as a referrer host or a utm_source. A crawler shows up in the User-Agent. This box does not send either one anywhere.
Reading this tab…
These buttons are not this tab. They run the same rules on a made-up referrer, utm_source, or User-Agent.
CDN Logs lab draws invented edge rows, including GPTBot. Those rows are not this site's log. AI referrals lab is where stored page views are counted. A zero there is a stored zero.
^(chatgpt\.com|chat\.openai\.com|chatgpt|openai|perplexity\.ai|perplexity|gemini\.google\.com|gemini|copilot\.microsoft\.com|copilot|claude\.ai|claude|you\.com|you|poe\.com|poe|meta\.ai|phind\.com|phind)$
Tap a source to filter the landing pages and fade the other bands. Tap it again to clear. Days are D1–D14, not a calendar.
Each day's stack is 100% of that day's example clicks. Crawlers are not in this chart.
Example clicks from every source.
Solid bar: example clicks. Faded bar: example crawler fetches. Same invented names, different rows.
Example vendor setup is marked below. eVar and event numbers are examples. This lab does not call Adobe.
The AI referral traffic lab charts a seeded example log of clicks and crawlers, and it classifies this tab's referrer, utm_source, and User-Agent in the browser. The lab does not call Adobe.
Build one seeded log
exampleTraffic uses mulberry32 with seed 20261107. Each of 14 days gets click and crawler rows for five assistants. The same seed always returns the same rows.
Share clicks by day
clickShareByDay sums click counts per assistant for each day. landings sums the selected kind by path. kindTotals splits clicks and crawlers.
Filter one source
filterSource keeps rows for the assistant button that is on. A second click clears the filter. The charts that take scoped rows follow that filter.
Classify this tab locally
readLiveVisit reads document.referrer, utm_source, and the User-Agent. classifyReferral labels a click. matchCrawler labels GPTBot, ClaudeBot, or PerplexityBot. The section is data-track-ignore, and this file does not call trackLabEvent.
The seeded loop. Click rows and crawler rows are separate.
export function exampleTraffic(): TrafficRow[] {
if (cached) return cached
const rand = mulberry32(20261107)
const rows: TrafficRow[] = []
const clickBase = [8, 6, 4, 5, 3] as const
const crawlBase = [5, 4, 3, 2, 2] as const
for (let day = 0; day < EXAMPLE_DAYS; day += 1) {
AI_SOURCES.forEach((source, index) => {
const wave = 1 + 0.35 * Math.sin((day + index) / 3)
const clicks = Math.max(0, Math.round((clickBase[index] ?? 2) * wave + rand() * 2 - 0.4))
const crawlers = Math.max(0, Math.round((crawlBase[index] ?? 1) * (0.8 + rand() * 0.4)))
if (clicks > 0) {
rows.push({ day, source: source.id, path: pick(rand, CLICK_PAGES), kind: "click", count: clicks })
}
if (crawlers > 0) {
rows.push({ day, source: source.id, path: pick(rand, CRAWL_PAGES), kind: "crawler", count: crawlers })This tab's referrer, utm_source, and User-Agent. The values stay in the browser.
function readLiveVisit(): LiveVisit {
const params = new URLSearchParams(window.location.search)
const next = {
referrer: document.referrer,
utm: params.get("utm_source") || "",
agent: navigator.userAgent,
}
if (
liveCache &&
liveCache.referrer === next.referrer &&
liveCache.utm === next.utm &&
liveCache.agent === next.agent
) {
return liveCache
}
liveCache = next
return nextThis lab does not queue its own event. The Google block shows an example of how you would track the interaction. Adobe is an example too.
This lab does not queue an analytics event of its own. After Accept, the site still queues virtual_page_view for the page, with page_path and page_title. Clicks inside a data-track-ignore region are not recorded. The push below is an example, not a call this page makes. This site does not call gtag('set') or Adobe.
Example only. This page does not push this object.
// Example only. This lab does not queue this event.
dataLayer.push({
event: "ai_click_share_example",
source: "chatgpt",
kind: "click",
})Example implementation. This site does not call Adobe. eVar45, prop45, and event45 are example variable numbers. In Launch, a data element reads the value and a rule sends it when the value changes. In the report suite, enable eVar45 with Most Recent allocation and expire it on the hit. A processing rule is unnecessary if the data element maps straight to that eVar.
Example alloy payload. _example is a placeholder schema, not a real tenant.
alloy("sendEvent", {
xdm: {
eventType: "web.webinteraction.linkClicks",
_example: {
aiSource: "chatgpt",
},
},
})Example hit. eVar45 is hit-scoped in this example. prop45 dies at the same hit.
s.eVar45 = "chatgpt"
s.prop45 = "chatgpt"
s.events = "event45"
s.linkTrackVars = "eVar45,prop45,events"
s.linkTrackEvents = "event45"
s.tl(true, "o", "ai_click_share_example")Each lab field, the GA4 name, and the example Adobe variable. Scope is the part that usually gets lost in the translation.
Assistant in the example
GA4. Not sent. The chart source is the seeded id, such as chatgpt.
Adobe. Example eVar45, expire on the hit.
Scope. Hit, if you tracked a filter click. The example source is not a user property.
Click or crawler
GA4. Not sent. kind is click or crawler on the example row.
Adobe. Example prop45 on the hit.
Scope. Hit. A visit-scoped kind would mix a crawler row into the rest of the visit.
This tab's referrer
GA4. Not sent. classifyReferral reads it for the box on the page.
Adobe. Example eVar45 if you sent the host. This lab does not.
Scope. Hit. The referrer of this view is not a user property.
User-Agent match
GA4. Not sent. matchCrawler returns a crawler id or null.
Adobe. Example prop45. Not sent.
Scope. Event, if you tracked the classification. It is not a visit trait.
Ask analytics in plain English. Preset questions become SELECT statements over the log. AI-assisted matching is optional, and the SQL stays on the page.
How search engines see this page: title, canonical, structured data, and the HTML before JavaScript, for any URL on this site. Free to fetch and compare.
The body of one page view, the gate it has to pass, and the record that comes back.