Picture a Live Search Agent answering a reader's question about your sector. It requests your article, receives the HTML, lifts out the paragraphs it needs and is gone in well under a second. Your analytics tag was sitting in that HTML as an unexecuted line of text, so nothing was sent to Google and no session exists. Even if the bot had triggered the tag, GA4 carries a second layer of protection that would have removed it: an automatic exclusion list of known bots. Between them, these two mechanisms mean the AI systems reading your content are structurally absent from the dashboard you use to judge your audience.
How does Google Analytics actually count a visit?
GA4 is client-side analytics. A small piece of JavaScript runs in the visitor's browser after the page loads and sends an event back to Google with details of the visit. The model has one assumption built into it: the visitor is a browser that executes code. Every number in the interface, from sessions to engagement time to conversions, is derived from events that script sent.
The same is true of the rest of the browser-side stack. Your ad auction, consent layer, affiliate link logic and paywall checks are all JavaScript too, so a visitor that skips the scripts skips all of them at once.
Why do AI bots skip the tag?
Most AI bots are not browsers. They make an HTTP request, take the HTML in the response and parse it as text. The largest public study of this behaviour remains the joint analysis by Vercel and MERJ, published in December 2024, which examined real crawler traffic across Vercel's network. It found that none of the major AI crawlers it tested rendered JavaScript, a list that included GPTBot, ClaudeBot, PerplexityBot and Common Crawl's CCBot. Some of them did download JavaScript files, but they treated them as text and never executed them. Applebot and Google's crawler infrastructure are the notable exceptions, because both render pages.
Rendering is expensive and slow, and a Live Search Agent is retrieving pages while a user waits for an answer. The incentive runs firmly against adding a headless browser to every retrieval, so this is a design choice rather than a gap that is about to close.
One category of AI traffic does behave differently. Agentic browsers, such as Perplexity's Comet and OpenAI's Atlas, are real Chromium browsers driven by a model, so they execute your scripts. That traffic can reach GA4, but it is a minority of AI activity on most sites and it arrives looking like an ordinary visitor.
Does GA4 filter out bots even when they are recorded?
Yes, and this is the part most publishers miss. Google's documentation states that "traffic from known bots and spiders is automatically excluded" in GA4 properties, identified using Google's own research together with the IAB's International Spiders and Bots List. Google also states that you cannot disable the exclusion and cannot see how much traffic it removed.
Two consequences follow. A bot that does run your tag and identifies itself as a known bot is dropped before it reaches a report. And you have no way to audit what was dropped, so the dashboard cannot tell you whether the figure is zero or substantial. The filter is useful for keeping spam out of your conversion data. It is the wrong instrument for understanding how much automated demand your content is attracting.
How big is the gap between analytics and reality?
Network-level data shows automated traffic at a scale that client-side analytics cannot represent. Cloudflare Radar figures reported by PPC Land on 5 June 2026, covering the seven days to that date, put bots at 57.4 per cent of HTML requests across Cloudflare's network, with AI training crawlers alone accounting for 50.6 per cent. Those numbers are a snapshot of one network over one week and the split between training and search traffic comes from a third party's breakdown, so treat them as indicative rather than definitive. Earlier measurements were lower: Cloudflare Radar's 2025 figure of 4.2 per cent of HTML requests for AI bots, as cited by Clickport, shows how quickly the share has moved.
Clickport's April 2026 study of its own customer sites found a median of 20 per cent of traffic blocked as bot traffic, with individual sites ranging from 2.45 per cent to 82.22 per cent. The spread is the point. Your own exposure depends on your content, your audience and who is retrieving it, which is why a generic benchmark cannot substitute for measuring your own requests.
What does this blind spot cost a publisher?
Editorial and commercial decisions get made on incomplete evidence. Articles that AI systems retrieve heavily can look quiet in analytics, because the retrievals leave no trace there. A page with modest human traffic may be one of the most-read sources in your category for assistants, and nobody on the team knows.
There is a revenue consequence as well. You cannot price, negotiate over or monetise an audience you cannot see, and every retrieval that happens unseen is content used without a record. That is the difference between a publisher who can credibly say "we see N thousand retrievals a week from these agents" and one who has only an absence in a dashboard.
How do you actually see AI bot traffic?
You have to measure below the browser, where the raw request exists before any rendering.
Server log analysis is the traditional route. Parse access logs, group requests by user agent and compare against published bot lists. The limitations are that user-agent strings can be spoofed, logs are large and noisy, and attributing a request to a specific AI company takes ongoing maintenance.
CDN-edge classification is more reliable because it sits in the request path. It can check a claimed bot against its operator's published IP ranges and signatures, and increasingly against cryptographic verification such as Web Bot Auth, then label the request in real time. blankspace does this as its analytics layer, typing each agent as a Live Search Agent, training crawler, search crawler or scraper, and attributing it to its owner before JavaScript would have run. The analytics are free and go live on the first request.
What should you look for once you can see it?
Start with three questions. Which agents visit most, and how is that volume trending week to week? Which pages do they retrieve, and do those match the pages your human analytics rank highly? And which of those retrievals are Live Search Agents answering a user in real time rather than bulk training crawls?
The third question matters most commercially. A Live Search retrieval is a high-intent read happening at the moment someone is researching a decision. In blankspace's own publisher accounts, Live Search is frequently the largest single category of AI traffic, though the mix will differ by site and should be measured rather than assumed.
Frequently asked questions
Do any AI bots run JavaScript?
A few do. Applebot and Google's crawler infrastructure render pages, and agentic browsers such as Comet and Atlas are full Chromium browsers. The large majority of training crawlers and Live Search Agents fetch raw HTML without executing scripts, so you cannot rely on client-side analytics to capture them.
Can I see AI bots in GA4 with a filter or setting?
No. For most bots the events are never sent, because the tag never runs, and no configuration can recover a visit that did not fire it. For bots that do run the tag and identify themselves, GA4's known-bot exclusion removes them automatically, and Google says it cannot be turned off.
Does GA4's AI Assistant channel show AI bot traffic?
No. That channel classifies human visits that arrive by clicking through from an assistant, using the referrer. It does not record the agent reads that produced the answer, and a large share of AI clicks arrive with no referrer at all.
Will server logs show me all my AI traffic?
Logs record the requests, but attribution is hard because user-agent strings can be spoofed and raw logs are noisy. Verifying bots against operator IP ranges and signatures, typically at the CDN edge, gives a far more reliable picture.
Does seeing AI traffic let me monetise it?
Seeing it is the prerequisite. Once agent retrievals are visible and classified at the edge, eligible Live Search retrievals can be monetised through content-layer advertising, which is the only mechanism that reaches bots that never run JavaScript.