Google Analytics was built to measure the web of 2005, refined for the web of 2015, and rebuilt again for the cookie-deprecation challenges of 2022. It has not yet reckoned with the web of 2025, a web where a growing share of content consumption happens before, around, and without the browser visit.
The assumptions baked into GA are quietly breaking down, one percentage point at a time. The bottom line is this: if you want to know whether your brand is actually working in the AI era, stop looking only at traffic. Start watching brand behavior as a real signal. Are people searching for you by name? Are they arriving already knowing who you are? Are they converting faster than the traffic numbers would predict? These are signs of life that reflect actual revenue, and GA's dashboard was never designed to show them.
The Foundation Google Analytics Was Built On
Let me be direct about something. Google Analytics was never really a universal measurement tool. It was a browser measurement tool. And for a long time, that was enough, because the browser was where everything happened.
Here is how it works. When someone visits your website, a JavaScript tag loads in their browser and fires a tracking event. It captures their session, their device, their location, where they came from, and what they did. The whole system rests on three assumptions:
- Traffic arrives via a browser
- A human is sitting behind that browser
- The human actually visits your website
In 2010, these assumptions held for nearly every visit. In 2025, they are breaking down faster than most marketers realize.
Assumption #1: Traffic Runs Through a Browser
How the Tracking Script Works
The GA tracking code, whether you are running the legacy analytics.js, gtag.js, or the current GA4 setup, executes inside the browser's JavaScript engine. It reads cookies, fingerprints the device, listens to page events, and sends data back to Google. It is elegant and effective when a browser is involved.
The problem is that AI traffic often skips the browser entirely.
The Channels GA Cannot See
Think about how AI systems actually interact with your content.
Command line and API calls. When a developer or an AI pipeline fetches your content via curl, wget, or httpx, no JavaScript runs. The request hits your server, the HTML comes back, and GA's script never fires. That visit is completely invisible.
bash
# This request leaves zero trace in Google Analytics
curl -A "GPTBot/1.0" https://yoursite.com/articleHeadless browsers. Tools like Puppeteer and Playwright can technically render JavaScript, but they usually block third-party analytics scripts or run with cookies disabled. GA might log something, but it loses all meaningful attribution data.
AI retrieval pipelines. A RAG (retrieval-augmented generation) system might pull your sitemap, parse your RSS feed, or call a REST endpoint directly. None of these produce a pageview in GA. None of them were ever meant to.
LLM crawlers. OpenAI's GPTBot, Anthropic's ClaudeBot, Perplexity's bots, Google's Gemini crawlers. These are fetching your content at scale, processing it, and making it available to millions of users. GA records zero of this. It does not even know it happened.
The result is that a growing share of your content's actual reach generates no signal in Google Analytics whatsoever.
Assumption #2: Visitors Have Persistent Identities
How GA Thinks About Users
GA4 was built around the concept of a unified user identity. Rather than just counting sessions, it tries to stitch together a single user journey across devices and visits using a few key signals:
- Client ID, a cookie stored in the browser
- User ID, an identifier you pass when someone logs into your site
- Google Signals, cross-device tracking via Google account login
This lets GA answer questions like: how many unique users came back this month, what is the typical path before someone converts, and which audience segments are most valuable?
These are genuinely useful questions. The problem is that AI agents do not answer them.
AI Agents Are Anonymous by Design
An AI agent querying your content has no Google account. It does not accept cookies. It does not log in. It might operate from rotating IP addresses, use generic user-agent strings, or route through proxies.
And even if you try to work around this with UTM parameters, those only work when a human with a browser actually clicks a tagged link. When an AI system retrieves your content and surfaces it to a user, UTM parameters are irrelevant. The agent is not clicking through. There is no click.
So from GA's perspective, the entity consuming your content at scale is nobody. Or worse, it gets flagged as a bot and filtered out entirely, which means GA is actively removing evidence of your AI-era reach from your reports.
Assumption #3: The User Visits Your Website
The Old Funnel
For two decades, the web worked in a predictable sequence. Someone had a question, they searched, they clicked a link, they landed on your site, they read something, and you measured the visit. That was the funnel, and GA was designed to instrument it.
That funnel is breaking.
The Visit No Longer Happens
Think about your own behavior when you use ChatGPT, Gemini, or Perplexity. You ask a question. You get an answer. You may never see the source URL. You may never visit the site that the answer came from.
Your content might be the reason someone got a great answer today. GA has no idea.
Google AI Overviews. When Google generates an AI Overview for a search query, it reads your indexed content, synthesizes an answer, and the user moves on satisfied without ever clicking through to your domain. You provided the knowledge. GA recorded nothing.
Conversational AI. When someone asks ChatGPT about a topic your site covers, the model draws on content it was trained on or has retrieved from the web. Your article might be the direct source of the response. The user never sees your URL. No session, no pageview, no conversion event.
Featured snippets and knowledge panels. This is not even a new problem. Featured snippets have been extracting and displaying your content without generating a click for years. AI just accelerated it dramatically.
Voice and assistant interfaces. Smart speakers and AI assistants surface information verbally without rendering a web page at all. There is no browser, no cookie, no GA event.
In this new reality your content is being consumed actively and at scale, but the consumption is happening before and around your website. The indexing and processing occurred before the chat. By the time your content reaches the end user, it has been transformed and delivered through a channel that Google Analytics cannot see.
What Google Analytics Cannot Measure in the AI Era
Here is an honest look at the blind spots:
| Traffic Type | GA Coverage | Reality |
|---|---|---|
| Human with browser, organic search | Full | Declining share |
| Human with browser, direct or referral | Full | Stable |
| cURL / CLI fetch | None | Growing fast |
| Headless browser | Partial or filtered | Growing |
| LLM training crawl (GPTBot, ClaudeBot) | None | Massive and accelerating |
| RAG pipeline content retrieval | None | Growing with enterprise AI |
| AI-generated answer citing your content | None | Becoming a dominant channel |
| Zero-click AI Overview | None | Significant share of searches |
| Voice and assistant interface | None | Growing |
| API-based content syndication | None | Growing |
The share of traffic in the blind spot categories is not static. It grows every quarter.
The UTM Workaround and Where It Breaks
To be fair to Google Analytics, the UTM parameter system was designed exactly for this kind of problem. Append ?utm_source=newsletter&utm_medium=email to a URL and GA reads the attribution on arrival. It works well for email campaigns, social posts, paid media, and partner referrals.
But UTMs have limits that AI simply bypasses:
You cannot tag training data. There is no UTM you can append to content that an LLM crawler indexed six months ago.
AI agents do not click. When an AI retrieves a URL and surfaces its content to a user, the agent is not clicking through. UTM parameters become irrelevant.
The channel does not exist. GA's channel groupings include Organic Search, Direct, Referral, Email, and Paid Social. There is no bucket for "mentioned in an AI response" or "retrieved by a RAG pipeline." You can create custom groupings and build workarounds, but you are fighting the underlying architecture.
What Comes After This
I want to be clear: this is not a eulogy for analytics. Measurement matters more now than it ever did, precisely because the attribution problem has gotten harder. But the tools and the mental models need to catch up.
Here is where smart teams are looking:
Server-side analytics. Moving tracking from the browser to the server captures a broader range of requests, including some non-browser traffic. Still limited for agents that carry no cookies, but meaningfully better than client-side only.
Server log analysis. Raw server logs capture every request, human, bot, agent, and crawler. They do not give you the behavioral richness of GA, but they give you the ground truth of what actually hit your server.
AI crawler monitoring. A new category of tools is emerging to track AI bot activity specifically, which models are crawling your site, how often, and which content they prioritize.
Brand and citation monitoring. If your goal is understanding real reach, tools that track your content appearing in AI-generated outputs (Perplexity citations, ChatGPT references, AI Overview appearances) are becoming essential complements to traditional analytics.
The dark traffic signal. When someone types your URL directly because an AI mentioned you, it shows up in GA as direct traffic. Marketers are increasingly studying that channel as a proxy for AI-driven awareness.
Google Search Console. It provides crawl and impression data that GA never sees. Worth treating as a primary source rather than an afterthought.
The Percentage Question
The title frames this as a question of percentage, and I mean that literally.
GA is not obsolete for everyone today. If you run a local service business and your customers find you on Google Maps and visit on a mobile browser, GA is still doing its job. If your entire funnel lives in the browser, GA4 remains a solid tool.
But if you are a content publisher, a B2B company, a knowledge business, or anyone whose audience increasingly finds them through AI-mediated channels, the percentage of real activity that GA captures is shrinking. Not to zero, but toward a minority share of your true reach.
When the majority of your content's actual impact is happening in channels your analytics tool cannot see, you are not measuring your business anymore. You are measuring a fragment of it and making decisions accordingly.
We may already be at that point for certain categories of content and audience.
The Measurement Gap Is a Strategy Gap
The real issue here is not technical. It is strategic.
When you look at a GA dashboard, you are making decisions about content strategy, channel investment, budget allocation, and what is working and what is not. If GA is showing you a shrinking and distorted slice of reality, those decisions are being made on bad data.
Content that looks like it is performing might be underperforming in the AI landscape. Content that looks invisible in GA might be generating real influence through channels that leave no analytics footprint.
The question is not whether GA will eventually adapt. It probably will.
The question is what decisions you are making today, based on the percentage of reality it currently shows you.