Blog

You Can’t Track the Model. You Can Only Measure the Market: AI, GEO and the New Measurement

Written by Sayf Sharif | Jan 23, 2026 5:57:10 PM

Tracking “Organic” Is Evolving

For many years, measurement of organic efforts has been a difficult, political, noisy, and imperfect process, but it was at least conceptually possible. We might have argued about the specific attribution models, but at least we agreed on the basic premise: users do things, platforms record those things, and we did our best to reconstruct reality from the logs.

That premise no longer holds the full truth.

In our new AI-mediated world, the "stable surface" we once observed is shrinking. We are seeing a transition away from consistent impressions and durable journeys toward a more fragmented discovery layer. The core problem we faced has changed: tracking these systems in a 1:1 deterministic way is becoming elusive.

The Measurement Problem

Users no longer always traverse a sequence of observable steps. They may ask a question, get a synthesized answer, and then, maybe, they act later. If at all. There are fewer pages to rank, positions to track, or referrers to inspect. Marketing influence obviously still exists, and demand is still being created, but the observable layer is only part of the story.

We now live in a world where events are often hidden, but effects remain visible.

The old paradigm assumed that you could observe every user behavior and assign it credit. The new reality is a hybrid. You still observe the clicks you can, but you must now also intervene in a system, observe the aggregate outcomes, and attempt to infer causal impact.

Measurement is no longer just about what happened; it’s about what changed because we acted. It’s an integration of traditional analytics and causal inference within a complex system.

You Can’t Measure the Model (But You Can Benchmark the Output)

The instinctive response of some marketers to AI is to try and recreate the old observability. Trying to determine a “ranking” in ChatGPT or how often your brand appears in Perplexity. While these are useful benchmarks and leading indicators (often called "Share of Model" or "Brand Citation Volume") we must be careful not to treat them as the same "truth" as a Google SERP.

The models don’t have a stable ranking or a fixed memory across all users. Any “AI Visibility Metric” is a helpful proxy, but it can be theater if not tied to revenue. These systems are probabilistic and personalized. Even if perfect instrumentation existed, there’s no fixed state to observe. You cannot measure the inner workings of the model; you can only benchmark the responses it generates.

You Can Only Measure the Market

The only thing that ever actually mattered was the market, and that’s something you can still measure. AI systems aren’t removing causality, they’re just removing visibility. The most honest measurement strategy, therefore, is to look at real-world deltas. What changes in your business after you act?

So what does serious measurement now look like? It's a stack designed for both tracking and inference.

  • Market-Level Outcome Metrics: Everything starts with the basics. Revenue, pipeline, and LTV. These aren’t proxies. Everything else exists to explain these movements.
  • Branded Demand Signals: This is your new organic surface area. Branded search volume, direct traffic, and unattributed inbound leads. If AI is creating latent demand, it shows up here first.
  • Real Experiments (Scalable to Your Size): If you can’t define a control, you aren’t measuring.
  • Forecasting & Counterfactuals: The question stops being “what happened?” and becomes “what would have happened if we didn’t do this?” The output isn’t just a dashboard; it’s a decision range under uncertainty.

 

What Breaks If You Keep Measuring the Old Way

The real danger is that organizations might continue making decisions as if 100% visibility still exists. If you apply pure attribution logic to a system that isn't fully observable, you will systematically over-invest in what is still measurable (like bottom-of-funnel ads) and under-invest in what actually drives demand (brand, narrative, AI discovery).

This is self-sabotage masquerading as optimization. You will train the company to believe a false model of reality. This is where the CFO Blind Spot occurs. CFOs want deterministic certainty, but providing "false sight" leads to bad capital allocation. The path forward is to show the CFO that while the path is probabilistic, the outcome (the market delta) is measurable and real.

What You Can Actually Measure About AI

You can’t measure every AI journey, but you can measure your brand demand lift, competitive displacement, and revenue deltas by market. You can also use new tools to audit Model Citations to ensure your narrative is reaching the training sets. Forget about trying to measure the interface alone; focus on measuring reality. Observe the world that AI shapes.

Effects Without Events

AI didn’t kill marketing; it just changed the lens we need to use. Human marketers will continue to shape the market, but we no longer get to watch every single "click" happen. In this world of effects without events, the most honest form of measurement is a blend: track the signals you can, infer the impact you create, and measure the market above all else.