Published: 11 June 2026 by Dan Williams
My thoughts started wandering, as they do, while reading an article by Duane Forrester, who spent years working on search at Microsoft and Bing.
The article itself was about semantic alignment, embeddings and vector search. The central argument was simple enough: whilst the ability to measure content alignment has improved dramatically, the measurements themselves remain approximations. Better approximations perhaps, but approximations nonetheless.
That idea stayed with me.
For years, much of the SEO world has been trying to understand search engines by observing their behaviour from the outside. Watching rankings, counting links, measuring keywords and tracking traffic. More recently, measurement has moved towards entities, embeddings, vector spaces, semantic alignment scores and all sorts of increasingly sophisticated ways to describe what might be happening beneath the surface.
The tools have changed. The measurements have improved. The discussions have become more technical. Yet I am not entirely sure the underlying problem has changed very much.
Search still seems to involve trying to understand systems that cannot be directly observed.
And somehow, that led me to think about Schrödinger’s Cat.
Not because Google is quantum mechanics. Thankfully. But because the famous thought experiment was not really about cats. It was about uncertainty, observation, what can be known directly, what must be inferred, and how easy it is to confuse a measurement with reality itself.
The more I thought about it, the more familiar it felt. Search engines, eBay, recommendation systems, artificial intelligence, and even marathon training all involve people trying to understand something happening inside a box they cannot fully open.
The Box
Imagine the box is Google, or eBay or any other complex system that people interact with every day.
The inputs are visible and the outputs are visible. Rankings, recommendations, purchases, traffic and all manner of other results can be observed. What remains out of reach, at least from the outside, is the ability to inspect every mechanism operating inside.
Sometimes research papers appear, very occasionally a leak emerges, and then people spend years running experiments and building theories. Those things can be useful, sometimes very useful, but most of what happens remains hidden from view.
Which is why so much attention ends up focused on the cat.
The Cat
The cat is whatever hidden machinery is operating inside the box.
Algorithms, ranking systems, retrieval engines, recommendation models, machine learning systems, and the collection of processes attempting to connect people with useful outcomes. For a long time, much of SEO has been an attempt to understand Google’s cat. Similarly, some eBay sellers spend time trying to understand eBay’s cat.
The names change and systems evolve, although the curiosity remains exactly the same.
What is that thing actually doing in there?
The honest answer is that nobody outside the box really knows. Observations are made and models are built. Occasionally showing glimpses of what is happening. But most of the time the cat is being inferred from its behaviour rather than observed directly.
And that is where things become interesting.
For a long time, the cat appeared to have a fairly simple diet: keywords, links and content. Much of the discussion centred around these things because they were the things that could be seen and measured most easily. If the words matched and enough links pointed at a page, good things often happened.
Of course, even then, the reality was probably more complicated than it seemed. But the model was useful.
Then things were seeming to change. Or perhaps, the understandings were changing.
The cat did not stop eating keywords, nor did it suddenly become vegetarian. It simply discovered other things it liked as well. And like most cats, it appears to have become a little fussy over time. The cat’s diet became broader, richer and more varied. Tasty morsels such as user behaviour, engagement, relationships between concepts, recommendation patterns, semantic similarity, and signals generated by people interacting with content were discovered by the cat.
And every time the cat discovered a new food source, somebody immediately started trying to measure these things.
The Shadows
The shadows are attempts to understand what the cat is doing inside the box: rankings, traffic graphs, click-through rates, keyword research, vector alignment scores, topical authority, watch counts, and every other chart, metric and dashboard used to infer what might be happening behind the scenes.
Some shadows are useful. Some are extraordinarily useful. The problem begins when those shadows start being treated as the thing itself.
A ranking is not always relevance and a click is not always satisfaction. A cholesterol reading is not the full picture of health. A six-mile run written on a training plan is not adaptation. A semantic alignment score is not the full understanding.
These things are measurements, and are attempts to describe reality rather than reality itself. They help build models, support decisions and navigate uncertainty, but they are not the thing being investigated.
And this, I think, is where people occasionally get themselves into trouble.
As the cat’s diet expanded, so did the shadows. Yesterday it was keyword density and today it might be vector alignment. Yesterday it was term frequency and today it might be cosine similarity. The measurements changed, but the temptation remained exactly the same: to believe the measurement itself was the thing being understood.
That was the part of Duane Forrester’s article that resonated most strongly with me.
A better measurement is still a measurement. A higher-resolution shadow is still a shadow.
That does not make it useless. Far from it. Some shadows reveal things that could not previously be seen. The mistake is believing they reveal everything.
Then Goodhart Walks Into The Room
There is another trap hidden inside all this.
Goodhart’s Law suggests that when a measure becomes a target, it ceases to be a good measure. The history of search is full of examples.
The moment people discovered keyword density, somebody started stuffing pages with keywords. The moment people discovered links, somebody started manufacturing links. The moment people discover a new shadow, somebody inevitably starts optimising for the shadow itself rather than the thing creating it.
The measurement becomes the objective. The model becomes the destination. Gradually, the relationship between the shadow and reality begins to break down.
The Training Plan On The Fridge
Running provides a useful example.
Every training plan is a model. Today the plan says six miles. Reality says poor sleep, a sore hamstring and thirty-plus degrees of heat. Or perhaps reality says fresh legs, excellent recovery and perfect conditions.
The number on the plan is useful and it represents a decision made using the best available information at the time. But it isn’t reality.
Treating the plan as reality is how people end up on a rainy day arguing with a piece of paper attached to their fridge.
The goal was never six miles, it was adaptation and the plan was simply an attempt to get there.
The map is not the territory. The model is not the reality. And neither is the shadow.
Useful Models From Incomplete Information
The more I think about it, the more I feel expertise has less to do with certainty and more to do with building useful models.
Nobody possesses complete information. Not runners, doctors, search engineers or eBay sellers. Everyone is working with fragments: measurements, signals, observations, experience and the occasional educated guess.
The skill lies in building a model that remains useful despite those limitations.
Sometimes the model needs updating and might turn out to be wrong. Sometimes it falls apart completely. Which is not failure – it’s part of the process.
Perhaps expertise is simply the ability to build useful mental models from incomplete information, whilst remembering they are still models.
Closing Thoughts
One of the strange consequences of modern technology is that there are now more measurements than ever before. More metrics, more signals, more dimensions, more dashboards.
Yet the need for judgement never disappears, it probably becomes more important.
The box remains closed. The cat remains hidden. And plenty of people are staring at shadows on the wall trying to work out what is happening inside.
Shadows are useful, measurements are useful and models are useful.
Just don’t mistake them for the cat!
— Dan Williams
Related Reading
You Can Finally Measure Content Alignment. That’s the Dangerous Part — Duane Forrester
Staying In Your Lane: Simpler Than You Think, More Complex Than You Imagine
Beyond SEO: Visibility, Discovery and Selling on eBay