SEO and science: why SEO is a grey area

Is SEO an exact science? No. Anyone who claims to know exactly how Google's algorithm works is overstating things. But SEO isn't guesswork either. In this article I explain why SEO is a grey area and how you can still make well-founded decisions.

Why nobody knows the algorithm exactly

Search engines keep the exact workings of their ranking systems secret to prevent manipulation: if everyone knew precisely which levers to pull, results would quickly fill up with pages that game the system rather than help the searcher. On top of that, these systems combine hundreds of signals and several layers of machine learning that are constantly adjusted. Even engineers at Google can't explain for every individual query exactly why page A ranks above page B.

What we do know

Fortunately, a lot is known. Google publishes guidelines for site owners, such as Search Essentials and its documentation on helpful, reliable content. Its staff regularly share clarifications, patents become public and the SEO community runs continuous research and experiments. In 2024, a leaked set of internal documentation also gave extra insight into the kinds of signals Google tracks.

SEO as a discipline of testing and measuring

Because certainty is lacking, a good SEO consultant works like a researcher. You form a hypothesis (‘this page underperforms because it doesn't answer the search intent’), make a change and measure the effect. Step by step you build up knowledge of what works for your website and your market. Experience across many different sites helps you choose the right hypotheses faster.

Correlation is not causation

A common mistake is drawing conclusions from relationships that merely coincide. Top-ranking sites often have many backlinks, but that doesn't mean every extra link automatically lifts your rankings. Perhaps those sites simply have better content, which naturally attracts links. When reading SEO studies, stay critical about the methodology and sample size.

What about AI search engines?

The same applies to AI assistants, even more so. Language models don't always give the same answer to the same question, and the sources they use vary. GEO therefore calls for repeated measurement across many prompts rather than a single snapshot. Here too, clear and trustworthy information is the best long-term strategy.