Why Google Can't Find Everything for You — and Why It's Not Google's Fault

The article connects Gödel's 1931 incompleteness theorems with the fundamental limits of internet search engines, showing that no finite set of rules can fully capture reality, which itself has no axioms. The author compares three approaches to search — lexical (word matching), semantic (closeness of meaning), and search over structured data — and explains why each necessarily overlooks something relevant, not because of an implementation flaw, but because of the nature of any rule-based system. The conclusion: combining the approaches narrows the limits considerably but does not remove them, because no finite description can fully encompass what it describes.

What do a 1931 piece of mathematics and an internet search engine have in common

Imagine you're searching the internet for grandma's recipe for svíčková. You know it exists — someone published it years ago on a small blog. You type in your query. Google returns a thousand results, but the right one isn't among them. You try different words. You rephrase. Nothing. And yet the recipe is on the web. Your browser would display it if we knew the address. But the search engine didn't find it.

Intuitively, we attribute this to the imperfection of technology. A better algorithm, more data, a smarter system — and the problem disappears. Except it doesn't. That every set of rules has built-in boundaries was something mathematics knew long before the internet existed.

Gödel and the limits of every system of rules

In 1931, the logician Kurt Gödel — born in Brno, working in Vienna — published a proof that shook the foundations of mathematics. His incompleteness theorems say, at their core, one thing: no sufficiently powerful system of rules can be both complete and consistent at the same time.

What does that mean concretely? If you have a system of rules — axioms from which you derive conclusions — and this system is rich enough that basic arithmetic can be expressed within it, then there will always exist true statements within it that cannot be derived from those rules. Completeness and consistency are two properties we want to have simultaneously — but Gödel showed that this is impossible. Either the system fails to reveal some truths, or it also derives falsehoods.

The key phrase is system of rules. Gödel's theorem does not speak about reality — it speaks about our models of reality.

Reality doesn't need axioms

And here is a crucial distinction that is often overlooked. Reality is not a set of axioms. It has no derivation rules. It simply exists — in and of itself, independent of the tools we use to describe it.

An apple falls from a tree regardless of whether we have Newtonian mechanics, general relativity, or no theory at all. The recipe for svíčková exists on that blog regardless of whether Google finds it.

Gödel's limits therefore do not concern the world itself. They concern the tools with which we try to describe, search, and understand the world. Every model, every theory, every algorithm — all of these are tools with defined rules. And as such, they have limits beyond which they cannot see.

A search engine as a system of rules

An internet search engine is exactly such a tool. It has a defined way of deciding relevance — a scoring function, rules for word matching, weights for individual signals. When you enter a query, the search engine runs through its rules and returns the results that follow from them.

Gödel's theorem does not apply to a search engine directly — it doesn't deal with arithmetical proofs. But the principle that Gödel revealed holds in a more general form: every finite set of rules has things it cannot capture. And that is precisely why relevant documents necessarily exist that the search engine will not find. Not because of an implementation error — because of a fundamental property of every system that operates on rules.

A concrete example: classical search works on word matching. If you search for "how to fix a dripping faucet," it finds pages containing those words. But what about a page that describes the problem as "solving a water leak from a tap"? It says the same thing, but in different words. The rules for word matching overlook it — not because they are bad rules, but because no finite set of word-matching rules can cover all the ways a single idea can be expressed.

Semantic search: different rules, different limits

Modern search engines solve this problem with so-called semantic search. Instead of word matching, they convert both the query and the documents into a mathematical space, where "dripping faucet" and "water leak from a tap" lie close to each other because they mean a similar thing.

It looks like a solution. But it isn't — it's just a shift of the boundaries. Semantic search has its own limitations. It works splendidly for common queries, where it has learned from millions of examples what is "close" to what. But it fails where context changes meaning unpredictably, where the subject is a narrowly specialized field with its own jargon, or where relevance depends on something that cannot be captured by proximity in a mathematical space.

Again: not because the system was poorly designed. Because any single method of deciding — however sophisticated — has limits, beyond which lie things it is not enough for.

Structured data: the strictest rules, the sharpest limits

There is yet a third approach to search, which looks like the most reliable of all. Instead of searching free text, you can search structured data — in catalogs, registers, knowledge graphs like Wikidata. Here facts are stored as precise relationships between entities. Karlštejn Castle is an instance of a castle, it lies in the Beroun district, it was founded by Charles IV in 1348. Pure structure, no ambiguity of words.

At first glance, this looks like an escape from the dilemma. No guessing of relevance, no semantic fog — either the datum is in the structure, or it isn't. But it is precisely this precision that exposes the weaknesses of the rule-based approach even more sharply.

The first problem is the closed-world assumption. Structured data operates on the rule: what is not in them does not exist. If you ask "which Czech castles stand on a sandstone rock," the query returns only castles for which someone filled in this property. Kost Castle does stand on a sandstone promontory in the Bohemian Paradise — but if that is missing from the structure, then for the system it does not exist. Reality is richer than any data structure, and always will be.

The second problem is the rigidity of the schema. You can only pose a query about properties that someone defined in advance. If there is no "geological bedrock" field, you cannot ask about it — regardless of how many castles are recorded. Here the system limits not only the answers, but also the questions that can be asked at all.

The third problem is the limits of classification. Structured data requires a taxonomy — the sorting of things into categories and subcategories. But reality has no natural taxonomy. Every classification is our model, and like every model it is incomplete. Things that fall between categories or intersect them in unexpected ways slip through this sieve.

Searching in structured data is thus doubly bound: by the schema, which delimits what can be asked, and by the data, which determine what can be found.

Three perspectives, one search

We therefore have three fundamentally different approaches to search. Lexical search looks for word matches — it fails when people say the same thing in different words. Semantic search looks for proximity of meaning — it fails when context changes meaning or when precise terminology is involved. Structured search looks in catalogs and graphs — it fails where data are missing or where the schema does not allow the question to be posed at all.

Each of them is a different tool with different limitations. And that is precisely why it makes sense to combine them: the shortcomings of one approach are compensated for by the strength of another. Lexical search reliably finds the exact term that semantic search would blur. Semantic search overcomes the language barrier that lexical search trips over. Structured search answers the precisely formulated question that both textual approaches can only guess at.

Not perfectly — the resulting combination is still a finite tool and has its own limits. But those limits are significantly narrower than those of any single approach.

This is not an engineering compromise. It is a consequence of how finite models relate to the reality they are trying to capture. Both mathematics and information theory show the same thing: no finite description can fully encompass what it describes.

What follows from this

Next time a search engine fails to find what you're looking for, know this: it's not just a technical shortcoming that the next version of the algorithm will fix. It is a property of every tool built on rules — whether it searches for words, meanings, or relationships — that tries to capture a world that does not fit into any finite description.

Kurt Gödel showed this for mathematics. For internet search the same holds in an analogous sense — with the only difference that, instead of arithmetical truths, we are searching for recipes for svíčková.

The concept, structure, and editorial line of the article are the work of the author, who prepared the content sketch, established the key theses, and directed the entire creative process. Generative AI (Claude, Anthropic) was used as a tool for research, fact-checking, and fleshing out the author's outline.

The author edited the outputs continuously, verified the key findings, and approved the final wording. No part of the text was published without human review. All factual data were verified against the publicly available sources cited in the text.

The procedure complies with the requirements of Art. 50 of EU Regulation 2024/1689 (AI Act) on the transparency of AI-generated content. #poweredByAI