--- type: "Article" title: "TF-IDF | topical authority content" description: "What TF-IDF and the vector space model are, what Google has said about TF-IDF, and why tuning copy to a TF-IDF tool is a weak SEO tactic." resource: "https://www.laurelinlabs.com/search-systems/tf-idf" tags: ["tf-idf", "topical-authority-content", "vector-space-model"] generated: { by: "laurelin-site-build/2026-09-02", at: "2026-09-02T12:00:00+00:00" } status: "stable" stale_after: "2027-03-02T00:00:00+00:00" --- # TF-IDF TF-IDF (term frequency, inverse document frequency) is a classic information retrieval weighting that scores a word highly when it is frequent in one document but rare across a collection. It underpins the vector space model, where documents and queries are compared as vectors. Google has never confirmed using TF-IDF as a ranking factor, and John Mueller has called it a fairly old metric. For a site, the practical lesson is to cover a topic in its own natural vocabulary rather than tuning copy to a TF-IDF tool score. ## What is TF-IDF? TF-IDF (TF-IDF) is a weighting from IR that measures how important a word is to one document within a larger collection. Term frequency rewards words a document uses often; inverse document frequency discounts words that appear in most documents, so words like "the" carry almost no weight and distinctive words carry a lot. The idea of inverse document frequency is credited to Karen Spärck Jones in [a 1972 paper](https://en.wikipedia.org/wiki/Tf%E2%80%93idf), and Wikipedia notes that variations of the weighting were often used by search engines as a central tool for scoring relevance. ## How does TF-IDF work inside the vector space model? In the VSM, each document and each query becomes a vector with one dimension per term. TF-IDF supplies the value in each dimension, and relevance is estimated from how close the query vector is to each document vector, usually with cosine similarity. According to [Wikipedia](https://en.wikipedia.org/wiki/Vectorspacemodel), its first use was in the SMART Information Retrieval System. Two limits matter for SEO. First, the IDF part depends on statistics from the whole collection, which for a web search engine means its entire index, so no third party tool can reproduce it. Second, the model is a bag of words: it ignores word order and cannot tell that two different words mean the same thing. Later models such as BM25 fixed some of the scoring weaknesses, and embedding models addressed the vocabulary problem. - Term frequency: how often the term appears in this document. - Inverse document frequency: how rare the term is across all documents. - Score: the two multiplied, summed over the query terms in the simplest ranking function. ## What evidence is there that Google uses TF-IDF? This page is labelled **General theory**. Google has not confirmed TF-IDF as a production ranking signal. When asked directly in 2019, John Mueller described it as a metric from information retrieval, said Google uses many techniques to understand which words on a page are relevant, and advised against focusing on it ([Search Engine Journal, April 2019](https://www.searchenginejournal.com/google-tf-idf/304361/)). A few weeks later he added that learning how information retrieval algorithms work is still useful, even if Google does not use them one to one ([Search Engine Roundtable, May 2019](https://www.seroundtable.com/google-learning-information-retrieval-useful-27520.html)). That is the honest position: TF-IDF explains the logic of lexical relevance, but nothing public says Google computes it the way SEO tools do. ## What does TF-IDF mean for your pages? Use TF-IDF as a diagnostic, not a target. A term gap analysis against top ranking pages can reveal a subtopic you forgot to cover, which is useful. Adding words until a tool's score turns green is not, because the tool's corpus is a handful of competing pages, not Google's index, and because repetition beyond natural use drifts towards keyword stuffing. A falsifiable test: if a TF-IDF tuning pass is what moves a page, you should see the change after recrawl without any other edit. In our experience the pages that move are those where the gap analysis exposed a missing subtopic that was then written properly. That is an estimate from practice, not a controlled result. ## Which Laurelin audit checks test for TF-IDF? TF-IDF cannot be measured from Google's side, so these checks measure the on-page proxy: whether vocabulary serves the reader or a score. The new check [Page copy padded with terms to hit a TF-IDF tool score](https://www.laurelinlabs.com/audit/topical-content/378-page-copy-padded-with-terms-to-hit-a-tf-idf-tool-score) looks for artificial term inflation. Related checks: [Keyword stuffing and over-optimisation](https://www.laurelinlabs.com/audit/topical-content/167-keyword-stuffing-over-optimisation) catches the end state of score chasing. [Content not matching search intent](https://www.laurelinlabs.com/audit/topical-content/168-content-not-matching-search-intent) matters more than term weights. [Low information gain versus competitors](https://www.laurelinlabs.com/audit/topical-content/161-low-information-gain-vs-competitors) is the better use of a gap analysis. [Padded content](https://www.laurelinlabs.com/audit/topical-content/173-padded-content) often follows term insertion. [Keyword stuffed title](https://www.laurelinlabs.com/audit/topical-content/138-keyword-stuffed-title) is the same habit in the title element. ## What are the key dates for TF-IDF? - 1972: Karen Spärck Jones introduces inverse document frequency as term specificity ([source](https://en.wikipedia.org/wiki/Tf%E2%80%93idf)) - 2019-04-20: Search Engine Journal reports John Mueller calling TF-IDF a fairly old metric ([source](https://www.searchenginejournal.com/google-tf-idf/304361/)) - 2019-05-05: John Mueller says learning IR algorithms is useful even if not used one to one ([source](https://www.seroundtable.com/google-learning-information-retrieval-useful-27520.html)) ## FAQ ### Does Google use TF-IDF? Google has never confirmed it. John Mueller called it a fairly old metric in 2019 and said Google uses many information retrieval techniques to judge which words on a page are relevant. ### Are TF-IDF SEO tools useful? They can highlight subtopics that competing pages cover and yours does not. They cannot reproduce Google's statistics, so treat their scores as hints, not targets. ### What is the difference between TF-IDF and the vector space model? The vector space model is the framework that represents documents and queries as vectors. TF-IDF is the most common way of filling in the values of those vectors. Hub: [Topical Content](/topics/topical-content.md)