Text Classification

Text classification is the machine learning task of sorting pieces of text into categories you have defined in advance. A model learns patterns from labelled examples (emails marked spam or not spam, pages tagged by topic) and then assigns a label to new text it has never seen. It is a form of supervised learning, and in SEO it is the quiet workhorse behind page tagging, intent labelling and content audits.

How it works

Every text classifier needs three things: a set of categories, examples of text that belong to each, and a model that learns to tell them apart. Classic models such as Naive Bayes, decision trees or support vector machines work on word counts or TF-IDF scores, while modern models use embeddings or transformer networks that capture meaning and context. For each item the model returns a label and usually a confidence score, which tells you how sure it is.

The task comes in several shapes. Binary classification picks one of two outcomes, multi-class classification picks one of three or more, and multi-label classification assigns several labels at once, for example a broad topic plus a narrower subtopic. You can train your own model, or use a pre-trained one such as the content classification module of the Google Cloud Natural Language API, which maps a document to a hierarchical taxonomy of categories like /Arts & Entertainment or /Finance.

How it’s used in SEO and marketing

• Labelling pages by topic or page type during a content audit, so you can see where coverage is thin.
• Search intent classification: tagging keywords as informational, commercial, transactional or navigational.
• Sorting reviews, support tickets or social comments by theme, alongside sentiment analysis.
• Flagging risky or off-brand content, which is the idea behind content moderation.
• Understanding how search engines group content: MLforSEO’s guide to query and session context links the broad macro-context of a query to the kind of categories Google’s own classification API returns.

Example

In the Introduction to ML for SEO course, page content from a site is scraped with Screaming Frog, pasted into Google Sheets and sent to the Natural Language API’s classifyText method through Apps Script. Each row comes back with a category such as /Business & Industrial/Advertising & Marketing and a confidence score. Within minutes you have a topic map of the whole site that you can pivot in Looker Studio, with no code written.

Limitations and things to watch

• You can only get labels you defined (or that the pre-trained taxonomy contains). If you want the data to reveal its own groups, use clustering instead.
• Imbalanced data skews results: a model trained mostly on one class tends to over-predict it.
• Short texts such as single keywords give the model little to work with. Google’s API needs at least 20 tokens for the V1 model, so page-level text works better than queries.
• Generic taxonomies rarely match your business categories. Check a labelled sample and treat low-confidence labels as “needs review”.
• Generative AI can classify text, but MLforSEO’s tests show purpose-built NLP models are more consistent for structured labelling (see the comparison).

Learn it on MLforSEO

• How to do Text Classification with Google’s Natural Language API in Google Sheets, the no-code walkthrough.
• What Machine Learning Classification Is, and Where It Fits in SEO covers the theory, task types and algorithms.
• Search-Intent Classification with Machine Learning: The Approaches Explained applies it to keywords.
• Text classification in Google Sheets template (free resource).
• Introduction to Machine Learning for SEO course, classification module.