Keyword Universe

A keyword universe is the complete, structured database of queries a brand cares about: its own ranking terms, competitor terms, questions, autocomplete suggestions and trend data, combined in one place and labelled so people can analyse it. In MLforSEO’s approach to semantic keyword research, the keyword universe is the backbone of the project. It is built around search intent, entities and query paths, and its value lies in how easily other teams can use it.

How it works

Building a keyword universe happens in three stages.

• Collect. Pull queries from several sources: your own Search Console data (ideally through the Search Console API, the Search Analytics for Sheets add-on or a BigQuery export), keywords for both business competitors and organic competitors (bloggers, affiliates, YouTubers), People Also Ask and other question data, autocomplete suggestions and trends. Merge and de-duplicate. As the course puts it, data in is data out: a thin starting list limits every later analysis.
• Label. Add the dimensions that turn a list into a dataset: branded versus non-branded, search intent, entities and EAV combinations, topics or clusters, personas, funnel stage, content format and SERP features. Most labels can be rule-based (a REGEXMATCH formula in Google Sheets for branded terms); others use machine learning, such as entity extraction or topic modelling.
• Deliver. Package the universe so teams can use it without the analyst: layered dashboards (for example in Looker Studio), documentation of sources and methods, and a short summary report of the highest-leverage opportunities.

How it’s used in SEO and marketing

• Content teams use it to decide what to write next and in which format.
• PPC teams use the intent and coverage labels to decide where to bid.
• Leadership reads the opportunity summary; product teams look for persona signals and unmet needs.
• SEO teams use it for keyword mapping, site architecture, internal linking and tracking how query paths and topics change over time.

Example

MLforSEO’s categorisation lesson builds a branded versus non-branded label in two ways: asking an LLM to extract brand names from a small keyword list and return them as a regex, or running the whole list through an entity analysis template and keeping the entities typed as organisations. Once intent, topic and entity labels are added, the most useful first view is usually a topic by intent matrix, which shows where coverage is dense, where it is thin and where the larger opportunities sit. As the MLforSEO post on categorisation notes, a 10,000-keyword universe with rich labels stays useful for a year or more, while an unlabelled one is read once and forgotten.

Limitations and things to watch

• Search Console does not show every query you rank for; anonymised queries are left out even in the BigQuery bulk export.
• Search volume is one signal among several. EAV combinations with near-zero volume can still have business value.
• Automated labels contain errors (sub-brands classed as non-branded, for example), so spot-check and refine them over time.
• Without documentation and a refresh plan, the universe depends on one analyst and quietly goes stale.

Learn it on MLforSEO

• What Does a Good Semantic Keyword Universe Actually Look Like? covers delivery and the common failure modes.
• How to Structure and Categorise Keyword Data for Semantic Analysis walks through the labelling layer.
• Transitioning from Traditional to Semantic Keyword Universe (Checklist) gives you the task list.
• The Semantic ML-enabled Keyword Research course builds a keyword universe from start to finish.