Semantic Keywords

Somewhere a plugin is grading your service page and docking points because "AC repair San Antonio" appears four times instead of seven. Meanwhile the page never explains what a capacitor does, never mentions a heat pump, and never answers the question every owner of a twenty-year-old unit asks first: fix it or replace it? The plugin is counting. Search engines are reading.

Semantic keywords are the words a knowledgeable page cannot avoid using. Write honestly about roof replacement and you will mention decking, underlayment, flashing, hail, the adjuster and the permit. Some of those are vocabulary. Some are entities, meaning named things: a shingle brand, a city, a building code. The rest are subtopics, the smaller questions inside the big one (cost, timeline, warranty, financing). Together they are evidence that a page covers its subject instead of repeating its title. The list comes out of keyword research, but it answers a different question: not what people type, but what a complete answer has to contain.

Google explains the mechanics in public. Its How Search Works pages, as of October 2026, describe language models that work out what a query means and a synonym system that can match a page to a search even when the exact words differ. The same pages call keyword matching the most basic relevance signal and then say it is not enough on its own: the systems also look for other relevant content beyond the repeated word. Google's SEO Starter Guide adds that keyword stuffing is against its spam policies. Say the main phrase clearly, once or twice, and spend the rest of your words on substance.

Which brings us to LSI keywords, the most durable folklore in the trade. Latent semantic indexing is real. It is a retrieval method patented in 1989 by Bell Communications Research that ran matrix math over a table of terms and documents to find documents sharing meaning without sharing words. "LSI keywords" is something else: a label stuck on lists of related words, usually by a tool that sells them. As of October 2026, Google's SEO Starter Guide does not mention LSI at all. The sensible idea underneath, cover the subject, survives. The name and the magic sprinkle do not.

So we do not hand you a word list. We take the query a page exists to win, read the pages already ranking for it and the questions in People Also Ask, and build a coverage map: the subtopics a full answer includes, the entities it names, the terms your customers use that your page never does. Then we lay it beside your page. The gaps tend to be plain once they share a sheet. Some belong on the page. Some deserve their own URL, which is where keyword clustering and the keyword map decide what goes where, so two pages never end up fighting over one query.

Coverage now pays on two fronts. AI assistants answer in full sentences and cite sources, and a page that names the entities and answers the sub-questions gives them passages worth quoting; our GEO versus SEO guide covers where the two disciplines overlap, and the AI visibility check shows whether assistants mention you today. Coverage also gives a site its wiring. A page that names its subtopics has natural places to link to the pages that treat them in depth, which is how scattered pages turn into sections.