Semantic SEO is the practice of building content around meaning, topics and the relationships between them, rather than around keyword strings, so a search system can recognise genuine coverage of a subject instead of counting how often a phrase appears.
- Search stopped matching strings a long time ago. Hummingbird in 2013, the Knowledge Graph in 2012, BERT in 2019 — the direction of travel has been consistent for over a decade.
- The unit is the topic, not the page. Coverage is judged across everything you publish on a subject, which is why a topical map beats a blog calendar.
- Depth is demonstrated, not claimed. Answering the awkward adjacent questions is the signal; saying “we are experts” is not.
- It is the substrate the AI layer reads. A generative system deciding whether you are a credible source on a topic is doing semantic evaluation, whatever we call it this year.
Ten years of the same direction
The move from strings to meaning is the long arc of search rather than a recent reaction to AI. The Knowledge Graph arrived in 2012 with “things, not strings”; Hummingbird rebuilt the ranking engine around query meaning in 2013; and in 2019 Google shipped BERT.
Pandu Nayak, then Google Fellow and VP of Search, described BERT on 25 October 2019 as “the biggest leap forward in the past five years, and one of the biggest leaps forward in the history of Search” — a model that could “consider the full context of a word by looking at the words that come before and after it.” Google said it would affect roughly one in ten searches in U.S. English at launch.
Which means the discipline now being resold under new acronyms was already the correct answer in 2019. What changed is the cost of ignoring it: a keyword-stuffed page used to rank badly, and now it also fails to be understood, quoted or recommended.
What semantic work actually is
Map the topic before you write anything. List every question a buyer asks across the whole decision, not the ten with search volume, all of them. The map is the deliverable; the articles are just its instalments. Most businesses have a content calendar where a topical map should be.
One page per intent, not per keyword. “Cost of a privacy fence” and “privacy fence price” are one page. Splitting them makes two thin pages that compete with each other and dilute the signal each was meant to send.
Cover the entities the topic implies. Materials, permits, HOA rules, seasons, warranty terms, the neighbouring trades. A system evaluating whether you know fencing is checking whether the concepts a fencing expert would mention are present, not whether you said “fencing” enough times.
Link by meaning. Internal links should follow the logic of the subject, so the structure of your site restates the structure of the topic. Related-posts widgets do not do this and never have.
Answer the questions with awkward answers. What it costs, how long it takes, when you are the wrong choice. Competitors skip these, which makes them both the cheapest to own and the most convincing evidence that you know the subject.
Then stop publishing for its own sake. Volume without coverage is how sites end up with four hundred posts and no authority. Fewer, planned, complete pages beat a feed.
Where this term ends and the next one begins
How you would know the semantics are landing
Count coverage, not rankings. Take your topic map and mark which questions you have actually answered on a page a stranger could find. The gap between the map and the marks is your real backlog, and it is usually more honest than any tool output.
Watch what you get found for that you did not target. A site with genuine topical depth starts appearing for long-tail phrasings nobody wrote a page for. That drift is the clearest evidence the meaning landed rather than the keyword.
Then check the AI layer. Being named when someone asks a topic-level question, not a brand question, is semantic coverage paying out. That is what the X/12 count measures, and the protocol is published so the number can be audited rather than believed.
This page defines a term. It does not sell anything. How this work is run inside an engagement is on the SEO service page, and the category it belongs to is mapped at AI visibility, defined.
Want the one-paragraph version? The Visibility Glossary keeps short definitions; this page keeps the practice.
Fair questions about semantic SEO
Is this just a fancy way of saying 'write good content'?
Partly, and the people who say so are not wrong about the writing. Where it differs is planning. “Write good content” produces good pages in no particular arrangement. Semantic work decides which pages exist, how they relate, and when the topic is finished, which is a structural decision made before anyone writes a word.
How many words does a semantically optimised page need?
There is no number, and any tool giving you one is measuring its competitors' word counts rather than your subject. The right length is however long it takes to answer the question completely and then stop. Padding to hit a target actively hurts, because it buries the answer a machine was going to extract.
Do I need to use LSI keywords?
“LSI keywords” is a term the SEO industry borrowed from an unrelated older technique, and Google has said plainly it does not use it. What is real is that covering a topic properly naturally produces the related concepts. Write the complete answer and the vocabulary takes care of itself; reverse-engineering a word list does not.
Should I delete thin pages?
Consolidate before you delete. Two thin pages competing for one intent usually want to be one good page with a redirect. Genuine deletion is right for pages with no intent behind them at all, but check what links to them first, because orphaning a page a customer bookmarked is a worse outcome than leaving it up.
How long before semantic work shows up?
Slower than technical fixes and more durable than either. You are accumulating evidence of coverage, which no single publication delivers. Expect months, expect it to be non-linear, and take the coverage inventory at the start so you can tell progress from mood.
See what the engines already think you cover
The free Visibility Check runs 12 real buying questions for your trade and market across three assistants, your X/12, the screenshots, and the first fix worth making. No sales call.
Get my free Visibility Check