What actually changed
AI Overviews didn't kill organic traffic everywhere. It killed the easy traffic — definitional queries, "what is X" questions, anything answerable in two sentences. For those queries the SERP now hands the answer back directly.
The traffic that survived (and in some niches grew) was traffic to pages that covered an entity, not just a keyword. That is the shift worth understanding.
What "entity SEO" means in 2026
An entity is the thing your page is about, in a way a knowledge graph would recognise — a product, a method, a person, a company, a place. Entity SEO is the practice of making sure your page covers that entity completely enough that AI systems and traditional ranking systems both pick it as the canonical source.
Concretely: not "10 tips for X" but "everything you need to know about X, with sources, an author, and the structured data to prove it".
A 3-step audit you can run this week
- List the entities your page should cover. Pull related entities from your top-3 ranking competitors and from the People-Also-Ask box. If any are missing from your page, that is a coverage gap.
- Check your schema. The page needs structured data that names the entity (Product, Article, HowTo, etc.) and links the author, organisation and citations correctly.
- Score your authoring signal. Named author with credentials, dated publication, dated last-updated, and outbound citations to authoritative sources. Pages without these signals are penalised in AI selection even if they rank traditionally.
How to score entity coverage
The crude but useful test: paste your URL into ChatGPT or Perplexity and ask "what is this page about". If the system can name the entity precisely and list 2–3 supporting facts from the page, your coverage is decent. If it produces a vague summary or hallucinates details, the page is undercovered.
What entity SEO replaces
It replaces the old "publish a 1,500-word piece around the keyword" workflow. That workflow still produces traffic, but a shrinking share of it. The shift is toward fewer pages, each covering its entity more completely, with the structured data and authoring signal to back it up.
For most content programs we run, that means cutting the publishing cadence in half and tripling the depth per piece.