The short version
- What still works: genuine expertise, clean technical foundations, useful content, corroboration from other sites. Unchanged for fifteen years.
- What changed: the payoff moved. Position 1 is worth less when an answer sits above it; a citation is worth more than it used to be.
- What stopped working: volume plays. Publishing 200 mediocre pages was already weak; against systems that synthesise, it is close to worthless.
- What never worked: keyword density targets, word-count minimums, and most of the advice built around gaming a scoring widget.
Every few years SEO announces that everything has changed. Usually it has not. This time something genuinely has — but it is narrower and more specific than the "SEO is dead" genre suggests, and separating the real change from the noise is worth doing carefully.
What did not change
- Being genuinely good on a subject. Still the foundation. Google’s own guidance on helpful, people-first content reads almost identically to how it did before generative answers existed.
- Technical accessibility. If a crawler cannot render and index your page, nothing downstream matters. See JavaScript SEO basics.
- Clear information architecture. Pages that are easy to understand structurally are easier for every consumer of them, human or machine.
- Corroboration. Other credible sites referring to you still signals something, whether as a link or a mention.
- Search intent. Matching what the person actually wants remains the whole game.
What genuinely changed
The second row is the one that breaks teams. Every SEO dashboard ever built measures sessions. If a meaningful share of your value now arrives as "the assistant recommended us and they came directly next week", your reporting will show a decline while your business does not. Expect to have that conversation with someone.
What stopped working
- Volume plays. Publishing large quantities of adequate content to blanket a keyword set. It was always a depreciating asset; against systems that synthesise across sources, thin coverage adds nothing worth citing.
- Pure aggregation. Pages whose only value was collecting facts that already existed elsewhere. That is precisely what an assistant does for free.
- Ranking without substance. Technically optimised pages with nothing behind them could rank. They do not get cited, because there is nothing to cite.
What never worked
Worth a separate section, because a lot of it survives in plugin interfaces and agency decks: keyword density targets, minimum word counts, LSI keyword lists, exact-match phrase repetition quotas, and "optimisation scores" that grade content against averages of what currently ranks. None of these was ever a ranking factor. They were proxies that correlated loosely and got mistaken for causes. See keyword density vs semantic coverage and editorial quality vs optimisation score.
What to do differently, in practice
- Audit for substance, not coverage. Ask which of your pages contains something not available elsewhere. Consolidate or delete the rest — see refresh vs new content.
- Front-load answers inside your existing long pages rather than writing new short ones.
- Fix your entity description. One plain sentence about what you do, used consistently everywhere.
- Add first-party evidence. Original data, real screenshots, tested numbers. This is the only durable moat against synthesis.
- Measure clicks and brand demand, not just average position. Watch branded search volume in Search Console as a proxy for assistant-driven awareness.
The one durable advantage
Anything an assistant can synthesise from other sources, it will. The content that survives is content it cannot produce: your own data, your own tested results, your own specific experience. That is not an SEO tactic — it is a reason to actually do things worth writing about.