What actually changed
A growing share of research now begins with a question put to a model rather than a query typed into a search box. The model answers, cites a handful of sources, and the buyer arrives — if they arrive at all — already partway through their decision.
It is tempting to treat this as a new discipline requiring new tactics. Mostly it is not. The systems generating those answers are still reading the open web, still relying on crawlable, parseable pages, and still weighting sources by the same broad signals of expertise and corroboration.
What has changed is the penalty for being unreadable. A page that a crawler cannot render, or whose claims cannot be attributed to anyone, was previously ranked poorly. Now it is simply not cited at all.
Structure for extraction, not just for ranking
Answer engines extract passages. That rewards pages where a specific question is answered in a specific place, in language that stands on its own.
In practice: use headings that state the question a reader actually has, answer it in the first two sentences underneath, and then elaborate. Avoid answers that depend on the three paragraphs above them for meaning. Define terms where you use them.
This is also better for human readers, which is the usual test of whether an SEO recommendation is sound. If a tactic only makes sense for machines, it will not last.
Who wrote it, and when, is now a ranking input
Content published with no named author, no date and no sources is difficult for any system to assess. Increasingly it is treated as low-confidence and excluded rather than ranked low.
Name a real author with relevant expertise, show when the page was published and last reviewed, and link to primary sources rather than to other summaries. Where a claim is contested, say so.
None of this is a trick. It is the ordinary practice of publishing responsibly, which has simply become measurable.
The technical floor has not moved
Crawlability, indexation, rendering, internal linking and speed still decide whether anything else matters. A brilliant topic architecture on a site that returns a JavaScript shell to a crawler earns nothing.
Check the boring things first: what a crawler actually receives, whether your important pages are internally linked from somewhere other than the sitemap, whether pagination and filtering are generating thousands of near-duplicate URLs, and how quickly the page becomes readable on a mid-range phone on a poor connection.
Fixing those is unglamorous and usually the highest-return work available.
Measure demand, not rankings
Position tracking is becoming less meaningful as results become more personalised and more synthesised. Non-brand organic sessions, qualified organic conversions and assisted conversions are more stable and closer to the thing you actually care about.
Set the baseline before the work starts, and be honest that this is a two-to-three quarter program. The compounding is the reason to do it; it is also the reason it cannot be rushed.