The most useful mental model for search in 2026 is a split screen. On one side, the classic results page still drives the majority of clicks. On the other, AI answers in ChatGPT, Perplexity, Copilot and Google’s own AI Overviews resolve a growing share of questions before a click ever happens. The strategic mistake is treating these as two projects with two budgets. They are one system with two outputs, and most of the work feeds both.
What transfers directly from classic SEO
More than the headlines suggest. Crawlability and indexation still gate everything, because answer engines retrieve from search indexes: a page Google cannot index is a page no AI Overview will summarise. Links and brand authority still decide which of several adequate sources gets chosen. Search intent still rules, since a page that misreads what the buyer wants fails on both surfaces. If you have spent years on clean architecture, genuine expertise and earned links, none of that is wasted. The one caveat is thin content: pages that survived on domain strength alone are the first casualties, because passage-level retrieval has no reason to pick an empty paragraph.
What is genuinely new
Three things. First, retrieval is passage-level, so a single well-written paragraph can earn a citation even when the page as a whole does not rank first. Second, unlinked brand mentions now carry weight, because language models learn brand associations from everything they read, not just from anchor text. Third, queries are getting longer and more conversational, since people describe problems to a chat box in full sentences they would never type into a search bar. Your keyword research needs a question layer it probably does not have yet.
Follow where the clicks actually go
Informational queries are losing clicks fastest, because a short summary satisfies them. Commercial and local queries still send traffic, since someone comparing suppliers wants detail, proof and prices that a three-line answer cannot carry. Plan accordingly: informational content earns citations and brand recall, commercial pages earn visits and revenue, and each should be written and measured against its own job rather than one blended traffic target.
Content that serves both surfaces
Write pages that answer one question completely, open every section with its conclusion, and keep claims specific enough to quote. Then go deeper than a model can: original data, first-hand testing, real screenshots and named trade-offs. Generic summaries are the one content type AI has made worthless, because the engines produce those themselves. What they cannot produce is your experience, and that is exactly what they need to cite.
A worked example: instead of another “what is technical SEO” explainer, publish the crawl waste you actually measured on a large site, what changed after the faceted navigation was blocked, and how long reindexing took. That page earns citations because nobody else has it.
The technical layer both surfaces read
Serve full HTML without requiring JavaScript, because several AI crawlers do not render scripts at all. Decide your crawler policy deliberately: GPTBot, ClaudeBot, PerplexityBot and Google-Extended each respect robots.txt, and blocking them removes you from those answers. Keep Organization, Article and FAQ schema accurate, publish an honest About page, and keep your facts consistent everywhere, since models notice contradictions between your site and the wider web.
Measurement without vanity
Rankings and sessions no longer describe the whole funnel. Add three measures: citation share across a fixed monthly prompt set, referral traffic from AI domains, and branded search volume, which rises when AI answers mention you and readers go looking. Then tie all of it to leads and revenue by landing page, because a citation that never produces a customer is trivia. A simple spreadsheet with one row per prompt per month is enough to show direction; tooling can come later once the habit exists.
The practical sequence
Audit indexation first, then rewrite your highest-value pages for passage-level clarity, then fix schema and crawler access, then start the monthly prompt tracking. In that order, each step makes the next one measurable. Teams that run this as one connected system are quietly taking positions on both surfaces while their competitors still debate whether AI search matters.