SEO Did Not Die. It Picked Up Two More Jobs.

The short version

Why the ladder metaphor breaks

Ladders imply order. You do the audit, then keywords, then technical, and somewhere around rung seven you "start doing GEO." That is not how any of this works in practice. A page with a broken canonical tag will not get cited by an AI system no matter how quotable the prose is, because the system never sees it. A page with perfect technical hygiene and vague, unsourced content will rank for a while and get skipped by every answer engine.

The other problem is that several rungs on the popular version are not really separate disciplines, and putting link building at the summit, as the final unlock for "SEO success," reflects the SEO of 2014 more than 2026, per Search Engine Journal's report on the Illyes comments. We have made the longer version of this argument before, in nine funerals for SEO since 1997 and in AEO vs GEO vs SEO.

What the ten items actually describe is three overlapping jobs.

The rest of this piece takes them in that order, because that is the order the evidence supports.

Test one: can this page rank?

This is the old job and it is still the foundation. Nothing about AI search removed the requirement that a page be found, fetched, rendered, indexed and judged relevant. Google's AI features documentation is explicit that AI Overviews and AI Mode draw only from pages that are indexed and eligible for a standard snippet, and that the same controls, from robots.txt to nosnippet and max-snippet, govern what those features can use. If you block Googlebot or noindex a page, you have opted out of AI visibility on Google, not just out of the ten blue links.

The first five rungs of the ladder, audit, keyword research, technical SEO, on page SEO and content optimization, are all in service of this one test. The useful update for 2026 is what "rankable" now has to include. Crawl budget matters more when AI crawlers from OpenAI, Anthropic, Perplexity and others are hitting your server alongside Googlebot. Structured data needs to match the visible content exactly, because Google says mismatched markup is a reason for it to ignore the page's data; we covered where that leaves structured data in AI search. Internal linking is how AI systems and crawlers alike discover which page on your site is the canonical answer to a given question, so a topic buried three clicks deep with no inbound internal links is effectively invisible, which is the case we make in internal linking for SEO and AI visibility.

Keyword research has changed the least in mechanics and the most in purpose. Volume still tells you what people type, but intent clusters now tell you what a conversational query will ask next, and AI Mode turns a single search into a multi step session. Mapping the follow up questions, not just the head term, is what earns a page the second and third citation in a long answer.

Test two: can this page answer?

Answer engine optimization got its name from featured snippets and voice assistants, and its core instruction has survived every interface change since: put the answer before the essay. The Pew data explains why this matters more than ever. With AI summaries appearing on roughly one in five searches in Pew's March 2025 sample and the traditional click rate nearly halving when they do, according to Pew Research Center, a page that makes the reader scroll to find the point loses to a page that states it in the first hundred words.

At Silverback we build every article around a self contained summary block at the very top, before the intro, precisely because AI systems lift the earliest complete answer on a page. This article does it. The FAQ answers in the frontmatter of our posts are written to stand alone if extracted. None of that is a trick. It is what a good reference page looked like before AI existed, made non negotiable by the fact that an AI system reading the page has no patience for a scene setting anecdote. The practical playbook is in how to optimize content for Google AI Overviews.

The AEO test for any page is simple. Read only the first paragraph under the main heading. Does it answer the question the page targets, with a specific figure, date or name? If not, the page is asking the reader, and the machine, to do work they will not do.

Test three: can this page be cited?

This is the new job, and it is the one where most sites are weakest, because it asks for something SEO never required before: your page has to be worth quoting.

The best evidence for what "worth quoting" means comes from the GEO paper by Aggarwal and colleagues, accepted at KDD 2024, which tested nine content changes across a 10,000 query benchmark and measured how often each version of a page was used in a generative answer.

The paper's authors also note that the strongest methods vary by domain, so the exact numbers will differ for a law firm and a running shoe brand. The direction does not.

Read those results next to the industry's obsession with new file formats and the gap is obvious. Google's John Mueller said in June 2026 that llms.txt is "purely speculative for now," that the file "has existed for years, yet none of the AI systems use it," as reported by Search Engine Journal, and that the most basic agentic optimization is simply this:

Don't block agents.

Meanwhile the thing that measurably works, adding real statistics with real sources, is the thing most content teams skip because it is slow.

Citability also has a trust layer that no content change can fake. Google's helpful content guidance frames it as three questions: is it self evident who wrote this, is it clear how it was produced, and is it obvious why it exists. That "who, how, why" test is the practical form of E-E-A-T, and it is the same thing an AI system is implicitly evaluating when it decides whether to attribute a claim to your page or to Wikipedia. Pew found Wikipedia, YouTube and Reddit were the most cited sources in AI summaries, and that .gov sites made up 6 percent of AI summary sources versus 2 percent of standard results. The machines already prefer sources with obvious provenance. Bylines, dates, methodology and outbound citations are how a commercial site earns the same treatment, which is why we argued that E-E-A-T matters more in AI search.

This is also where links belong on the ladder: not at the top as the final boss, but inside the trust layer. A link from a publication that genuinely references your data is a citation in every sense that matters. A hundred links from directories are noise. The goal was always credibility. The volume game is just over.

Where the money is starting to follow

There is a reason to care about the third test beyond visibility. Reporting on September 17, 2026 confirmed that Google's AI Contribution pilot pays participating publishers when their content meaningfully contributes to an AI generated response in AI Overviews, AI Mode or Gemini, with earnings tracked in a dedicated Google Search Console panel. It is small and early, dozens of publishers rather than thousands, and skewed toward small and mid sized sites. But it is the first time the largest search engine has put a dollar figure on being cited rather than clicked. That does not change the strategy. It confirms it.

The zero click numbers point the same way. SparkToro's Rand Fishkin, analyzing Similarweb clickstream data from January to April 2026, found that only about 32 percent of U.S. Google searches produced any click at all, down from 41 percent in 2024, and that 27 percent of the clicks that did happen went to Alphabet properties such as YouTube, Maps and AI Mode, per Search Engine Journal's coverage of the study. Fishkin flags real caveats: the data excludes Google's mobile app, where zero click behavior is more aggressive, and the panels differ across years. Even with those caveats, the trend is not in dispute. Attention is moving into the answer, and being the source of the answer is the only durable position left.

What to actually do: the one page test

Pick your single most important commercial page, not the homepage, and run the three tests in order.

Do that for one page this week and you will know which of the three jobs your site is worst at. That is more useful than any ladder, because it tells you where to spend the next month. If you want a second set of eyes on the diagnosis, that is exactly what our SEO and GEO work starts with.