no. but the version of the question worth answering is not "dead or alive", it is "what exactly stopped working", and there the honest answer has a real loss in it. ranking still works. the click that used to come with a ranking arrives less often, because the answer now frequently appears on the results page or inside an assistant. what did not change is the signal those systems use to decide who is worth quoting, and that signal is the link graph.
rankings still work. clicks per ranking are worth less. the underlying credibility signal, who links to whom, is unchanged and is what both search results and AI answers are built on top of. the strategy shift is from chasing traffic volume to being the source that gets named.
what people are actually asking
"is seo dead" is asked far more often than any other question in this category, and it spikes rather than trends, which tells you it is a reaction to news rather than a research task. underneath it are three different worries, and they have three different answers:
- will my traffic go away? partly, and unevenly. the pages most exposed are the ones whose entire answer fits in a sentence.
- is the work still worth doing? yes, but the payoff moved. being cited is now worth more relative to being clicked than it used to be.
- do the old signals still count? this is the only one with a factual answer rather than a judgement, and it is the rest of this post.
what the link graph shows
we run a backlink product on the open common crawl hyperlink graph, so the one thing we can speak to first-hand is whether the web's link structure is decaying. it is not. here is the current release as we index it, and the one piece of arithmetic worth doing on it:
current common crawl release, apr-jun 2026, as we index it domains (nodes) 121,100,000 domain-level links (edges) 3,900,000,000 mean links per domain 3.90e9 / 1.211e8 = ~32 the mean is the least useful number on this page. link distributions are heavily skewed: a small number of domains hold a large share of the edges and the median domain sits far below 32. quote the shape, never the average.
two honest caveats, because this is the part people quote out of context. the domain and edge counts are measured: they are the size of the release we index. the ~32 figure is arithmetic over those two numbers and nothing more. link distributions are heavily skewed, so a mean flatters the typical domain badly, and the median domain has far fewer than 32 referring domains. if you take one number from this post, take the shape rather than the average: a small set of domains carries a large share of the edges, which is exactly why which domains link to you outperforms how many.
that skew is also the whole reason a gap analysis is a better use of an afternoon than a link count. the useful question is not how many links you have, it is which credible domains link to a competitor in your subject and not yet to you. referring domains and backlinks are different counts and the difference matters more now, not less.
what genuinely changed
the real change is that a ranking and a visit have come apart. a result can be shown, read, and acted on without anybody arriving on the page. that is not a theory, it is what a zero-click answer is. the practical consequences:
- definitional queries are the most exposed. if the complete answer is one sentence, an answer box or an assistant can reproduce it and the page becomes optional.
- positions past the first few are worth less than the position number suggests. ranking eleventh has always been worth far less than ranking tenth. an answer occupying the top of the page pushes that cliff further up.
- the citation replaced the click as the prize. being named as the source of an answer reaches a reader who is closer to deciding than a casual visitor ever was.
what did not change
every system that answers a question has the same problem: out of millions of pages that mention the subject, which few are credible enough to repeat? nobody has found a cheaper answer to that than the link graph. a link is the only signal on the open web where one site spends its own reputation to vouch for another, and that property does not stop being useful because the interface changed.
the retrieval layer behind AI answers is built on crawled web pages and the link structure between them, the same raw material search indexes are built from. how heavily any particular system weights it is not published by any vendor, so treat confident numbers about it as marketing. the defensible statement is narrower and still useful: being linked from credible domains in your subject has not stopped helping, and no one has demonstrated a signal that replaces it.
if you want the mechanics of where that link data comes from and how far apart the vendors are, where ahrefs and moz get their backlink data walks through the crawlers behind each index.
so do backlinks still matter
yes, with one correction to how the question is usually asked. the thing that stopped mattering is the count. a domain with four hundred links from directories nobody reads is not ahead of a domain with nine links from publications its industry trusts, and it has not been for years. what matters is the identity of the linking domain, and the skew in the graph above is why: the edges are not interchangeable.
that reframing is good news for small sites, because it is winnable. you cannot out-publish a large competitor. you can find the eight domains that link to three of your competitors and not to you, and write to them. that set is small, specific, and almost always reachable.
what to do instead of panicking
- write things a summary cannot replace. original measurement, a real worked example, a number you generated yourself. if a two-sentence answer is a complete substitute for your page, the page was always fragile.
- be crawlable and quotable. plain html, a clear answer near the top of the page, and a claim stated in one sentence that a machine can lift without mangling.
- work the gap, not the count. the domains linking to competitors in your subject but not to you are a finite list, and it is the highest-yield outreach list available.
- stop scoring yourself on total backlinks. it was always a vanity number and the skew in the graph is the reason.
and if what you actually want is the mechanical version of this, the page-level habits that make a passage quotable and attributable, what actually makes a page citable by an ai assistant is the companion to this post rather than a repeat of it.
the fastest way to settle whether links still move anything for your site is to look at the ones you are missing. put your domain and a competitor into the free backlink gap analysis and read the list of domains linking to them and not to you. the first five gaps come back for an email, with no account to create.
faq
Is SEO dead in 2026?
No, but the traffic per ranking is worth less than it was, and that is a real change rather than a reassurance. Search engines and AI assistants increasingly answer the question on the results page, so a position that used to send a click now often sends nothing. What has not changed is how those systems decide who is credible enough to quote: they still read the web's link structure. The honest summary is that ranking still works and clicking through happens less often, so the value moved from traffic volume toward being the source that gets named.
Do backlinks still matter?
Yes, and the reason is structural rather than a matter of policy. A link is the only signal on the open web that one site spends its own reputation vouching for another, and no ranking system has replaced it with anything cheaper to trust. The current Common Crawl release contains 3.90 billion domain-level links across 121.1 million domains, so the graph is not shrinking. What changed is that raw link count matters less than which domains link to you, because the systems reading the graph weight the source.
Does AI search use backlinks?
The retrieval layer behind AI answers is built on the same crawled web and the same link structure that search indexes use, so link-derived authority is part of how candidate sources get chosen and ranked before a model ever summarises them. No vendor publishes the exact weighting, so treat any specific number you read as a guess. What is safe to say is that being widely linked from credible domains in your subject has not stopped helping, and nobody has demonstrated a replacement signal.
If AI answers the question, why rank at all?
Because AI answers cite sources, and the citation is the new position one. An assistant that answers a question and names your site as where the answer came from delivers the credibility that a blue link used to, often to a reader who is further along in deciding. The tactics that get you named are the ones that got you ranked: be genuinely the best source on a specific question, publish it somewhere crawlable, and be linked to by domains that already have standing.
What should I stop doing?
Stop buying traffic-volume forecasts as if they were revenue, and stop chasing keywords whose answer fits in a sentence, because those are the ones an AI answer absorbs completely. Stop treating total backlink count as a scoreboard. The work that still pays is narrower: questions with enough substance that a summary cannot replace the page, and links from domains that are themselves trusted in your subject.
writes the queries we run internally. ships one tactical post a week.
plus one when a new common crawl release lands. that is all.