Is SEO dead?

No. But the version of "no" worth listening to concedes the parts that are genuinely true: a large share of informational clicks really has been absorbed by answers that appear above the results, the studies measuring it are specific, and the people who lost that traffic are not imagining it.

Two things are true simultaneously, and most writing on this topic picks one and shouts. One: the click economics of search got materially worse in 2025, particularly for pages whose job was answering a question. Two: the discovery machinery works exactly as it always did - four sites launched between 8 and 25 July 2026 have accumulated 27,260 impressions and 136 clicks by 10 August 2026, having started from zero with no history, no links and no brand.

Those aren't contradictory. Search still finds you, indexes you and shows you to people. It converts fewer of those appearances into visits than it did three years ago. "Dead" is the wrong word for that; "repriced" is closer.

The strongest case for "SEO is dead" is not vibes. It's two findings, both from 2025, both worth taking seriously and stating precisely rather than dramatically.

The top result lost a large share of its clicks when an AI answer sits above it

A December 2025 Ahrefs study of AI Overview impact found that the top-ranking organic result lost roughly 58% of its click-through rate when an AI Overview appeared above it. Note what that measures: same page, same rank, different results page. The ranking didn't move. The value of the ranking did.

Most searches now end without anyone clicking anything

A 2025 analysis of search behaviour found the share of searches ending in no click at all rose from roughly 56% to roughly 69% inside a single year. That's a step change, not a drift, and it is the number to quote at anyone who insists nothing happened. A majority of searches were already zero-click before AI answers existed - people checking a fact, a score, a definition - but the majority got substantially larger, fast.

The damage is concentrated, not evenly spread

This is the part that gets lost. Absorption is worst exactly where a short answer finishes the job: definitions, "what is", "how does X work", basic how-tos. It is far weaker where the user needs to do something - buy, compare named options, check availability, sign up, download, get a price. An answer engine can answer a question. It cannot complete a transaction on your behalf. If your traffic was built on explaining things, you have lost real ground. If it was built on people trying to solve a problem with a tool, you have lost much less.

The strongest counter-evidence to "SEO is dead" isn't an argument. It's that brand-new domains with no history still get indexed and still accumulate impressions quickly, in 2026, not 2019.

itscheaper.in went from zero to 16,500 impressions in about 17 days, peaking near 2,400 impressions a day. That is a dead channel behaving with suspicious vigour. Every one of the four was indexed and being shown to real people inside a fortnight of launch, with no links, no promotion and no domain history.

The honest limits of this evidence

These sites cannot tell you how much AI answers cost you, and it would be dishonest to pretend otherwise. At an average position of 22, a page earns almost no clicks regardless of what sits at the top of the page, so the low click-through rates here are explained by position long before they need an AI explanation. What the data does establish is narrower and still useful: discovery, indexing and impression accumulation are alive and fast. The click side of the equation is genuinely worse, and the studies above are better evidence for that than these four sites are.

Nothing replaced SEO. The target moved, and the same underlying craft now aims at a slightly different outcome.

The last row is the one worth arguing about. You are now competing to be a source a model prefers over its own generation. Text indistinguishable from what the model would have written anyway gives it no reason to cite you. First-party numbers - your conversion rates, your Search Console curves, your failure rates - are the only content category that is structurally impossible to synthesise. This is covered in more depth in how to get cited by ChatGPT.

• Shift the content mix toward the transactional end. Comparisons, alternatives, pricing, "X for Y" pages, live data. These survive answer absorption because they end in a decision rather than a fact.
• Stop writing definitional posts as a traffic strategy. "What is X" is exactly the query class an AI answer completes. Write those for credibility or internal linking if you like, but don't put them on the revenue plan.
• Make every important claim self-contained. A paragraph that holds up when lifted out of your page - claim and support in the same place - is quotable. An argument spread across six paragraphs with pronouns pointing backwards is not.
• Watch impressions against clicks, not clicks alone. Impressions steady while clicks fall is the fingerprint of absorption. Both falling together is a ranking problem and has nothing to do with AI.
• Measure citations on purpose. Fix a set of buyer prompts, run them on a schedule across the assistants, and record who gets named. A citation produces no click and will never appear in your analytics, so if you don't go and look, it isn't happening as far as you know.
• Publish something only you know. The most reliable way to be quoted is to be the origin of a number. Every operating business has several and publishes none of them.

Stating this in advance is the only thing that separates a position from a posture. If the following happen, "SEO is dead" becomes the correct answer and this page gets rewritten with that conclusion and a date on it.

• New domains stop getting indexed and shown. If a site launched in 2027 cannot accumulate impressions in a fortnight the way these four did, the discovery machinery has actually changed and the whole argument above collapses.
• Transactional queries start getting absorbed too. If assistants begin completing purchases, comparisons and signups end-to-end without ever handing off to a site, the safe harbour described above disappears.
• Citations stop producing anything measurable. If being named in AI answers turns out to generate no visits, no signups and no recall over a long enough window, then the replacement outcome isn't an outcome and "dead" is fair.
• Our own four sites flatline. The impression curves above are five weeks old. If they plateau and clicks never follow through months three to six, that's evidence against the optimistic reading, and it will be published on the receipts page whether or not it's convenient.

The GEO Audit runs your category's real buying prompts across the major assistants, records who gets named instead of you, and returns the specific fixes - with a 90-day recheck so you find out whether they worked.

Or get the 30-day post-build playbook Send it