Is AI Search Killing Google? What Actually Happened to Search Traffic

Every few months a headline announces the death of Google, and every few months Google declines to die. So let us do the unfashionable thing and read the numbers slowly, because what they describe is more interesting than a funeral.

Is AI replacing Google search?

Measured by raw volume, not remotely. Google still processes billions of queries a day, an order of scale no assistant approaches. ChatGPT reached roughly 900 million weekly active users by February 2026, up from about 400 million a year earlier, according to Similarweb's generative AI statistics. That is a staggering ascent for a product barely three years old, but a weekly user of a chatbot is not the same thing as a daily search habit, and Google's habit is the most entrenched in the history of software. Anyone telling you Google is finished is selling something.

And yet the question refuses to feel settled, because everyone reading this has personally felt their own behaviour shift. You asked a chatbot something last week that you would have Googled in 2023. So did your clients. The honest framing is not replacement. It is rearrangement: which questions go where, and what happens after the answer arrives.

What actually happened to search traffic?

The clearest change happened inside Google itself. In May 2024, Google launched AI Overviews, its own generated answers sitting above the traditional results. In the twelve months that followed, the share of Google searches ending without any click grew from 56 percent to 69 percent, per SEO Sherpa's compilation of AI search statistics. Those overviews now appear on roughly 25 percent of searches, and we walked through their mechanics in our explainer on AI Overviews.

Sit with that 69 percent. More than two searches in three now end on Google's own page, question answered, no website visited. The assassin of the click was not ChatGPT. It was Google, pre-empting the disruption by disrupting itself. Websites lost traffic not because searchers left Google, but because Google stopped needing to send them anywhere.

The pain is not evenly distributed either. Informational content, the definitions, how-tos and explainers that once harvested clicks reliably, is precisely what an AI Overview digests best, so publishers built on that traffic felt the drop first and hardest. Queries with intent behind them, transactional, local, navigational, still push people through to websites at healthier rates. Whether the shift starves the open web that answers are built from is a live argument; what is beyond argument is that the click is no longer the default ending of a search.

How much traffic does AI search actually send?

Here is the number that deflates the hype, and it deserves to be quoted often: AI referral traffic accounts for roughly 1.08 percent of all website traffic across ten industries, according to Conductor's 2026 benchmarks as reported by SEO Sherpa. One percent. After three years of revolution talk, the assistants send about one visitor in a hundred. Of that thin stream, ChatGPT dominates utterly, driving about 87.4 percent of all AI referral visits.

If the story ended there, you could file AI search under interesting but immaterial and get back to your keyword rankings. It does not end there, because counting visits is the wrong way to count. AI assistants are not designed to send traffic. They are designed to finish the conversation. The visits they do send are the survivors of a filter that has already done most of a salesperson's job.

Why does the tiny traffic convert like mad?

Because by the time an AI referral lands on your site, the visitor has already been told, by a source they trust, that you are the answer. Conductor's 2026 benchmarks make the point with unusual violence:

Google organic visitors convert at about 1.76 percent, which is what browsing looks like. ChatGPT referrals convert at 14.2 to 15.9 percent, Perplexity at around 10.5, and Claude referrals at up to 16.8 percent. That is not an audience, that is a shortlist. A channel one-hundredth the size producing conversions at up to nine times the rate is not a rounding error. It is a different kind of channel entirely, closer to referrals from a trusted colleague than to search traffic.

Where did the buyers go?

To the beginning of the journey, which is where the real migration is happening. A G2 buyer-behaviour survey from March 2026 found that 51 percent of B2B buyers now start their research in an AI chatbot more often than in Google, up from 29 percent in April 2025. Seventy-one percent use AI chatbots somewhere in the research process. The first question, the framing question, the who-should-we-even-consider question, increasingly gets asked in a chat window. What buyers actually type there is its own study, and we collected it in the prompts buyers type into AI.

Meanwhile the money has noticed. The US market for generative engine optimization services is projected at around 365 million dollars in 2026, growing at roughly 42.9 percent a year, per Incremys' GEO statistics. Budgets follow buyers. They always have.

So is Google dead? The honest read

No. Say it plainly, because plenty of people in this industry will not: Google is not dead, not dying, and will likely remain the largest source of discoverable traffic for years. It answered the AI threat by becoming an answer engine itself, which preserved its audience while quietly starving the websites beneath it. If anything died, it was the old bargain, the one where Google sent you traffic in exchange for your content.

What follows from that is a portfolio, not a pivot. Search still carries the volume, so the fundamentals that earn Google rankings still pay rent. But the layer deciding who gets considered has moved up, into generated answers on Google and in the assistants, and that layer runs on different fuel: entities, evidence, third-party trust. We explained the discipline built for it in our plain-language guide to answer engine optimization.

Is the real shift about Google versus AI, or about where the click happens?

Framed as Google versus ChatGPT, the story writes itself into a boxing match, and boxing matches need a loser. Framed correctly, it is not a fight between two products at all. It is one shift, showing up in two places at once. Google's own AI Overviews and a separate chatbot like ChatGPT or Perplexity are doing structurally the same thing: reading several sources, composing an answer, and presenting it before anyone has to click anywhere. Whether that composed answer sits inside google.com or inside a chat window is, from the point of view of the person being described, close to irrelevant. Either way, the moment a name gets chosen or dropped has moved earlier than it used to sit, into the paragraph, away from the list of results underneath it.

That reframing matters because it stops you from betting the whole strategy on guessing which interface wins. If Google's AI Overviews had never launched, the same pressure would be arriving anyway, entirely through separate assistants. If chatbots plateaued tomorrow, the pressure would still be arriving through Google's own answer box, which by itself now covers roughly a quarter of all searches. We broke down the mechanics of that box on its own terms in our explainer on Google's AI Mode, and the honest reading is that Google built its own disruption on purpose, rather than wait for someone else to build it first.

How visibility works differently once the point of contact moves from a ranked list to a composed answer.

What does staying findable mean if you don't know which interface wins?

It means optimizing for the thing both interfaces share, rather than for either one specifically. Every generated answer, whichever product renders it, is assembled the same way: pull a handful of trusted sources, weigh them, compose a paragraph that names names when it can verify them. We mapped that assembly process end to end in the anatomy of an AI answer, and the sources it favors, the pages, mentions and structured data it reaches for before it will attach your name to a recommendation, are covered in where AI looks before it recommends someone. Build for that shared mechanism, evidence, consistency, third-party corroboration, and you are protected whichever interface a given searcher happens to be using this year.

The practical version is almost boring in its simplicity. Publish the specific, verifiable material that earns citations, keep your identity consistent across every profile a model might cross-reference, and stop treating "Google traffic" and "AI traffic" as two separate scoreboards. They are increasingly one scoreboard, measuring the same underlying question: does the machine, whichever machine it happens to be, know enough about you to say your name with confidence.

Why does everyone keep predicting Google's death?

Because the story is irresistible, and because the people telling it are often selling the replacement. The death-of-Google genre is nearly as old as Google: social search was going to kill it, then mobile apps, then voice assistants, each obituary written with total confidence and each quietly retracted. The pattern repeats because pundits confuse a change in behaviour at the margin with a change in the default, and defaults are the most durable force in consumer technology. Two billion people do not renegotiate a reflex because a better interface exists for some queries. They split their behaviour, gradually, query type by query type, which is exactly what the current data shows: research and advice migrating toward assistants, navigation and quick lookups staying loyally where they were.

There is also a measurement trap worth naming, because it flatters the dramatic version of the story. AI referral traffic is systematically undercounted: assistants answer without linking, users read a recommendation and then search the recommended name directly, and that visit gets booked as branded search or direct traffic rather than as the AI's doing. The 1.08 percent figure is a floor, not a ceiling, on AI's real influence. So the honest position holds two thoughts at once. The obituary writers are wrong about Google's death, and the complacent are wrong about nothing having changed. The influence moved; the traffic ledgers just have not caught up with where persuasion now happens.

What does this mean if you sell your expertise?

It means the arithmetic of visibility has flipped for exactly your kind of business. If you run a high-volume content site, losing 13 points of click-through to zero-click answers is a wound. But if you sell judgement, advice or skill, you never needed volume. You needed the right hundred people to hear your name at the moment of choosing, and that moment now happens inside an answer. One mention there, converting at shortlist rates, is worth a page of blue links. The mechanics of earning it start with being the person ChatGPT actually recommends, and if you would rather have the whole layer built for you, that is what our services exist to do.

The Monday-morning version, as a checklist. Keep publishing the clear, well-sourced material that earns Google rankings, because it is the same material AI engines cite. Ask the major assistants the ten questions your ideal client would ask, and record whether your name appears. Tag and track AI referrals in your analytics separately, however small the numbers look today, because their conversion rate is the argument for the rest of the work. And treat every prospect who says "ChatGPT suggested you" as a data point worth writing down. Channels are cheap to monitor before they are expensive to catch up on.

Google outlived its obituary. The click did not. Plan for the second fact, and the first one stops mattering very much.