The most important sentence in search is no longer here are the results. It is ask anything. That small shift tells you nearly everything. The old search engine was a switchboard operator, a useful bureaucrat pointing you onward. The new one wants to behave like a concierge, a researcher, a shopping assistant, a browser, and eventually a mild domestic servant with too much confidence. Its ambition is not merely to send you to the page. It is to finish the errand before you get there.
Google now describes search as moving “beyond information to intelligence”.[1] AI Mode breaks a query into subtopics, fires off a multitude of searches on your behalf, pulls the returns back into a synthetic answer, and then invites follow-up as if the web were merely the back office for a smoother front desk.[1] For research, it offers Deep Search. For errands, it offers agentic booking and buying. For shopping, it promises that the machine will keep watch and check out when the price is right.[1] The page of blue links is being treated as a quaint intermediate format, like a railway timetable in the age of the chauffeur.
Some of this is genuinely useful. If all you want is the date of the next full moon, the height of a mountain, or a quick sanity check on whether anchovies belong in puttanesca, one concise answer is perfectly civilised. The problem begins when the same logic expands from factual retrieval into the middle of journalism, criticism, reviews, recipes, how-to pages, and every other form whose value lies precisely in the visit. That is where the machine stops answering a question and starts pre-empting a relationship.
Google is redefining the loss as a better class of click
The company insists that nothing essential is being lost. In August 2025 Google said total organic click volume from Search to websites had remained “relatively stable year-over-year”, that it still sends billions of clicks to the web every day, and that clicks coming from AI-heavy results are now of “higher quality” because users are less likely to bounce straight back.[2] This is the sort of corporate phrasing that deserves to be held by the ankles for a moment. When an intermediary keeps more of the attention for itself but tells the suppliers that the remaining traffic is better behaved, it is not describing health. It is describing rationing in a soothing tone of voice.
Google’s defense is clever enough to matter. If users ask more questions, and if some of those questions are longer and stranger than classic search would ever have encouraged, then search can indeed claim growth while sending a thinner share of value onward. The pie becomes larger. The slice reaching the original publisher becomes smaller. One can see why Mountain View likes this arithmetic. It preserves the rhetoric of abundance while quietly revising the economics of being found.
The trouble is that outside Google’s house the numbers are much uglier. Ahrefs re-ran its study of 300,000 keywords using aggregated Search Console data and found that by December 2025 the presence of an AI Overview correlated with a 58% lower click-through rate for the top-ranking page.[3] Semrush’s study of 200,000 AI Overviews found an average of roughly eleven links inside an overview, but only about 20% to 26% overlap between those cited links and the ordinary top ten results.[4] So even when Google says it is still linking out, the links are being redistributed inside a format that no longer treats ranking as a meaningful promise of traffic.
The visit is being converted into upstream inventory
This is the mechanism people keep softening with euphemisms like discovery or enhanced search experience. The simpler description is that the open web is being turned into supply. Publishers, forums, archives, merchants, and independent writers still do the expensive part: reporting, testing, comparing, explaining, photographing, reviewing, maintaining. Search increasingly performs the glamorous part: the smooth summary, the organised answer, the polished surface that receives the gratitude.
Once you see it in those terms, the rest of the design choices make perfect sense. Google says AI Mode uses “query fan-out” to issue many searches simultaneously on the user’s behalf.[1] That sounds like depth. Economically it means one user gesture can trigger a harvest across many sites while returning as a single synthetic object. Search ceases to be a path through the web and becomes an extraction layer above it. The user sees convenience. The publisher sees a citation where a visit used to be.
This is why the click-through studies matter more than the tone of the product demos. Ahrefs’ 58% figure is not a mere line wobbling inside an SEO trade deck; it is a description of leverage moving uphill.[3] Digiday reported data from nineteen Digital Content Next member companies showing median year-on-year Google Search referral declines of 10% overall, 7% for news brands, and 14% for non-news brands across an eight-week stretch.[5] The same reporting notes that publishers cannot cleanly refuse Google’s AI use without risking their presence in search, which is rather like being told you may object to the dinner guest’s manners only on condition that he also repossess the road to your house.[5]
The consequence will not be the death of the web, and one should avoid that sort of melodrama. It will be something more tedious and therefore more believable. The big brands will keep enough direct traffic, legal muscle, and name recognition to survive. Reddit and YouTube will absorb what still leaks out, because platforms already fat with habit tend to receive the leftovers of every regime change. The middle will thin. Specialist publishers, independent review sites, patient how-to archives, and small magazines will continue doing the work while receiving a steadily worse bargain for having done it.
Search is no longer neutral infrastructure
There is a more philosophical mistake buried underneath the traffic debate. For twenty years we spoke about search as if it were public-minded infrastructure with advertising attached. It was never that innocent, but it did at least preserve one discipline: to find the thing was still to leave. That outward motion kept the wider web economically alive. AI search weakens that discipline. It turns orientation into substitution.
The result is a subtle but serious change in what writing on the web is for. Under the old arrangement, a good page had the chance to become a destination. Under the new one, the page risks becoming training data with a logo attached. The machine does not need to misquote you to damage you. It only needs to absorb enough of your utility that the reader never arrives. In that world, originality is not just stolen; it is skimmed.
One sees the same instinct in every AI convenience pitch of the last year. Meetings should be summarised before anyone digests them. Search should answer before anyone reads. Shopping should complete before anyone compares. The public language is freedom from friction. The underlying project is to remove the human interval in which judgment used to happen. The web, poor old creature, is simply the largest available quarry.
The adult response is separation, not nostalgia
I do not particularly care for the romanticism that pretends the ten blue links were a golden age. Much of classic search was spam, arbitrage, copied sludge, and listicles written by people who ought to have been prevented from touching a keyboard. Fine. But the answer to that disorder cannot be a settlement in which one platform quotes the whole town while keeping the crowd at the gate.
The practical response is dull, legal, and necessary. Search crawling and AI extraction should be separable by default, not philosophically but technically. If a site wants to be indexed for discovery but not harvested into synthetic answers, that choice should not amount to commercial suicide. Citation inside an AI answer should also become more accountable: clearer attribution, usable logs, and a measurable route back to the source rather than a decorative footnote meant to flatter regulators.
Publishers, for their part, need to stop confusing SEO survival with audience strategy. The sites that endure will be the ones people deliberately seek out: newsletters, memberships, podcasts, forums, communities, archives, magazines with an actual voice, writers whose names mean something before the query is typed. Search will still matter. But trusting it as the generous middleman of the old web now looks naive.
A search engine that answers before it sends is no longer organising the web. It is enclosing it. And once the visit becomes optional, the people making the pages will eventually become optional too.
Sources
[1] Google, “AI in Search: Going beyond information to intelligence”: AI Mode rollout, query fan-out, Deep Search, agentic capabilities, shopping, and personal context.
https://blog.google/products-and-platforms/products/search/google-search-ai-mode-update/
[2] Google, “AI in Search is driving more queries and higher quality clicks”: Google’s own defense of AI-heavy search, including its claims about stable traffic, billions of clicks, and “quality clicks”.
https://blog.google/products-and-platforms/products/search/ai-search-driving-more-queries-higher-quality-clicks/
[3] Ahrefs, “Update: AI Overviews Reduce Clicks by 58%”: 300,000-keyword study using aggregated Google Search Console data, updated through December 2025.
https://ahrefs.com/blog/ai-overviews-reduce-clicks-update/
[4] Semrush, “We Studied 200,000 AI Overviews: Here’s What We Learned”: prevalence, link counts, and overlap between AI Overview citations and top organic rankings.
https://www.semrush.com/blog/ai-overviews-study/
[5] Digiday, “Google AI Overviews linked to 25% drop in publisher referral traffic, new data shows”: Digital Content Next member data, publisher referral declines, and the opt-out problem.
https://digiday.com/media/google-ai-overviews-linked-to-25-drop-in-publisher-referral-traffic-new-data-shows/