Sometime in the past few weeks, a particular kind of post started appearing with unusual frequency on marketing LinkedIn. SEO professionals, content strategists, and agency founders, many of them people who have spent careers building expertise in search optimisation, are publicly questioning whether what they know still applies. The trigger was a cluster of events arriving in quick succession: SparkToro's data showing web traffic down 46% in three years, Google quietly deprecating FAQ rich results while publishing a guide that debunks a significant portion of the GEO/AEO playbook, and Rand Fishkin declaring that traffic is now a terrible goal.

The industry is not panicking about one thing. It is panicking about three things simultaneously, while treating them as the same crisis. They are not the same crisis.

The traffic argument

SparkToro and Datos have been publishing clickstream data on this for years, but the cumulative picture became impossible to ignore in 2025. Zero-click searches account for 56% of Google desktop searches. Google generates paid or organic clicks for just 44% of queries, down significantly from over 60% a decade ago. For news publishers, the deterioration has been faster: Google Search traffic to news publishers dropped from 51% of all publisher traffic in 2023 to 27% in the fourth quarter of 2025, a loss of nearly 24 percentage points in two years (Datos & SparkToro, 2026).

Fishkin's response to this data is Zero-Click Marketing: stop optimising for the click and start optimising for presence, awareness, and brand authority on the platforms where audiences already exist. The argument is that Google is now answering almost two-thirds of all queries without producing a click, and many of the 40% that do send traffic go to branded or navigational terms, meaning someone already knows where they want to go (SparkToro, 2025).

This is a legitimate observation about a real structural shift. The problem is what it gets turned into: a prescriptive alternative strategy that skips over the question of how any brand acquires the presence and authority required to benefit from zero-click visibility in the first place. The Condé Nast CEO directing all brands to operate as if search traffic will be zero is a reasonable institutional response for an organisation with a century of editorial authority across some of the most recognised media brands in the world. It is a considerably less useful directive for a company that nobody has heard of yet.

Zero-click is an outcome of having built something worth knowing about. It cannot be adopted as a tactic by organisations that have not done that work.

The Google argument

On 15 May 2026, Google published a new guide on optimising content for generative AI features in Search. The guide includes mythbusting of common AEO and GEO misconceptions, and makes the case for why SEO best practices continue to be relevant for success in generative AI features on Search. In the same period, FAQ rich results were deprecated entirely (Google Search Central, 2026).

A substantial industry had grown around exactly the practices Google is now explicitly calling misconceptions: reformatting content into FAQ structures, rewriting headers as questions, artificially shortening content blocks for LLM consumption, and producing high volumes of AI-generated pages to capture long-tail queries. Sites with high-volume AI content without subject matter expertise saw traffic drops of 50 to 90% following the 2025 helpful content updates (Google Search Central).

The SEO professionals who built practices around these tactics are not being punished for being incompetent. They are being punished for being too competent at the wrong thing: optimising for a model of how search works that Google has spent several years progressively invalidating.

The technical argument

The most precise diagnosis of what is actually happening comes from Pedro Dias, writing in The Inference. Language models exist because the web is a mess. The transformer architecture handles this by treating language as sequences of tokens. There is no parser inside the model looking for schema tags. There is no preference for FAQ markup. The model reads the words. That is the mechanism (Dias, 2026).

The GEO and AEO vendors selling structured optimisation frameworks for AI visibility are, in Dias's argument, committing a category error. Schema has real, well-defined uses: classical Google search uses it for rich results, the Knowledge Graph uses it for entity disambiguation, voice assistants pull structured fields from it. Schema cannot reach into a transformer and improve its comprehension of prose. The model is not architected to read schema as schema.

What the academic paper most frequently cited by the GEO/AEO industry actually found is worth reading carefully. The methods that produced the largest visibility lifts were: adding citations from credible sources, adding quotations from relevant sources, adding statistics, and improving fluency. The paper's authors note in plain terms that techniques effective in search engines may not translate to success in this new paradigm. Schema, structured data, FAQ markup, and heading hierarchy are not tested in the paper, because they are not the optimisation surface the paper studies (Aggarwal et al., KDD 2024).

Dias's most useful line on the whole industry: "A more honest version of the pitch would be: hire someone competent at writing for the web. That sentence does not fit on a pricing page."

The symptom beneath the symptoms

What these three conversations share is not the specific technical question of how AI search works, or whether zero-click is a strategy, or whether schema markup matters. They share an exposed assumption that has quietly governed a large portion of the marketing industry for decades: that there is a lever, and that if you identify it and pull it correctly, visibility follows.

Classical SEO was built on this assumption. The lever was keyword density, then backlinks, then technical site structure, then content volume, then structured data. Each iteration was a new lever theory, and each one worked until the engine adjusted, or the technique scaled to the point of producing exactly the kind of content the engine was trying to filter out.

The cycle goes: vendor sells the controllable-discipline frame, agencies adopt it, content teams scale production around the prescriptions, AI-generated articles get pumped out at volume because the prescriptions are easy to template. Some of it ranks for a while. Most of it tanks because the prescriptions were never the mechanism. GEO and AEO are the current cycle. The pillars and percentages and pyramids are this cycle's templates.

The panic of 2026 is not a crisis of search. It is the belated recognition that the lever was never as connected to the outcome as the dashboard suggested.

What this means for brands that are not Condé Nast

The honest answer to the zero-click question, to the GEO/AEO question, and to the Google update question is the same answer, and it is the one the industry finds least useful to sell: create something people find genuinely worth knowing about, and do it consistently enough that traces of your existence accumulate across the places where AI systems and human beings look for things.

Dias identifies the technical correlate of this precisely: write well, cite evidence, be specific, build authority through demonstrated expertise over time. The GEO paper's findings, stripped of the vendor layer, say exactly this. Google's guidance, stripped of the search console implications, says exactly this. Fishkin's zero-click argument, stripped of the tactical prescriptions, says exactly this.

The brands that accumulate genuine visibility in AI-mediated search are not the ones that optimised their schema markup for LLM consumption. They are the ones whose names appear in the places AI systems look because other people put them there: in discussions, in citations, in recommendations, in content written about the brand rather than by it.

Brand building was never optional. The lever theory just made it seem that way for a while.

‍

Photo by Shantanu Kumar: https://www.pexels.com/photo/smart-phone-with-a-google-search-page-on-its-display-16564263/

‍

References

• Dias, P. (2026). The Whole Point Was the Mess. The Inference. Link
• Fishkin, R. (2025). Why Do We Need Zero Click Marketing? SparkToro. Link
• Fishkin, R. & Natividad, A. (2026). Zero Click Marketing: How to Build Brands and Earn Customers on a Traffic-Starved Web. Damn Gravity (forthcoming). Link
• Datos & SparkToro. (2026). State of Search Q4 2025. Link
• Google Search Central. (2026). Optimizing for Generative AI Features. Google Developers. Link
• Aggarwal, S. et al. (2024). GEO: Generative Engine Optimization. KDD 2024. Link

‍