In the US, 68% of Google searches now end without a click, up from 60.5% in 2024. That volume hasn’t disappeared: it’s resolving inside the results page or a chat window before anyone reaches a website, and the traffic that still arrives converts far better than it used to.
Every marketing team is asking some version of the same question: is AI killing search? The honest answer is more useful than yes or no. Google is absorbing AI into how it answers queries, not losing them to a competitor, and that shift changes which signals are worth building for.
Search isn’t shrinking. It’s being answered before it’s clicked
The headline fear, that AI chatbots would simply replace Google, hasn’t played out literally. What’s happening instead is more disruptive for marketers: the same search volume increasingly resolves inside the results page or the chat window, never reaching a website at all.
In February 2024, Gartner forecast a 25% drop in traditional search volume by 2026 as AI chatbots absorbed queries. That specific number didn’t materialize. Google still holds well over 90% of search market share, and ChatGPT’s roughly 880 million monthly users have largely added a new destination rather than replacing the old one. The real shift is structural, not volumetric.
SparkToro’s Rand Fishkin has tracked the zero-click share of US Google searches from roughly 50% in 2019 to over 68% in early 2026, and BrightEdge data shows AI Overviews now trigger on close to half of all tracked queries, up 58% year over year.
At the same time, the traffic that does arrive from AI sources is proving unusually valuable. Adobe’s analysis of over a trillion US retail site visits found AI-referred traffic up 393% year over year in Q1 2026, and 138% in May alone. That traffic converts about 42% better than non-AI traffic, with visitors spending 53% longer on-site and viewing 23% more pages. Fewer visits, but each one is a more qualified visitor who has already been pre-sold by an AI answer.
The tag stack is being rewritten, not deleted
Marketers keep asking whether a new “AI schema” or a magic file will get them cited. Google’s own guidance is blunt: no such tag exists. What’s changing is which of the existing signals still carry weight, and which have quietly gone the way of <meta name="keywords">.
| Signal | Status in 2026 | What’s actually happening |
|---|---|---|
<meta keywords> |
Dead | Ignored by Google since the 2000s, and cited by Google itself as the precedent for why self-declared AI tags won’t work either. |
llms.txt |
Inert for ranking | Google’s John Mueller confirmed Search doesn’t use llms.txt for ranking or AI features. Ahrefs found 97% of published llms.txt files get zero requests. The narrow exception is developer-docs sites, which use it to help AI coding assistants navigate a known codebase. |
<title> / meta description |
Still functional, less controllable | Google now writes its own description roughly two-thirds of the time, and is testing Gemini-generated snippets. Write these for accuracy and clarity, not keyword density: you’re increasingly writing raw material for a rewrite, not the final snippet. |
Schema.org (FAQPage, Product, Organization) |
Still core | No dedicated “AI Overviews schema” exists, but structured data still sharpens entity understanding and rich-result eligibility. AI features draw from the same index and quality signals as standard Search. |
| Open Graph / social meta tags | Rising | AI systems are increasingly lifting OG titles and descriptions directly into generated answers and previews when a page is cited, a secondary but growing signal for how content gets summarized. |
| robots.txt AI crawler rules | New must-check | GPTBot, ClaudeBot, PerplexityBot, Google-Extended and Applebot-Extended each need explicit allowance. The best content in the world is invisible to a model that was never allowed to fetch it. |
| E-E-A-T: named experts, original data, sourcing | Rising fastest | This is the real currency of the shift. Content built on original data, named expertise and verifiable sourcing is what gets extracted and cited. Markup can’t substitute for it. |
| Third-party citations / digital PR | Rising sharply | A brand fact repeated in a high-authority outlet that an AI model actually retrieves from now outweighs the same claim on the brand’s own site. Digital PR is becoming an AI-visibility lever, not just an awareness one. |
What the research is showing
Four independent data sets, enterprise adoption, consumer sentiment, retail traffic and brand strategy, all point the same direction, with one shared warning: adoption is outrunning measurable return.
| Source | Headline finding | The catch |
|---|---|---|
| McKinsey, 2025-26 surveys | 79% of organizations now use generative AI, up from 33% in 2023. AI could add an estimated $463B in marketing productivity, over 60% of it from agentic AI specifically. | Only about 6% of organizations qualify as “high performers” actually capturing bottom-line value. CMOs ranked gen AI just 17th of 20 priorities for 2026, despite calling it a top-3 growth investment. |
| NielsenIQ, Consumer Outlook 2026 | 42% of consumers now use AI tools somewhere in their shopping journey. | Trust is softening: the share who find AI search “more helpful” than traditional search fell from 82% to 54% in twelve months. NIQ frames 2026 consumers as cautious by default, which raises the bar for AI answers to be accurate, not just fast. |
| Adobe Digital Insights, Q2 2026 | AI-referred traffic to US retail sites is up roughly 14x since Adobe began tracking it in October 2024, and converts meaningfully better than traditional channels. | Most retail sites still aren’t structured to be easily read by AI systems. The traffic reward exists well ahead of most brands’ technical readiness to earn it. |
| Harvard Business Review, 2025-26 | Brand strategy is shifting from persuading a person to being accurately understood by a model, introducing “Share of Model” as a metric to sit alongside share of voice. | LLMs frequently misrepresent nuanced positioning (HBR’s example: luxury brands), meaning being cited isn’t automatically a win if the model gets the brand wrong. |
What practitioners are saying
Beyond the institutional research, the people building AI-visibility tooling and running search practices day to day are converging on a few blunt points.
Direct clicks are no longer the only currency. Brand awareness and influence on the platforms an audience already uses now count as search performance, whether or not they drive a visit.
— Rand Fishkin, SparkToro, on reframing zero-click search
There is no equivalent of Search Console for AI platforms yet. Attribution is the piece of this shift nobody has solved.
— Wil Reynolds, Seer, on the 2026 AI-search measurement gap
Success in AI search shouldn't be measured by appearing in one personalized prompt. It's better tracked as your share of inclusion across a whole topic area, grounded in real answers.
— Aleyda Solis, on redefining success metrics for the AI-search era
The through-line: crawlability and technical foundations still gate everything else, the old rank-tracking KPI set is running out of road, and the industry doesn’t yet have reliable tooling to close the loop. That’s exactly the gap new AI-visibility platforms like Profound, Peec AI and Semrush’s AI toolset are racing to fill, by tracking how often, how prominently, and how accurately a brand is named across model outputs.
AEO vs. GEO: one discipline, two layers, no settled name
Clients increasingly ask which one they need. The honest answer: there’s no academic or industry consensus distinguishing the terms as of mid-2026, and most practitioner content uses AEO and GEO interchangeably. Where a useful distinction does hold, it maps to two different jobs inside the same funnel, not two competing strategies.
| AEO — Answer Engine Optimization | GEO — Generative Engine Optimization | |
|---|---|---|
| Optimizes for | Being the extracted, single answer: featured snippets, People Also Ask, voice assistants, and the direct-answer boxes inside AI Overviews and AI Mode. | Being cited inside a synthesized, multi-source answer: ChatGPT, Perplexity, Gemini and Claude weaving a brand's facts into a generated response alongside competitors. |
| Mechanism | Concise Q&A structure, FAQ and HowTo schema, one clean answer per question. | E-E-A-T, original data and third-party citations. |
| Speed and defensibility | Mechanical and cheap to copy. Table stakes, not a moat. | Earned, not built. Slower to win, and harder for competitors to fast-follow. |
"SEO makes a page eligible, AEO makes the answer extractable, GEO makes the brand citeable."
Practically, these aren’t competing budget lines, they’re sequential. AEO is hygiene: schema, crawler access and clean answer structure that should already be in place. GEO is where durable advantage lives, because it draws on the same E-E-A-T and citation signals this data already flags as rising fastest and rising sharply, signals that compound and are far harder for a competitor to fast-follow than a snippet format. Kognis’ recommendation: budget AEO as a fixed cost, GEO as the growth investment.
The Kognis playbook: what to build now
In priority order, foundational fixes first, measurement and content investment second.
Audit AI crawler access
Confirm GPTBot, ClaudeBot, PerplexityBot, Google-Extended and Applebot-Extended aren't blocked in robots.txt. This is a five-minute check that silently disqualifies otherwise-strong content.
Keep structured data current, for the right reason
Maintain Organization, Product, FAQPage and Author schema because it sharpens entity clarity for the standard index AI features draw from, not because a hidden "AI schema" exists. Skip llms.txt beyond developer-docs use cases: it has no proven ranking value.
Write for extraction, not keyword density
Structure key claims as clean lists, comparison tables and clearly attributed statistics: content built to be lifted whole into an answer, anchored in original data and named expertise rather than volume of pages.
Split the budget: AEO as hygiene, GEO as growth investment
Treat direct-answer formatting (schema, concise Q&A, snippet targeting) as a fixed-cost baseline every page should already meet. Direct incremental budget toward GEO: original research, named-expert bylines, earned citations, the signals that actually compound and are hard for competitors to copy.
Redirect PR and link-building toward AI-retrieved sources
Earned mentions in outlets that AI models actually train on or retrieve from now double as a citation-visibility lever, not just a brand-awareness one. Digital PR becomes a GEO channel.
Stand up a parallel AI-visibility measurement layer
Track brand mention frequency, position and sentiment across ChatGPT, Gemini, Perplexity and Google's AI features alongside classic rank tracking, via tools such as Profound, Peec AI or Semrush's AI visibility suite, since no platform-native "Search Console for AI" exists yet.
Reset client KPIs before traffic dips do it for you
Sessions and clicks will structurally decline even for well-optimized content. Bring visibility, citation accuracy and sentiment into the reporting model now, so a lower session count doesn't read as a failed campaign.
Don't discard classic SEO fundamentals
Google's own guidance is that AI Overviews and AI Mode draw from the same organic index, judged by the same quality signals. GEO is a layer built on top of technical SEO and E-E-A-T, not a replacement for them.
None of this argues that search stopped mattering. It argues that the definition of a successful search result has split in two: one version still sends a visitor, the other just needs to get the brand's facts right inside someone else's answer. Brands that keep measuring only the first version will read a structural shift as a traffic crisis, right up until a competitor's citation shows up inside the answer instead of theirs.
Sources and methodology
We build visibility inside AI search, not just Google rankings
We audit crawler access, structure content for extraction, and build the citation signals that get a brand named inside ChatGPT, Perplexity and AI Overviews.
Book a discovery call