State of
AI Search
2026
as we enter september 2026.
How AI-generated answers rewrote the rules of business visibility, and what the summer's measurement breakthroughs, ChatGPT's ad expansion, and a new gap between citation and recommendation mean for your brand right now.
of queries in Google's AI Mode end without a click.
You can rank #1 and still be completely invisible to the buyer.
The Data That Changes the Maththe gap between visible and chosen.
The gap between being visible in AI search and being chosen by it has widened into the defining story of the second half of 2026.
The Tools You Asked For Are Finally Hereno more measuring in the dark.
For most of 2026, brands have been told to measure AI visibility with no real instrumentation to do it. That changed over the summer.
Google Search Console: AI Impression Data
Announced June 3, 2026 and rolled out to all sites worldwide by the end of August, Search Console now reports impressions specifically from AI Overviews and AI Mode, not just traditional organic results. For the first time, you can see how often your content actually surfaces inside an AI answer, not just infer it from traffic drops.
Bing Webmaster Tools: AI Citation Share
Microsoft shipped a parallel metric in Bing Webmaster Tools the same month: an AI Citation Share score (your citations divided by total citations for a given query) showing how often your domain is cited relative to competitors on a topic.
The "wire up measurement now" recommendation from our last report is no longer aspirational. These tools exist today. If you haven't connected them, that is now the single fastest gap to close.
The Crawlers Are Being Sorted, and So Is the Spamquality and access, both tightening.
Two parallel developments from the summer change how you should think about both content quality and crawler access.
Spam Updates Have Tripled
Google shipped its third confirmed spam update of 2026 (March, June, and August 18–21) in the time all of 2025 saw one (August 26–September 22, 2025). The target is consistent with our last report's warning: thin, templated, AI-generated content built to rank a keyword rather than answer a question. If your content strategy leans on volume over depth, this is the year that stops working.
Crawler Access Just Got More Complicated
Cloudflare changed its default bot-blocking rules for new domains, effective September 15, 2026: on any page running ads, AI Agent and AI Training bots are now blocked by default, while Search and retrieval crawlers remain allowed. This is a meaningful nuance to our earlier "don't block AI crawlers" guidance: the crawlers that power AI answers are a different category from the bots that scrape for model training, and the distinction now matters at the infrastructure level, not just in your own robots.txt. Review your settings, and your host's or CDN's defaults, with this distinction specifically in mind.
The Google Earthquakewhere it started, five months ago.
Sundar Pichai called it "the biggest change in more than 25 years" of Google Search. For once, the hyperbole was earned. Here is what changed at I/O 2026 and what it means for brand visibility.
The Search Box Is Completely Reimagined
Google's traditional search box has been replaced by an intelligent input that expands as you type, anticipates intent, and surfaces AI-powered suggestions that go well beyond autocomplete. The redesign reflects behavioral data Google itself confirmed: AI Mode queries are now three times longer than traditional Google searches, and follow-up queries are up 40% month over month. People are having conversations with search, not typing keywords.
Gemini 3.5 Flash Is Now the Default Engine Globally
Starting at I/O, Google made Gemini 3.5 Flash the default model powering AI Mode for all users worldwide. Described as delivering "sustained frontier performance for agents and coding" at four times the output token speed of comparable frontier models, this is not a minor model bump. The underlying engine analyzing and synthesizing content for AI Mode is now faster, more capable, and more deeply integrated into every search interaction.
AI Mode Is No Longer Experimental
Google confirmed at I/O that AI Mode has graduated from Search Labs. It is now a mainstream product, not an opt-in experiment. AI Overviews reach more than 2.5 billion users globally. AI Mode has crossed 1 billion monthly active users. AI-mediated search is the default experience for the majority of the world's internet users.
The May 6 Structural Update to AI Overviews
Two weeks before I/O, Google quietly rolled out its most significant AI Overviews update since the feature launched. Inline citations now appear next to the specific text they support rather than clustered at the bottom. Hover previews on desktop now show site names directly. A new "Expert Advice" block pulls first-hand perspectives from forums, social media, and review sites. AI Overviews now appear in roughly 25.8% of all U.S. searches and in approximately half of all informational queries. Organic click-through rates where AI Overviews appear have dropped by as much as 61%.
Agentic Search: Gemini Spark and Background Tasks
Google introduced Gemini Spark, an AI agent that can work on tasks in the background even when the user's device is closed, and deliver synthesized updates when conditions are met. Information agents in Search can now monitor topics, track market movements with custom parameters, and alert users proactively. Users will increasingly delegate research tasks to AI agents rather than performing searches themselves. The implications for content discoverability and the entire concept of "driving traffic" are still being worked out, but they are significant.
Antigravity 2.0 and Custom Mini-Apps in Search
Antigravity 2.0 is Google's agent-first development platform. Within Search, it powers Custom Mini-Apps: interactive, task-specific interfaces that can be generated on demand directly inside the search results page. A user searching for a mortgage calculator, a trip planner, or a comparison tool no longer needs to click through to a website. Search builds it for them.
Universal Cart and Agentic Commerce
Google announced Universal Cart, described as "a truly intelligent shopping cart" that aggregates products across retailers into a single checkout experience within Search. This is Google's answer to the commerce challenge posed by ChatGPT's shopping integrations and Perplexity's PayPal-powered in-chat purchasing. The shopping journey is being absorbed into the search interface itself.
May 2026 Core Update: Rewarding AI-Ready Contentthe e-e-a-t filter got sharper.
Running in parallel with I/O was a Google May 2026 Core Update sending significant ranking fluctuations across verticals. The signal is clear: Google is accelerating its preference for content that demonstrates genuine expertise, real-world experience, and structural clarity that AI systems can parse and cite.
The sites experiencing the most volatility share common weaknesses: thin product pages without substantive reviews, templated copy, weak brand presence across the web, and content that was built to rank a keyword rather than to answer a question authoritatively.
The update reinforces a pattern that has been building for twelve months. E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) is no longer a content quality guideline. It is an AI-citation filter. If an AI system cannot confidently extract and attribute your content, it will not.
Volatility is highest for thin product pages, templated copy, and content built to rank keywords rather than answer questions. The sites gaining ground share one characteristic: their content is structured for extraction, not just discovery.
ChatGPT Advertising: From Pilot to $1 Billionaccelerated faster than anyone predicted.
Our last report covered ChatGPT ads through the May 5 self-serve launch. What's happened since is the story.
ChatGPT ads officially launched for U.S. users on Free and Go tiers, available initially through a managed enterprise pilot with a $250,000 minimum spend.
OpenAI opened self-serve buying through the OpenAI Ads Manager, added cost-per-click bidding alongside CPM, and launched a measurement pixel and Conversions API. The minimum spend dropped dramatically.
OpenAI shipped eight ad-platform updates in a single week: conversion-optimized bidding (oCPC), automatic advanced matching, a Bulk API, richer product cards showing price and star ratings, and intraday budget pacing.
Ads launched in the UK, Mexico, Brazil, Japan, and South Korea.
Self-serve access opened across 31 European markets, plus India, the Middle East, and North Africa. The same day, OpenAI confirmed ChatGPT Ads had reached $1 billion in annualized revenue run rate, in under 200 days from launch.
Minimum spend reduced from $250,000 to $50,000, then eliminated for smaller advertisers through the May self-serve Ads Manager launch.
Ads appear as clearly labeled "Sponsored" blocks beneath organic AI responses. They never influence what ChatGPT actually says in its answers. Formats include a 40-character headline, 150-character description, and clickable link. Some product queries trigger a shopping card format showing price and availability.
Targeting is intent-based and contextual, built on the current conversation's topic, chat history, and prior ad interactions. Critically, advertisers receive no user data: no names, emails, IP addresses, or precise locations. Generic brand messaging performs poorly here. Intent-matched, specific offers win.
This is no longer a pilot channel to watch cautiously. It's a global, self-serve, increasingly sophisticated ad platform with real budget behind it. If your paid strategy hasn't included a ChatGPT Ads test yet, the "wait and see" window has closed.
Perplexity: Growth, and Growing Scrutinythe platform keeps shipping, the risks are real too.
Perplexity's product velocity continued through the summer, but the full picture now includes real headwinds worth knowing about.
Perplexity Computer (February 2026)
Perplexity's most consequential launch of the year is an agentic AI designed to complete complex multi-step tasks with limited human supervision. The financial impact was immediate: ARR topped $450 million in March and an estimated $500 million in April, a 335% year-over-year increase from roughly $148 million in mid-2025.
The Comet Browser Goes Global
Perplexity's Comet browser, built on Chromium with Perplexity search natively integrated, launched for free worldwide download after a phased rollout. It is now available on Windows, macOS, Android, and iOS. For GEO practitioners, Comet matters because it creates a direct-to-user channel that bypasses Google's search index entirely. Users browsing in Comet are getting Perplexity-powered answers at the browser level.
Commerce Integrations
Perplexity launched in-chat shopping powered by a PayPal integration, allowing U.S. users to purchase products directly within a Perplexity conversation. Combined with the Snapchat partnership, Perplexity is expanding its surface area far beyond its own app.
The Unified AI Workspace
In May 2026, Perplexity debuted a unified AI workspace that coordinates 19 models to plan, delegate, and execute projects from a single conversation. The Sonar API, the developer-facing side of Perplexity's search infrastructure, has become one of the most widely adopted AI search APIs, powering at-scale content generation and customer support tools embedded in third-party products.
Continued Product Expansion Through the Summer
Personal Computer, Perplexity's agentic assistant, expanded to Windows on July 28, reaching well over a billion devices, though the feature remains limited to Max and Enterprise Max subscribers.
Not every market is accelerating. Multiple trackers reported Perplexity's app downloads in India fell more than 90% between February and July 2026, after a free-Pro promotional partnership with telecom carrier Airtel lapsed in January. Growth in the U.S. and product headlines shouldn't be read as universal, and Perplexity itself hasn't issued a public statement on the India numbers.
Perplexity is currently defending separate active suits from multiple major publishers, including Dow Jones, the New York Post, Britannica, Reddit, the New York Times, and CNN. In the Reddit case, a judge denied most of Perplexity's motion to dismiss on July 31, 2026, allowing Reddit's core DMCA anti-circumvention claims to proceed while dismissing some other claims. For brands, this matters less as a legal footnote and more as a real question: how a platform sources and licenses the content it cites is now an active, contested issue, not a settled practice.
Claude: The Quiet Powerhouse for Small Businessstill holding up, unchanged.
While Google and OpenAI dominate the headlines, Anthropic's Claude has become one of the most practically useful AI tools for small and mid-sized businesses, not primarily as a search engine, but as a day-to-day business assistant that is increasingly shaping how people find and evaluate brands.
Claude is used heavily for research, drafting, analysis, and decision-making. When a small business owner asks Claude to recommend accounting software, compare local contractors, or explain what to look for in a marketing agency, the brands Claude surfaces and describes become part of that person's consideration set. That is a GEO opportunity most small businesses have not woken up to yet. No Google search required.
From a practical standpoint, Claude is also one of the most accessible AI tools for businesses without large technical teams. It handles customer email drafts, proposal writing, competitive research, content outlines, and internal documentation without requiring any setup or integration. The barrier to getting real value out of it is low, and the time savings are immediate.
First, make sure your brand is represented accurately and positively across the web so that when Claude researches your category, it finds the right information. Second, consider using Claude directly as a business tool. The businesses getting ahead in 2026 are not just optimizing to be found by AI. They are using AI every day to work faster and serve customers better.
AIO, AEO, GEO, and AI SEO in 2026citation and recommendation, redefined.
The terminology has settled. The community has largely converged on four overlapping but distinct optimization disciplines. Understanding how they interact is now essential for any brand with a digital presence.
SEO: Still the Foundation, Now AI-Pressured
Technical site health, crawlability, site architecture, page speed, internal linking, and backlinks remain the necessary foundation. AI crawlers use the same sitemaps that traditional bots have relied on for years. A structurally broken site will fail in AI search the same way it fails in traditional search. What has changed is the interpretation layer: AI systems no longer scan pages for keyword density. They evaluate meaning, context, and how well a topic is explained.
AIO: AI Overview Optimization
AIO refers specifically to optimizing content to appear within Google's AI Overviews and AI Mode responses. The May 6 update made the citation dynamics clearer than ever: content needs to be structurally parseable, clearly attributed to an authoritative entity, and specific enough that the AI can extract and quote it with confidence. The new "Expert Advice" block is a new AIO target: community content, forum posts, and practitioner perspectives are now being surfaced alongside editorial content.
AEO: Answer Engine Optimization
AEO is the tactical discipline of becoming the single source of truth for a specific answer. It targets featured snippets, People Also Ask boxes, voice search responses, and zero-click results. AEO-optimized content is structured to answer questions completely and immediately in a format that can be extracted without context. The distinction between AEO and AIO: AEO optimizes for the answer being extracted; AIO optimizes for the brand being selected and cited.
GEO: Generative Engine Optimization
GEO is the broadest discipline: ensuring your brand is cited, mentioned, and trusted across all generative engines, including ChatGPT, Perplexity, Gemini, Claude, and whatever emerges next. GEO prioritizes citation over ranking. The new signal is citation worthiness: can an AI extract your content cleanly, attribute it confidently, and present it as a trustworthy source? The most important GEO diagnostic in 2026 is to test how your brand is represented in each major AI platform directly. Ask about your product category. See if you appear, and if the description is accurate. That is your GEO baseline.
GEO's citation-worthiness test now needs a second question alongside "can an AI extract and attribute my content?" — namely, "when this AI is asked to recommend a brand in my category, do I actually get named?" Section 01's up-to-30x gap data confirms these are separate outcomes: optimizing for one does not guarantee the other.
AI SEO: The Integration Layer
AI SEO is the umbrella term for the integrated practice of optimizing for all surfaces simultaneously. The companies winning in 2026 run all four disciplines in parallel, with unified measurement that tracks organic traffic, AI Overview citations, voice and featured snippet captures, and AI-referred conversions as separate signals. Traditional metrics no longer tell the full story. Brands need referrer segments for AI sources and direct attribution surveys, because direct clicks alone are no longer representative of AI-assisted discovery.
What to Do for the Rest of 2026the rest of 2026, mapped.
The developments of the last nine months point to a clear set of priorities. These are not theoretical. They are the actions that determine whether your brand earns a place in AI-generated answers or is excluded from the consideration set.
Audit Your AI Citation Baseline
Test your brand across Google AI Mode, ChatGPT, Perplexity, and Gemini. Understand what each says about you, how often you appear, and whether the descriptions are accurate. This is now as important as a keyword ranking audit.
Build Citation-Worthy Content, Not Just Rankable Content
The goal is no longer to get a page to position one. The goal is to have content that AI systems choose to extract, quote, and attribute. That means clear structure, authoritative sourcing, consistent entity signals, and genuine depth, not length.
Do Not Block AI Crawlers
Review your robots.txt and your CDN/host's default bot rules. Ensure OAI-SearchBot, PerplexityBot, and Google's Gemini crawlers, the bots that power AI answers, have access. New infrastructure-level defaults (see Section 03) increasingly distinguish these from AI training bots; know which category your rules affect before you change anything.
Wire Up Measurement
This is no longer a "get ahead of it" recommendation, it's now table stakes. Google Search Console's AI impression data and Bing's AI Citation Share metric are live today (Section 02). If you haven't connected both, that's the first thing to fix.
Think About the Agentic Future
AI agents are starting to complete tasks for people: researching, comparing, and buying on their behalf. Google, Perplexity, and OpenAI all have agent products live or in testing right now. The question is no longer whether this shift is coming. The question is whether your brand is one the agents will choose.
Test for Recommendation, Not Just Citation
Being quoted in an AI answer and being the brand an AI recommends are different outcomes (Section 01). Add a recurring test to your audit process: ask each platform directly, "Which brands do you recommend for [category]?" not just "What is [my brand]?" Track whether you're named, not just whether you're cited.
Start With Reality,
Not Assumptions
Before you optimize anything, you need to know where you actually stand - not in traditional search rankings, but in AI-generated answers across the platforms where your buyers are researching right now.
Start optimizing for questions. You cannot fix what you do not measure.
Previous Versions
Each version is preserved as published. New versions are issued when the landscape changes significantly enough to warrant a full revision.