Google Agentic Search Hits 75M Users and Mueller's 9 Canonical Override Scenarios

Two stories today reshape how SEOs think about visibility and indexing. Google's AI Mode — now at 75 million daily active users — expanded agentic restaurant booking to 8 new countries on April 10, 2026, turning booking aggregators into invisible backend pipes. Meanwhile, John Mueller published a detailed breakdown of the 9 specific scenarios where Google overrides your rel=canonical tag, finally explaining canonicalization failures that have puzzled technical SEOs for years.

Google AI Mode goes global: 75 million users and counting

Google AI Mode launched in the US in May 2025. Eight months later, it hit 75 million daily active users — a number Google disclosed in its Q4 2025 earnings. As of April 10, 2026, agentic restaurant booking expanded beyond the US to 8 additional countries: Australia, Canada, Hong Kong, India, New Zealand, Singapore, South Africa, and the United Kingdom.

The expansion isn't just geographic. Google added 8 booking platform partners as backend inventory providers: OpenTable, TheFork, SevenRooms, ResDiary, Mozrest, Foodhub, Dojo, and DesignMyNight. Users describe what they want in natural language — "quiet Italian place for 4, Saturday evening, walkable from King's Cross" — and the AI queries all 8 platforms simultaneously, surfaces available options, and completes the reservation without leaving Google.

Two engagement metrics stand out. First, AI Mode queries doubled quarter-over-quarter in Q3 2025. Second, sessions in AI Mode run 3× longer than traditional search queries. Users aren't just asking questions — they're completing multi-step workflows that previously required navigating to third-party sites.

The financial commitment behind this

Google's 2026 capital expenditure is projected at $175–185 billion, roughly 6× the $30 billion pre-AI spending level. Search revenue hit $63 billion in Q4 2025, with year-over-year growth accelerating from 10% (Q1) to 17% (Q4). The infrastructure constraints Pichai flagged — wafer production capacity, memory supply shortages, and data center permitting delays — suggest the scaling bottleneck is physical, not strategic.

Monetization timeline

• January 2026Direct Offers (advertising format) launched inside AI Mode.
• February 2026Shopping and travel ads integrated into AI Mode results.
• March 2026Universal Commerce Protocol (UCP) checkout deployed — agent-to-agent transactions. (Full UCP vs OpenAI breakdown)
• April 10, 2026Agentic restaurant booking expands to 8 countries with 8 partner platforms.
• May 19–20, 2026Google I/O — expected next wave of agentic features announced.

The aggregator extinction event

The structural impact of agentic booking isn't about restaurants. It's about what happens to every intermediary platform when Google intercepts user intent at the search layer and completes the transaction natively.

Consider the before and after:

OpenTable, TheFork, and the other 6 partners are now providing real-time inventory data through headless APIs. The user never visits their site. The aggregator's brand, their comparison UX, their upsell flows, their retargeting pixels — all bypassed. The competitive advantage shifts entirely to API speed, uptime, and data completeness.

What this means for local SEO

Local businesses that previously relied on aggregator visibility now need to think about two surfaces: traditional local pack results and agentic booking results. The requirements diverge:

• Structured data accuracy becomes non-negotiable — Google's agent queries structured data to populate booking options
• Real-time availability APIs that partner platforms can query at machine speed
• Google Business Profile completeness — the agent pulls cuisine type, price range, ambiance, and accessibility from GBP attributes
• Review velocity and sentiment — early evidence suggests the agent weights recent reviews heavily when matching "vibe" requests

WordPress 7.0 and the agent-ready web

The agentic search shift isn't limited to Google's ecosystem. WordPress 7.0, in development now, includes capabilities specifically designed to connect websites to AI agent systems. The stated goal: enable the transition from a human-centered web to an increasingly agent-centered web where every user has their own AI agent performing tasks on their behalf.

For SEOs, this means the technical requirements for visibility are expanding. A page optimized for human readers and traditional Googlebot isn't necessarily accessible to an AI agent that needs to complete a task, not just read content. Agent-accessible sites need:

• Machine-readable action endpoints — structured APIs or markup that an agent can invoke (book, buy, schedule, compare)
• Real-time state — agents can't work with cached inventory pages; they need live availability, pricing, and capacity
• Explicit capability declarations — agents need to know what actions a site supports without crawling every page

The UK search trend data illustrates the behavioral shift: "when to book a table" queries surged 140% in 2026 — but many of those queries now resolve entirely within AI Mode without generating a click to any external site.

Mueller's 9 scenarios: why Google ignores your canonical tags

In a Reddit thread this week, Google's John Mueller published the most detailed public explanation of canonicalization override behavior to date. The core message: rel=canonical is a hint, and Google's systems override it in at least 9 documented scenarios.

Mueller added a critical caveat: "These systems aren't perfect." (For more on how Google's crawl infrastructure is shifting, see why LLM bots now out-crawl Googlebot.) Canonicalization errors often resolve as Google's systems re-crawl and recognize content differences — but for sites with thousands of URLs, the interim impact on indexing and traffic can be severe.

The JavaScript rendering trap and the bot-challenge pitfall

Scenarios 7 and 8 deserve special attention because they affect the largest number of sites and are the hardest to diagnose without deliberate testing.

The JavaScript rendering failure (Scenario 8)

Modern JavaScript frameworks — React, Vue, Angular, Next.js with client-side rendering — typically serve a minimal HTML shell that JavaScript populates after load. If Google's Web Rendering Service (WRS) fails to execute the JavaScript (timeout, JS error, missing API response), it sees:

<html>
  <head><title>My App</title></head>
  <body>
    <div id="root"></div>
    <script src="/bundle.js"></script>
  </body>
</html>

That shell is identical across every URL on your site. Google's canonicalization system interprets all of them as duplicates and collapses them to a single canonical — often an arbitrary page. The result: most of your pages are de-indexed, and the canonical chosen may not even be your most important page.

The bot-challenge pitfall (Scenario 7)

This one is increasingly common as more sites deploy aggressive bot protection — and with Google's new spam enforcement policies, technical misconfigurations carry higher stakes than ever. If your Cloudflare, Akamai, Sucuri, or custom WAF configuration triggers a JavaScript challenge or CAPTCHA page for Googlebot, every URL on your site returns the same challenge HTML. Same problem as the JS failure: mass canonicalization to a single URL.

The insidious part: your monitoring sees human visitors loading the real page, so you don't realize Googlebot is getting blocked. The only symptom is a gradual indexing decline that's easy to attribute to other causes.

What to do this week

For agentic search readiness

For canonical hygiene

The bigger picture: search as execution layer

Today's two stories — agentic booking and canonical tag overrides — look unrelated on the surface. They're not. Both reflect the same underlying shift: Google is moving from linking to doing.

Agentic search means Google doesn't just point users to a booking site — it completes the booking. Canonicalization means Google doesn't just read your rel=canonical tag — it overrides it when its own analysis disagrees. In both cases, the common thread is Google asserting its own judgment over publisher intent.

For SEOs, the response is the same: verify what Google actually sees and does, don't assume it follows your instructions. Check rendered HTML, not source HTML. Check whether users complete transactions on your site, not just whether they land on it. The gap between what you tell Google and what Google decides to do is widening — and the only defense is systematic verification.

Google I/O 2026 on May 19–20 will likely announce the next wave of agentic features. If restaurant booking is the test case, expect travel, services, healthcare scheduling, and e-commerce checkout to follow. The playbook for visibility in agentic search is being written right now — by the sites that make their data agent-accessible first.

Frequently asked questions

About the author

Francisco Leon de Vivero

Francisco is VP of Growth at Growing Search and a global SEO expert with 15+ years of experience across enterprise, ecommerce, and international search. Former Head of Global SEO Framework at Shopify, speaker at SEonthebeach and UnGagged, and Canadian and European search awards judge.

LinkedIn · YouTube · Get in touch