I asked a friend last week how she found a local plumber. She answered: “I asked a chatbot.” That sentence used to sound like a punchline. It doesn’t anymore.
The past two years have quietly rewritten the rules of discovery. Google — long synonymous with “find it online” — is morphing from a link-ranking service into an AI agent that summarizes, decides and, in some cases, completes transactions for users. That change matters because it breaks the tidy funnel advertisers and publishers once relied on: query → click → page → conversion. Now the funnel can stop at the answer.
From ranked links to task-doers
SEO veterans saw it coming, but the speed surprised even them. Seasoned consultant Marie Haynes has traced the arc: Google’s signal set shifted toward user-satisfaction and expertise over the past decade, with big inflection points in 2017, the Helpful Content rollout in 2022, and dramatic AI-driven updates during 2024–2025. What used to be a list of links is becoming a conversational, multi-turn surface where Google’s AI can even make purchases for users (with permission).
Those agent capabilities are not abstract. Google has been testing features that let AI track prices and execute buys on a user’s behalf, and it has experimented with seamless transitions from AI summaries into a conversational “AI Mode.” The company is also folding deeper model-driven features into the product suite: if you want to see the direction Google is taking, look at the work around AI Mode and its agentic booking experiments and the move to plug Gemini-style research across Gmail and Drive (the kind of integration that lets an LLM find, summarize and act on documents across a user’s workspace). These changes shift value from page views to being the knowledge source an AI trusts.
Why clicks are evaporating (and why advertisers are alarmed)
A practical consequence: traffic that used to go to websites is being consumed at the platform layer. If an AI gives a satisfactory answer, fewer users click through. Combine that with the explosion of short-form video, messaging groups, and private communities — places where traditional display/search ads are hard to buy or measure — and you get a squeeze on ad effectiveness.
Small advertisers report steeper CPCs and weaker returns; big-budget players can still stomp the auction by out-spending everyone else. That creates a K-shaped outcome: well-resourced brands get more reach, smaller operators are forced to accept worse unit economics or find alternative channels. At the same time, some users are fleeing ad-laden, AI-generated rubbish toward privacy-minded or old-school search alternatives; others are simply asking chatbots first and never reaching a results page at all.
Attention has splintered — and so must marketing
People still look for things online, but the how and where have fractured. Some patterns worth noting:
- Short-form video and social feeds: discovery increasingly happens inside TikTok, Instagram Reels and YouTube Shorts, where the ad formats and creative rules differ wildly from search ads.\
- Private and pseudo-private channels: Discord servers, WhatsApp groups and closed forums are where recommendations circulate now — hard to reach via traditional programmatic buys.\
- Conversational search: chat-driven engines (and assistants plugged into email, docs and messaging) answer users directly. If your content doesn’t appear in the snippets an LLM relies on, you don’t exist to that user.
- Make genuinely useful, experience-rich content. Token author bios and recycled lists don’t cut it. Demonstrate firsthand knowledge, publish original data, or create tools and visual assets that are hard to replicate.\
- Think in fragments that agents can cite. Agents assemble answers from spans of content. Structure pages so they expose clear, labeled facts, Q&A blocks, and metadata that an extractor can use.\
- Grow direct signals. Email lists, repeat visitors, app users and local citations are currency. Those engagement metrics feed automated systems that determine trust.\
- Build for tasks, not just traffic. If Google’s agents can book a slot for a user, can your site offer the API or schema that lets an agent verify availability? Experiment with appointment APIs and lightweight integrations — the platforms are signaling preference for results that help machines act.\
- Invest in relationships and earned mentions. E‑E‑A‑T in an agent world is often about what respected voices say about you. Partner with local press, specialists and podcasts; an authoritative mention can become an agent’s citation.
At the same time, alternatives to a Google-only strategy are proliferating: privacy-focused engines, paid niche indexes, and AI-first competitors that serve sourced answers. The result is a multi-engine, multi-interface world.
What to do now — real, actionable moves
If you run a business that depends on online discovery, stop assuming “just keep doing SEO.” The models driving discovery have changed; so should your playbook.
Two product moves worth exploring: the agent marketplaces and integration points platforms are building. Think less about “ranking” and more about being a reliable data provider for the assistant layer — whether that means exposing structured product feeds, supporting ticketed booking flows, or enabling simple verification APIs.
The downside: accuracy, ads inside answers, and privacy
This new architecture is not a panacea. Agents hallucinate, and the temptation to monetize succinctly delivered answers is enormous. Expect ad formats to migrate into summaries and conversational outputs — perhaps subtly. There’s also the privacy dimension: agents that crawl your personal inbox or drive to answer questions raise new consent and data‑use questions.
That’s why businesses should diversify: own your customer relationships, surface verifiable assets (research, transcripts, photos), and treat platform exposure as one channel among several rather than the whole strategy.
A messy, multi-engine future
We’re midway through a chaotic rebalancing. Old assumptions — that search equals clicks, that links alone prove authority — are crumbling. The paths to discovery now run through feeds, agents, private rooms, and alternative indexes. None of that is stable yet; standards, marketplaces and policies are still being written. That means the window to experiment is wide.
If you are a marketer, product lead, or founder: start treating “appearing in assistants” as a product problem. If you’re a publisher, focus on unique value that an agent would need to quote. If you sell locally, invest in tooling that makes it trivial for a machine to confirm your availability and reputation.
This is not the death of search so much as a new kind of search — one that acts. And that’s a very different target to shoot at.
(For background on agentic booking experiments and the deeper Workspace integrations reshaping where answers come from, Google’s recent moves around agentic booking and AI Mode and the Gemini workspace integrations are good places to scan.)