Search has changed more in the last three years than in the previous twenty. The combination of large language models, retrieval, and real-time web access has produced a fundamentally new way to find information — and the implications, both good and bad, are still working themselves out.
The new search landscape
Google's AI Overviews, after a rocky launch, have stabilized into a genuinely useful feature for most informational queries. Perplexity has carved out a strong position as the default for users who want a thoughtful, cited answer instead of a list of links. OpenAI's ChatGPT Search and Anthropic's Claude with web access have made conversational search a mainstream behavior. Smaller players — You.com, Brave Search's Summarizer, Kagi's Ultimate — each have loyal user bases.
The result is that 'I searched for it' increasingly means 'I asked an AI,' especially for the question types AI is good at: how-to queries, comparison shopping, fact-finding, and exploratory research.
What AI search does well
For well-formed informational queries, the best AI search experiences are genuinely better than the traditional ten-blue-links interface. They synthesize across multiple sources, surface relevant context the user did not know to ask for, and answer the actual question rather than just pointing to pages that might contain the answer. For complex research, the ability to ask follow-up questions in context is transformative.
Where it still fails
AI search struggles with three categories. Navigational queries — 'I want to go to this specific site' — are still served better by traditional search. Real-time queries about breaking events are sometimes fast and sometimes hopelessly behind. And anything that requires deep engagement with primary sources — academic research, legal work, in-depth investigation — is still better served by traditional search followed by careful reading.
The leading products have largely figured out how to route these query types appropriately, but the user experience can still be confusing when the system picks the wrong mode.
The open web problem
The most consequential question hanging over AI search is what happens to the content ecosystem that makes it possible. If AI answers questions directly, fewer users click through to the sources, and the sites that produce the content lose the traffic that funds them. Publishers, knowledge bases, and independent creators are all feeling this pressure.
Several models are being tried. Direct licensing deals between AI providers and major publishers. Per-citation payment schemes. Improved attribution and click-through from AI answers. Subscription bundles. None of them has fully solved the underlying economic problem, and the long-term sustainability of the open web remains genuinely uncertain.
Implications for content creators
If you produce content for the web in 2026, the practical reality is that some fraction of your audience now consumes your work through an AI intermediary, often without ever visiting your site. The most effective response combines several strategies: write content that AI summarizers will cite and link to, build direct relationships with your audience through email and other owned channels, and ensure your most valuable content is behind some form of relationship — subscription, account, or community — that survives the disintermediation.
Implications for SEO
Traditional SEO is not dead, but it has been joined by a new discipline sometimes called GEO — generative engine optimization. Writing in a way that AI summarizers can easily understand and cite, providing clear structured data, and earning the kind of trust signals that AI providers use to rank sources are all becoming as important as classical link-building.
The good news is that most of the practices that make content visible to AI also make it useful to humans. The bad news is that gaming the system is now harder, faster-moving, and more opaque than it was in the classical SEO era.
Where this is going
AI search will continue to gain share, especially for the query types it serves well. Traditional search will continue to matter for everything else, and for the foreseeable future, the best products will integrate both. The open question is whether the economic model that supports the underlying content can evolve fast enough to keep producing the information that makes any of this work. That is the most important problem in the industry right now, and it is far from solved.