AI Search: How Artificial Intelligence Is Changing Search in 2026

Originally published on The Daily Flare.

KEY TAKEAWAYS

• Google’s AI search reaches over 2 billion monthly users, making AI Overviews the most widely used AI answering product in the world.
• ChatGPT search, Perplexity and Copilot Search in Bing answer questions with cited sources, so the source a query pulls from matters more than ever.
• Independent measurement shows AI answers cut outbound clicks sharply, so websites now compete for citation inside the answer, not just ranking below it.

AI search is replacing the classic list of blue links with direct answers. Google, OpenAI, Microsoft and Perplexity now answer full questions in plain language, with cited sources attached.

From links to answers: the core shift

Traditional search worked like a directory. You typed a few keywords, the engine returned ten links, and the work of reading, comparing and deciding sat with you. AI search flips that. The engine reads several pages for you, compares them, and hands you a finished answer with references attached.

Google started the mainstream shift in May 2024, when it launched AI Overviews widely across US search results. Instead of a page of links, many queries now show an AI-generated summary at the top of the results, built from a handful of cited sources. At Alphabet’s Q2 2025 earnings, CEO Sundar Pichai said AI Overviews had more than 2 billion monthly users in over 200 countries, which made it the most widely used AI answering product on the planet almost overnight.

The follow-up came in March 2025 with AI Mode, an experimental interface that gives fuller, more conversational answers with more links than the Overviews box. It first opened to Google One AI Premium subscribers, then to all US users over 18 in May 2025, and by October 2025 it had expanded to more than 40 countries and territories. AI Mode is a separate experience you opt into: ask a long, multi-part question and get a synthesized answer with follow-up options, instead of scrolling through pages of results.

What makes these systems tick under the hood is a technique Google calls query fan-out. One question gets broken into several subtopic searches that run in parallel, pulling in a broader cross-section of the web than a single query would reach. The model then assembles the answer from those retrieved pages. That architecture matters because it changes which pages get seen: the pages quoted in the answer get the attention, not necessarily the pages ranked first in the classic results.

How today’s AI search products compare

Google does not stand alone. Four distinct approaches to AI search have emerged, each with a different idea of what an answer should look like.

Methodology note: this comparison is based on each company’s official announcements through October 2026. Product behavior changes quickly; where third-party measurements are cited below (Pew Research Center, Chartbeat), they are named in the text. No market share or revenue figures are included because no official current numbers are available.

Perplexity, founded in 2022 and led by CEO Aravind Srinivas, was one of the first consumer products built around cited AI answers rather than a link list. Its answer engine runs a live web search for each question and writes a synthesized reply with numbered citations to its sources. In 2025 it added Deep Research, a longer mode that runs many searches and returns a sourced report, and Comet, its own AI browser, which launched for subscribers in July 2025 and became free to download worldwide that October. At Bloomberg Tech in June 2025, Srinivas said the platform handled about 780 million queries in May 2025, growing more than 20 percent month over month.

ChatGPT’s web search took a different route: it brought search into a chatbot people already used. OpenAI spent 2025 tightening it up. Shopping features arrived in April 2025, with visual product cards showing images, prices, ratings and direct links to buy, and results chosen independently rather than as ads. A September 2025 update targeted factual accuracy, shopping intent detection and answer formatting. At the GPT-5 launch that August, OpenAI said the new model was 45 percent less likely to contain factual errors than GPT-4o when web search was enabled. With 800 million weekly ChatGPT users as of October 2025, according to the company’s own developer event, even a small search feature inside ChatGPT reaches an enormous audience.

Microsoft’s answer is Copilot Search in Bing, launched April 4, 2025 by the Bing team. It blends traditional and generative search: for straightforward questions it produces concise answers with prominent citations, and for complex topics it builds digestible summaries of the key points with links embedded directly in the sentences. Two features stand out. “See all links” shows every source used in a summary, and “See reasoning” explains how the response was generated. Follow-up questions stay on one page, so an entire line of inquiry lives in a single interface.

All four share the same engine design, called retrieval-augmented generation. The model first retrieves live web pages, then writes the answer grounded in what it retrieved, citing the sources. This is the same family of techniques behind how AI is changing medicine, where models reason over research literature rather than just reciting memorized text. Retrieval is what keeps AI search honest: every claim should trace back to a page you could open yourself.

What AI search means for the web

The biggest consequence is visible in traffic numbers. When the answer arrives without a click, fewer people visit the pages behind it. A Pew Research Center study published in July 2025 found users clicked on sources cited directly in AI Overviews in just 1 percent of visits to pages containing those features. The consultancy Chartbeat reported that news publishers saw roughly a 40 percent drop in Google-origin visits between July 2025 and July 2026. Google has pushed back on the framing, saying its AI features send billions of clicks to websites every week, but the two claims describe different halves of the same picture: plenty of total clicks, far fewer per site.

Google has been adjusting. On May 6, 2026 it added five outbound link features to AI Mode and AI Overviews, including inline links placed next to the specific text they support and a section of first-hand perspectives with creator context. On October 15, 2026 the company went further in a post titled “More ways to connect with the web from Search”: inline links appear higher in responses more often, news carousels of timely articles are rolling out to more locations and languages, and the company is testing a dedicated section for official websites at the top of answers. It also said use of its Preferred Sources feature, which lets users pin the publishers they trust, grew more than sevenfold during 2026.

For everyday users, the change cuts both ways. On one hand, a question that once meant twenty minutes of open tabs now gets answered in seconds. Ask how a battery technology compares and you can get specs side by side, down to details like silicon-carbon phone batteries, without visiting five review sites. On the other hand, an answer is only as good as the sources the model picked. Two users can get different answers to the same question depending on what the retrieval layer found that day. The wise habit is to treat an AI answer as a briefing, not a verdict, and click through when the decision matters.

For websites, the game has changed from ranking to being cited. A page can rank first in the organic results and still not be quoted in the AI answer above it. The material that gets cited tends to share traits: clear, extractable answers to specific questions, named authors, original data, and first-hand experience that a model cannot manufacture. This echoes a shift seen across the field, where practitioners building with AI changing programming report that concrete, tested examples beat generic descriptions.

Looking ahead, the direction is clear. Search is becoming a conversation with follow-ups, not a one-shot query. Shopping is moving into chat with product cards and price tracking. Deep research modes are replacing hours of manual reading. The link list is not dead, it is becoming the machinery behind the answer rather than the answer itself. Whether that makes the web richer or flatter depends on one thing that has not changed: someone has to write the pages the answers come from.