Answer Engines Redefine Search

Google currently has about a 90% share of the overall search market and has dominated the industry for over 20 years. A kink in that dominance appeared in November 2022, when OpenAI introduced ChatGPT, inaugurating the era of AI search. The popularity of ChatGPT propelled the integration of generative AI into search and search summaries, and soon competitors like Microsoft Copilot, Perplexity AI, Anthropic’s Claude, and xAI’s Grok were launched and began to gain traction in search as well. In short order, artificial intelligence was redefining not only online search, but also how people find, process, and engage with digital information.

These new AI search engines, also called answer engines, leverage large language models (LLMs) to deliver direct, conversational answers to complex queries, support multi-step reasoning, and enable users to refine their searches interactively. This evolution moved traditional search from keyword-matching toward AI systems that understand context, intent, and multimodal input, shifting the user experience from sifting through links to engaging in a dialogue with AI.

Traditional search engines primarily work by indexing and retrieving webpages that are most relevant to a user’s keyword query. They provide a ranked list of links and the user is typically expected to sift through these search results, visit external websites, and synthesize information themselves to fully resolve their question.

In contrast, generative answer engines leverage LLMs to directly interpret user queries in natural language and dynamically generate concise, synthesized answers on the spot. Instead of merely listing sources that users click on and explore as in traditional search, AI search engines pull relevant information from multiple contexts, and assemble brief, self-contained answers on the fly. Answer engines still cite sources, but since they provide a comprehensible answer, the need for users to visit the sites from which the information has been gathered is greatly reduced.

In response to the encroachment of answer engines, Google made some changes itself. In May 2024, Google launched AI Overviews (AIOs), a search feature that provides users with AI-generated summaries in response to their queries, which usually appear at the very top of a search results page, where the highest-paying advertisers are usually displayed when Google returns traditional search results. Similar to answer engines, these overviews synthesize information from multiple sources using Google’s generative AI technology, specifically the Gemini language model, to deliver concise, informative answers along with links to relevant websites for further reading. The goal of AI Overviews is to give users immediate, comprehensive responses to complex queries, reducing the need to click through multiple search results, exactly like answer engines.

“AI Overviews are scaling fast,” said Nick Eubanks, vice president at Semrush, a digital marketing platform. “In May 2025, Semrush data showed AIOs appeared in nearly 20% of all Google queries, up from just 6% in January. Growth has been especially sharp in categories like travel, health, and science; travel alone saw a 3,100%+ increase in AI Overview appearances since September 2024.”

In a Q2 2025 earnings call with investors, Alphabet CEO Sundar Pichai said AI Overviews now have over two billion monthly users, up from the 1.5 billion users reported in Q1.

“Search is no longer ‘one size fits all’,” Eubanks said. “With AI search, users now see conversational answers, subtopics, and visual tools instead of a page full of links. It’s designed to deliver faster, more direct responses while keeping users on the platform. AI search engines like ChatGPT and Perplexity are pushing those expectations even further. They’re built around understanding intent and context, not just matching keywords.”

How Answer Engines Threaten Publisher Traffic

A study by Pew Research Center showed that just 1% of Google users who encountered an AI Overview clicked on a link in the summary itself. Pew found around two-thirds of users either browsed elsewhere on Google or left the site entirely without clicking a link in the search results after viewing an AIO. After viewing a search page with an AI Overview, 26% of users ended their browsing session entirely after reading the AI summary, according to the study.

If it sounds like generative AI search summaries are choking off Web referral traffic to online publishers, that is because they are. According to data from TollBit, a platform that enables publishers to monetize and license their content for use by AI bots and applications, AI search engines drove just 0.04% of total external referral traffic to publishers’ sites in the first quarter of 2025, while Google’s traditional search engine drove 85% of external referral traffic to these sites.

Web referral traffic from search engines is the lifeblood of online content publishers, because it directly fuels ad revenue from visits to publishers’ websites, as well as audience growth and engagement. This flow of inbound traffic not only drives ad impressions and subscription opportunities, but also signals content relevance, helping to boost rankings in search algorithms. Without the steady stream of referrals, the related ad revenue for publishers is at risk.

As a result, publishers are intensifying efforts to prevent unauthorized use of their content by AI search engines, which are aggregating and summarizing vast amounts of data from publishers’ sites. In response to declining website traffic and loss of advertising revenue, publishers have pursued both legal action and licensing agreements to protect and monetize their intellectual property. Meanwhile, technical measures—such as blocking AI web crawlers and using AI monetization services like TollBit—are being put in place to halt or regulate AI access to publisher content.

There is also a question of whether AI firms can freely use publishers’ premium content to train their models without direct compensation, a dispute with potentially far-reaching consequences for the media industry and the broader Internet. Some publishers have struck deals; The Atlantic and Dotdash Meredith, publisher of People magazine, have negotiated agreements with OpenAI. Lawsuits, such as The New York Times’ case against OpenAI and Microsoft, are ongoing, even as The New York Times struck a deal with Amazon to license its content across Amazon’s platforms. Washington Post owner Jeff Bezos, who is still the largest individual shareholder of Amazon, licensed the Post’s content to OpenAI, allowing its content to be used by ChatGPT. OpenAI has also inked media deals with the Associated Press, the Financial Times, and The Wall Street Journal, among other outlets, so publishers are taking steps to capture lost revenue.

Generative AI and Information Retrieval

Content publishing, online advertising, and search engine optimization are not the only markets being roiled by AI search. The information retrieval field is being upended as well.

“I recently attended the ACM Special Interest Group on Information Retrieval Annual [SIGIR] Conference, and one prominent theme that emerged is retrieval-augmented generation,” said Ryen White, vice president of applied science at Microsoft Research.

Retrieval-augmented generation, or RAG, is an AI technique that combines information retrieval from external sources with large language model text generation, allowing the model to generate responses grounded with relevant information beyond its original training data. RAG plays a significant role in ensuring language models remain current, and a considerable portion of research at the recent SIGIR Conference addressed enhancements to RAG methods.

“Google indexes might crawl a publisher’s articles 1,000 times a day,” said Toshit Panigrahi, TollBit’s CEO and co-founder. “However, with AI applications, the same articles might get accessed tens of thousands of times.”

Panigrahi explained that the correlation observed on the TollBit network between indexing patterns around current events and traffic spikes from AI retrieval-augmented generation bots meant the AI RAG-bots were not just indexing but actually responding to user queries in real time.

“In order to answer a search question, AI applications need to load the articles into context,” Panigrahi said. “It’s just not enough to index the articles anymore.”

From Answer Engines to Action Engines

“Human usage of search engines may have reached its peak,” Microsoft’s White said. Search engine outputs are being utilized increasingly by AI agents, where the results serve as inputs for LLMs to perform downstream tasks such as providing answers, interpreting tasks, or executing actions. This trend represents a decline in the necessity for direct human interaction with search engines, particularly as AI continues to assume a broader range of responsibilities previously managed by users.

“Traditionally, search engines present a ranked list—often 10 blue links—tailored to human cognitive limitations,” White said. However, he said, if the primary consumer of such content becomes an AI agent rather than a person, these constraints become less relevant.

“For instance, an LLM could process hundreds, thousands, or even millions of documents without requiring ranked results. This transition raises several important questions about how search systems should be designed when their primary audience is no longer human,” White said. The role of search engines is further changing in the context of their integration into larger AI-driven processes, which includes generating comprehensive answers, performing complex actions, and other advanced functions.

Ultimately, “We are presently evolving from search engines to answer engines,” White said. “The next step is evolving from answer engines to action engines. These are effectively agents that can perform either the whole task or aspects of the task on a user’s behalf.”

White said rather than just mapping queries to documents, modern search now translates intentions into decisions and actions, thereby broadening the scope of search technology.

The Rise of Answer Engines

The search industry is at an inflection point as answer engines are transforming the user experience through conversational interactions, greater personalization, synthesized answers, and eventually, agentic capabilities that complete tasks autonomously.

As more users bypass traditional search engines and use AI-powered search as their default method for seeking information, answer engines could supplant traditional search as the primary means by which information will be accessed in the future. Semrush, for instance, expects AI search visitors will surpass traditional search visitors in 2028.

Traditional search methods will not entirely disappear—Google will not be going the way of Lycos, AltaVista, or Ask Jeeves anytime soon—but answer engines powered by generative AI will become the first choice for search.

John Delaney is a freelance technology writer based in Hurley, NY, USA