AI Is Moving Between Users and Original Sources
Contents
Generative AI systems increasingly provide direct answers to questions that previously required users to open websites, compare search results, or consult original documents. Google Search displays AI Overviews for queries when its systems determine that generative AI can provide a useful summary, while ChatGPT Search can produce answers containing citations to web sources. Google states that AI Overviews combine information from multiple sources and provide links for further investigation.
This changes the sequence through which information is accessed. Instead of moving directly from a search engine to a publisher, database, government page, company website, or other original source, a user can receive a synthesized answer first.
The change is also visible in internet infrastructure. Artificial intelligence has become commercially significant enough to affect domain-name demand, including registrations using the .ai domain extension. The .ai country-code top-level domain belongs to Anguilla but is widely used by artificial-intelligence companies and products.
AI Answers Reduce Visits to Source Pages
Research provides measurable evidence that AI-generated answers can change browsing behavior.
Pew Research Center analyzed 68,879 Google searches conducted by 900 U.S. adults in March 2025. AI-generated summaries appeared in approximately 18% of the searches examined.
The study identified several differences between searches with and without AI summaries:
- Users clicked a traditional search result in 8% of visits containing an AI summary.
- Users clicked a traditional search result in 15% of visits without an AI summary.
- Links included directly inside AI summaries received clicks in only 1% of visits containing those summaries.
- Browsing sessions ended after 26% of search pages containing an AI summary, compared with 16% of pages containing only traditional search results.
These figures document a reduction in direct interaction with source websites when an AI-generated answer appears.
The behavior is not limited to a small number of search users. Pew found that 58% of participants encountered at least one Google search containing an AI-generated summary during the March 2025 study period. A separate Pew survey conducted in February 2026 found that 60% of U.S. adults said they read AI summaries displayed at the top of search results.
Summaries Are Not Primary Sources
A primary source contains original evidence or information rather than a later interpretation of it. Examples include legislation, court decisions, scientific papers reporting original research, government statistics, company filings, interview transcripts, photographs, historical records, and statements issued directly by an organization.
An AI-generated response has a different function. It produces text from information processed by a model and, in some systems, information retrieved from external sources. The generated answer therefore represents an additional informational layer between the user and the underlying material.
That distinction applies even to ordinary practical questions. Information about social-media settings, for example, can be summarized by an AI system or explained by a secondary website such as this guide on how to hide Instagram followers. Neither format replaces Instagram’s own documentation as the primary authority on the platform’s current features and settings.
AI Can Preserve Sources Without Guaranteeing Accuracy
Modern AI information systems can display citations rather than presenting answers without attribution. ChatGPT Search can provide inline citations and a Sources panel, while Google describes AI Overviews as summaries containing links that allow users to investigate information further.
Citation availability does not establish that every generated statement is correct. OpenAI’s documentation states that search results and citations can be incomplete, outdated, or incorrect and recommends checking whether a cited source actually supports an answer. Google similarly states that AI Overviews can contain mistakes and advises checking important information in more than one place.
The verification process therefore contains separate factual questions:
- Does the cited page exist?
- Does it contain the information attributed to it?
- Is the information current?
- Is the source authoritative for the specific claim?
- Does the generated summary preserve qualifications, dates, definitions, and numerical context contained in the source?
A citation answers only the first question automatically when the link resolves successfully. The remaining questions require examination of the source.
Source Selection Also Shapes AI Answers
AI-generated search summaries do not draw equally from every available website. In Pew’s 2025 analysis, Wikipedia, YouTube, and Reddit collectively represented 15% of sources cited in Google AI summaries. Government websites represented 6% of cited AI Overview sources, while news websites represented 5%.
Google also states that pages must be indexed and eligible to appear in conventional Google Search with a snippet before they can qualify as supporting links in AI Overviews or AI Mode.
Consequently, information presented through AI search is affected by both generation and retrieval. A system first identifies material available to its retrieval process and then constructs an answer from that material.
The Primary Source Problem Is Measurable
The primary source problem is not the disappearance of original information. Government databases, research papers, corporate filings, documentation, journalism, and other source material continue to exist.
The measurable change concerns user behavior. AI-generated summaries increasingly appear before source links, and users presented with those summaries click source pages less frequently. Pew’s observed reduction from a 15% traditional-result click rate without an AI summary to 8% with one quantifies that change.
AI can therefore function as the first information layer while the primary source remains available behind it. The resulting information process depends on two separate activities: obtaining an AI-generated answer and verifying consequential claims against the original evidence.