AI-powered search efficiency has made huge gains, reducing hallucinations and more

AI-assisted search has finally matured into a reliable research tool after years of disappointing performance. Since early 2023, various companies have attempted to combine large language models with search capabilities, but these systems frequently hallucinated information and couldn’t be trusted. Now, in 2025, several major players have released genuinely useful implementations that can reliably conduct online research without the rampant fabrication issues that plagued earlier versions.

The big picture: OpenAI‘s search-enabled models (o3 and o4-mini) represent a significant advancement by integrating search capabilities directly into their reasoning process.

• Unlike previous systems, these models can run multiple searches as part of their chain-of-thought reasoning before producing a final answer.
• This approach has dramatically reduced hallucinations while maintaining useful responses, creating a more trustworthy research assistant.

Key developments: Multiple AI companies have launched “Deep Research” capabilities with varying degrees of success.

• Google Gemini, OpenAI, and Perplexity have all released implementations that can generate lengthy, citation-rich reports on complex topics.
• The recent upgrade to Gemini 2.5 Pro has significantly improved Google’s offering, though the multi-minute wait for comprehensive reports isn’t ideal for all use cases.

The competitive landscape: Not all AI search implementations are created equal, with significant differences in transparency and performance.

• Google Gemini doesn’t reveal what it’s searching for, creating trust issues despite Google presumably having the best search index.
• Anthropic’s Claude added web search a month ago but uses the Brave search index, which may be less comprehensive than competitors, and lacks the integrated reasoning approach of OpenAI’s implementation.

Why this matters: The sudden improvement in AI-assisted search threatens to fundamentally reshape how people interact with information online.

• Users may increasingly bypass websites altogether, getting information directly from AI chatbots instead of visiting the source sites.
• This shift could upend the economic model of the web, intensifying the already active legal battles over AI training data and content usage.

Behind the numbers: The author reports a personal decline in Google search usage as these new AI search capabilities improve.

• The pattern suggests a potential widespread behavior change if these tools continue to prove reliable for everyday information needs.
• This shift is happening precisely when the tools have crossed the threshold from “interesting but unreliable” to “actually useful for regular tasks.”

Reading between the lines: While these tools still make mistakes, they’ve reached a level of reliability where users might skip fact-checking for lower-stakes tasks.

• This represents a significant inflection point in AI capability that could accelerate adoption and disruption.
• The economic impact on content creators and web publishers could be substantial as traffic patterns shift toward AI intermediaries.