AI-Powered Search

Introduction: Search Has Split Into Three Lanes

For 25 years, "searching the internet" meant one thing: Google. Type keywords, scan blue links, click through to find what you need. That model is fragmenting rapidly, and the implications for how clinicians find information are profound.

Consider this scenario: A patient asks about a supplement they saw promoted on social media for "reducing inflammation." In the old model, you'd Google it, wade through sponsored results, scan a few pages, and piece together an answer. Now you have options: ask Perplexity for a synthesized answer with citations, use ChatGPT's deep research to get a comprehensive report, or check if Google's AI Overview gives you what you need instantly.

In 2025, you have three distinct approaches to finding information online:

• Traditional search (Google, Bing): Returns links for you to explore—still dominant with 89.7% market share
• Answer engines (Perplexity, Google AI Overviews): Synthesize information and give you answers with citations
• Deep research (ChatGPT, Claude, Gemini, Perplexity Pro): Spend minutes autonomously researching, then deliver comprehensive reports

Each has different strengths. Knowing which to reach for—and when to just Google it—will make you dramatically more efficient at finding the information you need. This module will help you navigate this new landscape with practical guidance for clinical scenarios.

Part 1: Answer Engines—Perplexity and the New Search

What Perplexity Is

Perplexity calls itself an "answer engine" rather than a search engine—and the distinction matters. Instead of returning a page of links for you to explore, it synthesizes information from multiple sources and gives you a direct answer with inline citations.

Ask Perplexity "What are the current first-line treatments for community-acquired pneumonia?" and you get a structured response citing UpToDate, IDSA guidelines, and recent literature—not ten blue links to sift through. Every factual claim includes a numbered citation you can click to verify.

The growth has been remarkable: Perplexity now handles 780 million monthly queries, up from 230 million just a year ago. With a $14 billion valuation and 15-30 million active monthly users, it's become a serious alternative to traditional search—particularly for research-oriented tasks.

How It Works

Perplexity combines real-time web search with large language models. When you ask a question:

1. It searches the web for relevant sources (often 5-15 for quick search, many more for Pro)
2. It reads and processes those sources, extracting relevant information
3. It synthesizes an answer, citing each claim with numbered references
4. It suggests follow-up questions based on related topics

The result is conversational research. You can ask follow-ups, refine your question, and drill deeper—all while seeing exactly where the information comes from. Unlike traditional search, where you're doing the synthesis mentally as you click through links, Perplexity does that cognitive work for you.

This shift matters because it changes the skill required. Instead of crafting keyword queries and evaluating source credibility from snippets, you're crafting natural language questions and evaluating whether the synthesis is accurate. The literacy required is different—not necessarily easier, but different.

Quick Search vs. Pro Search

Perplexity offers two modes:

For quick factual questions, Quick Search is fine. For anything requiring synthesis across multiple sources—literature review, differential diagnosis research, guideline comparisons—Pro Search is worth the wait.

Perplexity for Clinical Research

Perplexity is particularly useful for clinicians when you need to:

• Stay current: "What's new in GLP-1 agonist research in 2025?"
• Find guidelines: "Current AHA recommendations for anticoagulation in AFib"
• Research patient questions: "Safety of turmeric supplements with blood thinners"
• Explore differentials: "Causes of elevated ferritin with normal iron studies"
• Find grey literature: Institute reports, white papers, professional society statements
• Compare treatments: "Head-to-head trials comparing SGLT2 inhibitors"

The University of Michigan Library specifically recommends Perplexity for finding current grey literature—think tank reports, institute white papers, professional society statements—that traditional academic databases often miss. It's also useful for tracking down peer-reviewed articles as a starting point for exploration, though it can't access full text behind paywalls.

Practical Examples

Here's how Perplexity handles real clinical queries:

Perplexity returns a structured response covering the DPP trial results, ADA guideline recommendations, recent meta-analyses on cardiovascular outcomes, and emerging research on metformin's effects beyond glucose—each claim with a citation. You can then ask follow-ups: "What about in patients over 65?" or "Any concerns with renal function?"

Compare this to Google, where the same query returns a mix of Mayo Clinic patient information, sponsored pharmacy results, and news articles—requiring you to click through multiple sources to piece together the same information.

Perplexity's Limitations

Despite its strengths, Perplexity has real limitations that matter for clinical use:

• Still hallucinates: A 2025 Wordstream study found a 13% error rate (better than Google AI Overviews at 26%, but not zero). That's roughly 1 in 8 responses containing some inaccuracy.
• Source quality varies: It sometimes cites sources that plagiarize other articles, or relies on outdated information. One analysis found Perplexity occasionally citing the same underlying text multiple times through different websites, creating an illusion of corroboration.
• Struggles with structured data: Restaurant hours, local business info, real-time schedules—Google is still better. Ask Perplexity for the nearest open pharmacy and you'll often get a generic answer rather than actual local results.
• Can't access paywalled content: Won't see full text of most journal articles without institutional access. It can find that a study exists and summarize the abstract, but can't read the full methodology or results sections.
• Time-sensitive information: For rapidly evolving topics, Perplexity sometimes uses outdated or less trustworthy sources, particularly when recent high-quality sources haven't been indexed yet.

Always verify important clinical information against primary sources. Use Perplexity to find those sources quickly, not to replace reading them. The citations are the feature—click them, verify them, and use your clinical judgment to assess whether the synthesis accurately represents what the sources actually say.

Part 2: Google AI Overviews—AI Meets Traditional Search

What Changed

Google hasn't ceded search to Perplexity. Instead, it's added AI Overviews—AI-generated summaries that appear at the top of search results for many queries. These use Google's Gemini model to synthesize information from multiple sources, and they're now rolling out to over a billion users globally.

Search for "symptoms of pulmonary embolism" and you'll see an AI-generated summary before the traditional blue links—covering the classic presentation, risk factors, and when to seek emergency care. Google is trying to give you the best of both worlds: synthesized answers plus the ability to explore sources through the traditional link-based interface below.

This represents Google's acknowledgment that the search paradigm is shifting. Users increasingly want answers, not just links. But Google's approach is additive—AI Overviews enhance traditional search rather than replacing it. The blue links, ads, Maps integration, and vertical search features all remain.

How AI Overviews Compare to Perplexity

The key insight: Google AI Overviews are better integrated with Google's ecosystem (Maps, Flights, Shopping, Calendar), while Perplexity is better for pure research and synthesis. Google is faster and more seamless; Perplexity is more thorough and better cited.

The Accuracy Problem

The higher error rate for Google AI Overviews (26% vs. Perplexity's 13% in one study) deserves attention. Google's AI Overviews are optimized for speed and integration, not deep research. They can be confidently wrong in ways that are hard to detect without clicking through to sources.

For medical queries specifically, Google has implemented additional safeguards—AI Overviews for health topics often include disclaimers and may link directly to authoritative sources like Mayo Clinic or CDC. But these safeguards aren't perfect, and the convenience of a quick answer can discourage the verification that clinical questions demand.

When Google Still Wins

Despite all the AI advances, traditional Google search remains better for:

• Local information: "Pharmacy near me open now"—Google's Maps integration with real-time hours is unmatched
• Real-time data: Flight status, stock prices, sports scores—structured data that updates continuously
• Navigation: Directions, business hours, contact info—tap to call, tap to navigate
• Shopping: Price comparisons, product availability, reviews—vertical search with purchase intent
• Specific websites: When you need a particular site, not a summary— "UpToDate atrial fibrillation" gets you there faster than asking for a summary
• Images and videos: Finding specific visual content, medical imaging examples, procedure videos

The rule of thumb: if you need to do something (book, buy, navigate, call), use Google. If you need to understand something, consider Perplexity. If you need to understand something comprehensively, use deep research.

Part 3: Deep Research—When You Need a Full Investigation

What Deep Research Is

Both Perplexity and the major chat models (ChatGPT, Claude, Gemini) now offer "deep research" modes that go far beyond quick searches. These features autonomously research a topic for minutes—reading dozens or hundreds of sources—then deliver comprehensive reports with citations throughout.

Think of it as having a research assistant who spends 15-30 minutes pulling together everything relevant on a topic, synthesizing it, and presenting it in a structured format. The AI iteratively searches, reads documents, reasons about what it's learned, identifies gaps, searches again, and refines its understanding—mimicking how a human expert might research an unfamiliar topic.

Perplexity's deep research, for example, achieved a 93.9% accuracy score on the SimpleQA benchmark—far exceeding standard search or quick responses. For complex topics that require synthesizing multiple perspectives, this thoroughness matters.

When Deep Research Makes Sense

Deep research isn't always the right choice—it takes minutes instead of seconds. But for certain tasks, the investment pays off:

• Preparing presentations: Grand rounds, case conferences, educational talks
• Complex clinical questions: When you need multiple perspectives and comprehensive coverage
• Staying current on a topic: "What's changed in heart failure management in the past year?"
• Understanding controversies: When expert opinion is divided and you need to understand why
• Writing or research: Review articles, grant backgrounds, policy documents

How the Major Players Compare

Enabling Search in the Big Three

If you're using ChatGPT, Claude, or Gemini, here's how to make them search the web:

With Search vs. Without Search

Understanding when models are searching matters for interpreting their answers. By default, ChatGPT, Claude, and Gemini respond from their training data—they're not searching the web unless you specifically ask them to or enable search features.

For clinical questions where currency matters—new guidelines, recent studies, drug interactions—always ensure search is enabled or use a dedicated search tool. A question like "What's the current recommendation for aspirin in primary prevention?" will get different answers depending on whether the model is searching or relying on potentially outdated training data.

Part 4: Choosing the Right Tool

A Decision Framework

Clinical Scenarios

Here's how this plays out in practice:

Part 5: Verification—The Non-Negotiable Step

No matter which tool you use, verification remains essential. AI search tools make finding information faster, but they don't make it automatically accurate. The speed and convenience can actually make verification more important, not less— because the barrier to acting on unverified information has never been lower.

Why Verification Still Matters

A 2025 Frontiers in Digital Health study compared ChatGPT, Gemini, Copilot, Claude, and Perplexity on clinical practice guideline questions for lumbosacral radicular pain. The finding: no AI chatbot provided advice in absolute agreement with CPGs. All had errors, omissions, or outdated information.

This finding is consistent across studies. AI tools can synthesize information effectively, but they lack the clinical context to know when guidelines don't apply, when patient-specific factors change the calculus, or when the latest evidence hasn't yet been incorporated into their sources.

This doesn't mean the tools aren't useful—they dramatically accelerate finding relevant information. But they're research accelerants, not replacements for clinical judgment. Use them to find information faster, then apply your expertise to evaluate and contextualize that information.

Verification Checklist

• Check the citations: Click through to source documents. Are they saying what the AI claims? AI synthesis can subtly distort meaning, especially for nuanced recommendations.
• Check the dates: Is this based on current guidelines or outdated sources? A 2019 source might predate important trials that changed practice.
• Cross-reference: Does UpToDate, DynaMed, or another authoritative source agree? If there's disagreement, investigate why.
• Consider the source quality: Peer-reviewed journal vs. institutional website vs. blog post vs. forum? AI search tools don't always distinguish quality reliably.
• Trust your training: Does this align with what you know? If not, investigate further. Sometimes the AI is right and your knowledge is outdated—but sometimes the AI is wrong.
• Consider the patient context: Even accurate general information may not apply to your specific patient's situation, comorbidities, or preferences.

Building a Verification Habit

The challenge isn't knowing that verification matters—it's actually doing it consistently when the AI-generated answer seems reasonable and you're pressed for time. Some strategies:

• Make it automatic: Before acting on any clinical information from AI, open UpToDate or your preferred authoritative source in another tab
• Verify the key claims: You don't need to verify every sentence—focus on the specific recommendations that would change your management
• Be especially careful with: Dosing, contraindications, drug interactions, and anything that differs from your training or usual practice
• Note the limitations: When sharing information with patients, acknowledge that you're providing general information and their specific situation may differ

Resources

Essential Reading

Tools to Try

Podcasts & Videos