ChatGPT is rapidly moving beyond chatbots into healthcare systems, advertising networks, and consumer planning tools, but real-world tests reveal the AI still struggles with context-dependent details that can cost users thousands of dollars. As OpenAI scales ChatGPT across hospitals, ad platforms, and travel services, the expansion exposes both the promise and the persistent blind spots of large language models in high-stakes domains.

Why Does ChatGPT Struggle With Real-World Vacation Planning?

When a travel journalist asked ChatGPT, Google Gemini, and Perplexity to plan an identical five-day Disney World vacation for a family of five traveling from Philadelphia in late March 2027, the results diverged dramatically. ChatGPT estimated the trip would cost $7,500 to $9,000, with a realistic target of $8,500. Gemini came in higher at $10,050 to $13,350, while Perplexity's full package with premium add-ons reached $10,545 to $13,955. The difference between ChatGPT's estimate and Perplexity's upper range was more than $5,000.

The culprit was a single overlooked detail: timing. ChatGPT recommended Disney's Art of Animation Resort family suite but underestimated the nightly rate by roughly half. The March 30 through April 3 dates fall immediately before Easter Sunday in 2027, placing the vacation during one of Disney World's busiest spring-break periods. Gemini factored in this seasonal pricing pressure and estimated the same suite at $4,000 to $4,500 for five nights, compared to ChatGPT's initial $1,800 to $2,600 estimate.

Gemini also caught a pricing rule that ChatGPT missed: children ages 10 and older pay adult ticket prices at Disney World. In this family of three children ages 12, 10, and 6, only one child qualified for a discounted ticket. That single rule change rippled through the entire budget. When the journalist showed ChatGPT the competing estimates, the AI revised its recommendation upward to approximately $10,000, with $11,000 available as a cushion, bringing it much closer to Gemini's calculation.

Despite the budget differences, Perplexity generated the itinerary the journalist found most useful. It suggested visiting Magic Kingdom on days one and five, with Hollywood Studios, EPCOT, and Animal Kingdom in between. This structure made sense for a family with young children because it avoided cramming all Magic Kingdom attractions into a single exhausting day during peak season. Gemini suggested a different park order with Park Hopper flexibility, while all three AI assistants agreed that selective use of Lightning Lane (Disney's paid skip-the-line service) made more financial sense than buying it for every day.

How Is ChatGPT Expanding Into Hospital Systems?

OpenAI's ambitions extend far beyond vacation planning. On September 1, 2026, OpenAI announced that ChatGPT for Healthcare now integrates with Epic Systems' electronic health record (EHR) platform, one of the largest medical record systems in the United States. The integration gives authorized clinicians at UCSF Health read-only access to patient notes, lab results, medications, specialist records, and encounter histories for more than 325 million patients in Epic's system.

The integration is carefully constrained. Clinicians can pull authorized Epic data into ChatGPT to review a patient's history before an appointment, or in supported deployments, use ChatGPT directly within the EHR interface without leaving the patient chart. However, ChatGPT cannot update medical records, place orders, message patients, or override existing chart permissions. The read-only restriction is not merely a technical limitation; it is a legal and clinical safeguard that prevents the AI from directly altering patient care workflows.

OpenAI has prepared the ground for this expansion for months. In July 2026, the company made Health in ChatGPT broadly available to U.S. users, allowing people to connect their own medical records and wellness data for personal health conversations. OpenAI reported that more than 300 million people use ChatGPT for health-related questions each week, establishing consumer demand for AI-assisted health information.

"The pilot is meant to help clinical teams understand what has changed across a complex patient record and spend less time synthesizing data," stated Suresh Gunasekaran, president and CEO of UCSF Health.

Suresh Gunasekaran, President and CEO, UCSF Health

The Epic integration positions OpenAI in direct competition with Microsoft's Nuance DAX, an AI tool for ambient clinical documentation that is already embedded in many hospital workflows. The real battle is not whether AI reaches the patient chart; it is whose system hospitals trust with the closest view of sensitive patient data. OpenAI's healthcare workspace includes role-based access controls, single sign-on (SSO) authentication, audit logs, data residency options, and Business Associate Agreement (BAA) support for HIPAA compliance, the federal privacy law governing patient records.

What Does ChatGPT's Ad Business Reveal About Its Scale?

OpenAI's advertising ambitions are expanding just as rapidly. The company announced that ChatGPT Ads reached a $1 billion annualized revenue run rate in less than 200 days, a milestone that underscores the speed at which OpenAI is monetizing its user base. On August 31, 2026, OpenAI expanded ChatGPT Ads to the Middle East and North Africa (MENA) region, giving regional advertisers access to self-service campaign tools through Ads Manager. The expansion also includes newly added markets in India and Europe, bringing ChatGPT Ads to more than 40 countries.

ChatGPT Ads operates differently from traditional search advertising. Instead of appearing alongside keyword searches, ads appear alongside relevant conversations within ChatGPT itself. The platform uses the context of a user's current conversation to determine which ads may be relevant, and ads are clearly labeled and remain separate from ChatGPT's answers. OpenAI emphasizes that advertising does not influence ChatGPT's responses and that advertisers do not receive access to users' private conversations.

The advertising platform increasingly resembles a performance marketing channel. It now offers cost-per-click (CPC) and outcome-optimized bidding, geographic and platform targeting, custom audiences, product feeds, Pixel and Conversions API capabilities. According to OpenAI, CPC and outcome-optimized bidding now account for the majority of campaigns on the platform. The company reports that tens of thousands of advertisers use ChatGPT Ads, with small and medium-sized businesses making up a material share of the advertising business.

OpenAI shared performance examples to demonstrate the platform's effectiveness. One ecommerce advertiser achieved a 3x return on ad spend over 28 days, while one technology partner found that more than 80 percent of ad-driven ChatGPT traffic came from new customers. These figures are based on examples shared by OpenAI rather than independent benchmarks, so they should be interpreted as case studies rather than representative averages.

How to Evaluate AI Tools for High-Stakes Decisions

• Verify Seasonal and Contextual Pricing: AI models trained on historical data may not account for peak seasons, holidays, or time-sensitive pricing changes. Always cross-check AI estimates against current market conditions and ask the AI to explain its assumptions about timing and demand.
• Check for Hidden Rules and Eligibility Thresholds: Many services have age cutoffs, qualification requirements, or category-specific pricing that AI systems may miss or misapply. Request that the AI explicitly state any rules or thresholds it is using and verify them against official sources.
• Compare Multiple AI Systems for Critical Decisions: Different AI models prioritize different information and may catch different details. For decisions involving significant money or health outcomes, running the same query through multiple AI systems can reveal blind spots in any single model.
• Treat AI Estimates as Starting Points, Not Final Answers: AI-generated budgets, medical summaries, and travel plans should be treated as frameworks to refine, not as definitive answers. Use them to structure your research, then validate key assumptions with primary sources.

The three expansions of ChatGPT reveal a consistent pattern: OpenAI is moving the AI into domains where accuracy, context, and trust matter enormously. In vacation planning, a $5,000 error is frustrating but recoverable. In healthcare, a missed detail in a patient record could affect clinical decisions. In advertising, the stakes involve brand reputation and customer data. Each domain requires different safeguards, and the vacation planning test suggests that ChatGPT's core limitation remains unchanged: the AI can synthesize information and generate plausible-sounding answers, but it struggles with context-dependent details that require real-time knowledge or domain-specific rules.

OpenAI's strategy appears to be scaling first and refining guardrails second. The company is expanding ChatGPT Ads globally, integrating with hospital systems, and positioning itself as a general-purpose AI assistant across consumer and enterprise domains. The vacation planning test suggests that users and organizations adopting these tools should do so with clear eyes about what ChatGPT can and cannot do reliably. The AI excels at generating structured plans, summarizing information, and identifying relevant options. It struggles with context-dependent pricing, rule-based eligibility, and real-time market conditions. For now, ChatGPT works best as a starting point, not as a final decision-maker.