Is AI recommending your health system?

If your team shows you an AI answer recommending your health system, ask to see the next few answers to the same question.

When we asked about primary care in Los Angeles three times, UCLA Health and Cedars-Sinai appeared every time, while other organizations appeared only once or twice. A favorable screenshot tells you little about how consistently your system appears for the services you want to grow.

Our research across 30 U.S. metros found that recommendations varied sharply by service and market.

Start with the service you want to grow

Health systems received 99.1% of local recommendations for heart surgery, compared with 58.1% for primary care, among providers whose affiliations we could confirm. These counts included hospitals, doctors and practices.

That difference gives leaders a reason to examine services separately. A strong presence for hospital-based care tells them little about how their primary care practices compare with nearby alternatives.

Understand the local competition

The most-recommended system’s share of all local recommendations averaged 37.9% in small metros and 23.8% in the ten largest. Yet health systems collectively received a larger share in the largest metros, where several systems split those recommendations. A larger share for the leader is different from a larger share for systems overall.

One possible explanation is local concentration: a system providing a larger share of available care may also occupy more of the answer.

The names being recommended matter too. In the largest metros, only 1.1% of local recommendations named an individual clinician, compared with 25.1% in small metros. That distinction matters when a physician leaves: a recommendation for that doctor may no longer bring a prospective patient into the system. Leaders should track whether their practices and service lines are also being recommended, as part of building patient acquisition that is less dependent on individual physicians.

Establish a baseline before making changes

Across repeated where-to-go questions for the ten services analyzed here, local provider lists differed in 85% of answer pairs. Sometimes one provider was added or replaced. All answers came from one day; this was variation between responses, not a change in performance over time.

Before commissioning changes, ask how often your system and its local alternatives appear across repeated, comparable questions. That baseline is what lets a team investigate a persistent gap rather than react to an isolated answer.

Look for improvements that scale

For a system with thousands of physicians, the practical starting point is a priority service line across its markets. Have marketing and service-line leaders compare the recommendations with where the system actually offers care, then investigate whether public information accurately describes that network. Compare the pages with the system’s current service inventory, so the team can distinguish missing information from care that is genuinely unavailable at a location.

Look for problems in the shared data and templates behind service and location pages. If several orthopedic locations omit the procedures available there, for example, correcting the common source could improve information across the network. The same approach applies to outdated practice affiliations after acquisitions. Check external directories and other cited sources for conflicting information as well.

These are potential levers to test. Ask for a pilot that identifies the affected locations, assigns responsibility for correcting the information, and compares repeated AI answers before and after the changes. Use the results to decide whether to extend the work to other services or markets.

Methodology

We collected 3,000 answers across 30 U.S. metros and 30 service lines on Oct. 4, 2026, using gpt-5.6-luna with web search through OpenAI’s API, not the consumer ChatGPT app. The findings discussed here use the original ten services, with each where-to-go question asked three times per metro. Each provider appearance counts as a recommendation, including repeats across answers. Shares were not adjusted for local provider supply or system size.

About the author

Nathan Woo is the founder of Halcy.