ZS surveyed 9,000 patients and 1,400 HCPs in the US, Germany, and China to get a read on the patient journey in 2026. They found a market that is primed for explosive AI growth.

In the US 42% are online before they walk into the doctor's office for their appointment. Many continue the research after diagnosis and more than half ask for a medication by name.

AI is positioned to take more healthcare duties because a minority of people have tried it, but those who have like it nearly as much as their primary care physician.

Taking the US example only 18% of people have tried AI for health information, but of those that do 89% found it helpful. That's only 7% lower than those who find their PCP helpful. When they tell their friends there will be more AI users coming online.

The contrary argument is that even after nearly 30 years of Google, the search engine still only has 37% of Americans using it for health information according to ZS. So, perhaps the AI usage will "top out" early.

The contrary argument to the contrary argument is that other studies have found that 71% of Americans use search engines for health information "often" or "occasionally." This indicates that the ZS study might be underreporting usage (please stop -Ed.).

See the ZS report for their take but keep that 52% number for your next planning deck to defend DTC spend.

Populus Health, with the DHC Group and PM360, studied the state of virtual care in 2026. For this study the team included 1,093 patient interviews, 125 HCP surveys, and 63 pharma marketer conversations earlier this year.

The group is keeping the full report close to the chest, but we were able to gather some stats from coverage in PM360 and Medical Economics. 

When asked what constitutes a "virtual visit" the three groups in the study had somewhat different opinions. All agreed that Video followed by phone are the top qualifying channels.

Pharma marketers will be interested to see that 92% of HCPs said that prescription management is seen as an appropriate use of virtual care.

The specialties that most support Rx management are:

When looking at chronic conditions. the top specialties that say virtual care is appropriate are:

Last, but not least, are the specialties that say virtual visits are appropriate for follow-up care:

This study in Nature is a "narrative review" of studies looking at AI usage for mental health. The finding seems to be that results vary so much that comparison is difficult if not impossible.

There are massive differences between different studies on AI use for mental health. In this chart the center of the circles is the average usage, the size is the population in the study, and the placement is by population type. 

The average of this mess is 27%.

Likewise when looking at "willingness to use AI for mental health" the numbers are all over the place. Here, we have charted them and ordered them by study size.

An article from YouGov last week shows the power of search for healthcare. When asked where they had turned to first in the past three months most generations said search. Only Boomers said their HCP.

The above is only a selection of the information sources available. The full article includes "none" which was only 14.5%, indicating that 85.5% of Americans looked up health information in the past three months.

The IAB published two reports last week on AI. One was consumers' use of AI technology and one was publishers' views on it. 

The IAB triggered our sarcasm response by selecting only AI-using consumers and then making a big deal about how much they use the tools. They do disclose that respondents had to use AI monthly to qualify but it's buried in the footnotes of charts rather than prominently in the titles where it belongs.

With 49% of Americans saying they use LLMs this data misses half of Americans when discussing trust in the technology.

The Usage Frequency of AI-Using Americans

83% of Americans who choose to use AI also like the technology, how surprising (ok, they get it -Ed.).

The top three tasks for which consumers use AI are search-related. Google must be pleased. 

A companion report looks at digital publishers and their reactions to LLMs. The IAB surveyed 110 US employees familiar with their companies' LLM policies. The respondents were much more positive than might be anticipated based on the reduction in click-through traffic from the tools.

This response isn't a Pollyanna reaction to AI. These respondents understand what is at stake with the shift to LLM answers from general search, 95% say that there is an impact on their site traffic.

98% think there is some effect from LLMs, and almost half (45%) are seeing reduced click-throughs from the services.

An updated study published in arXiv runs popular general purpose and medical models through a test called NOHARM. This test goes beyond rating answers and includes whether the medical AI advice could cause harm in realistic clinical consultations.

AMBOSS is popular in teaching circles and it has the top model for success in this test. Runners up include Doximity and OpenEvidence.

All of the top models use retrieval-augmented generation (RAG) tactics to ensure they have the best data. It must be noted that GPT 5.6 and 5.5 give them a run for their money.

A small, qualitative study found in the Journal of the AMA shows that patients can be ok with AI-generated messages from an online portal if they are reviewed by a physician. The study was qualitative and included only 40 patients but found:

Ah, the ever-present mantra of "more study needed." Still, this study provides good input for anyone creating patient-facing AI systems including chatbots.

Gallup and West Health looked at how Americans with medical expenses see their employment and health benefits. They found that these Americans put a high value on those benefits, to the point of staying in jobs they would rather quit.

These charts require context. The population here is indicated by the category value and the percent is the number currently working in a job they would rather quit.

The latest quarterly data from CB Insights shows VC activity in the AI market down but still strong. Year-over-year quarter 2 was 2.6x for dollar value on a smaller volume of deals.

Gallup showed two charts based on data collected from 5,065 U.S. adults, in May 28-June 5, 2026. One was obesity and diabetes and the other was GLP-1 usage. We charted both together.

The dashed lines are placeholders for data prior to 2024 (Gallup only provided three years of actual GLP-1 use data. These would actually have a long tail extending back to the first GLP-1 receptor agonist in 2005 but the result is the same, recent usage has grown almost 4x.

When it comes to branded vs. compounded GPL-1s the audiences are somewhat different. Branded users are less likely to have experimented with different products.

All ages use the smartphone for audio, the only group that has the in-car radio come close is Boomers.

A study by Pangram looked at over a million social posts and ran the company's detection tool on the text. An important caveat: the content came from "users who opted into sharing their data for research." It is likely these users have opinions on the use of AI in social posts and so the content exposed to the tool may not be completely representative.

Regardless, LinkedIn has the most AI-generated longform content.

This "lead" is for both shortform and longform content (though the tool only tests text that is 50+ words).