A Salesforce consumer survey reports 61% of consumers now use AI for health information, up from 2% in 2024. The jump tracks administrative friction and trust conditional on human oversight, not a delegation of medical judgment.
The phone tree takes fifteen minutes, the patient portal throws an error, and the front desk asks the patient to call back after lunch. So she opens ChatGPT, describes a symptom she's not sure about in plain English, and decides whether the visit can wait. She never tells her doctor. That bypass, not a sudden trust in AI medicine, is what a much-cited consumer survey is actually counting.
The survey is Salesforce's 2026 Connected Health Consumer Report, with the underlying PDF putting the data on the table. The widely circulated summary says 61% of consumers now use AI for health information, up from 2% in 2024. Two things to flag before that number settles into anyone's mental model. The survey is sponsored by Salesforce, which sells agentic AI to healthcare providers, and the survey base is 3,200 consumers worldwide, not the "US adults" implied by the headline that has been making the rounds. The US-only cut, if it exists in the report, is a slice of a global poll.
The forcing function sits in the middle of the report, not the lede. 58% of patients will delay or skip necessary care because scheduling is too difficult. Read against that, the 61% stops looking like a verdict on AI and starts looking like a verdict on the access layer in front of it. When the phone tree breaks, patients route around it. AI is the route.
The 61% also measures more than the framing suggests. "Use AI for health information" lumps ChatGPT queries, voice assistants, AI summaries inside search, and the new generation of provider-deployed patient-portal agents into one bucket. A patient who asks Siri whether ibuprofen is safe with a prescription has used AI for health information. So has a patient who reads an AI-generated visit summary inside a hospital portal. The category is broad enough that a 30x jump over two years is more plausibly an access story than a belief story.
The trust finding, read closely, runs the same direction. Patients say they trust their provider's AI agents more than public AI tools, but the trust is conditional, not absolute. It rises when the AI is integrated into the provider's portal and when a human clinician stays in the loop. It falls when the AI is freestanding. That is not a vote of confidence in algorithmic medicine. It is a vote of no confidence in the administrative layer that sits in front of it, with provider-integrated AI as the closest available substitute.
Salesforce has a direct commercial stake in how this story lands. The company sells agentic AI for healthcare, and the survey, the corporate blog summary, and the newsroom framing all point the same way: trust, adoption, agents, providers. Independent polling from Pew, KFF, the AMA, or a peer-reviewed source has not corroborated the 2%-to-61% jump. Until it does, the figure should be read as one vendor's reading of its own market, not a population-level shift.
Even with that caveat, the data points to something real. Two years ago, "call the office" was a tolerable answer. It isn't anymore. Patients are setting a new floor: if the phone tree eats fifteen minutes and the portal throws an error, they will ask a chatbot, and they will trust the chatbot more if their own provider put it there. The next inflection point is providers versus an access layer that has already started breaking on its own.
That sets up what to watch. If independent polling replicates the 2%-to-61% figure within the next two quarters, the friction workaround has hardened into a default. If it doesn't, the figure is a vendor's snapshot of its own market: real, narrower than the headline, and useful mostly as a marker of how loudly providers are being told to fix the front door.