Health Information Sources

In these survey responses, doctors remain the most widely named information source, while AI tools already stand level with Google search. Those roles can overlap: the survey lets one response name both a doctor and a chatbot.

The distinction matters when information leads to a decision about a symptom, a test result or whether to seek care. Research discussed below shows why a fluent AI answer is not enough to establish clinical accuracy. The survey itself records where people say they get information, not whether the answer was right or what care followed. Against that limit, the pattern is clear: 54.0% name a doctor, while about one in five name AI and a similar share name Google.

What the survey asked#

WHI 2026 asked: “Where do you primarily get your health-related information?” Respondents could tick any number of eight boxes.

6,996 responses answered this multi-select question, so the shares do not add to 100%.

The clinician’s lead is not close: 54.0% against 19.8% for the next-named source. Nothing here shows the doctor being displaced.

The pair in the middle is newer. AI tools were selected in 1,382 responses and Google search in 1,380, two apart on the same base of 6,996. ChatGPT, the tool the AI option names, launched in 2022 (Ayre et al., Medical Journal of Australia, 2025); Indian patients were looking health questions up online before ChatGPT launched, and a 2019 survey of 110 patients at a tertiary hospital in Davanagere, central Karnataka, found 91.8% of them used Google for health information and 73% had used what they found to self-diagnose (Shekhar & Javalkar, IJCMPH, published 2020). WHI 2025 reported no comparable measure, so these results cannot show how quickly the newer tool drew level.

The second-largest answer is not a source at all. 24.8% of responses ticked “I don’t actively seek health information”, more than named AI, search, or family and friends. Of the responses selecting it, 261 also named at least one source, so people could choose both. This answer records a box that was ticked; it does not mean those people avoid doctors, ignore symptoms or go without care.

These are survey responses, not verified unique people. IT, software and SaaS account for 3,084 of the 9,097 submissions, so the results should not be read as a national rate.

Workers in their 20s name AI most and a doctor least#

These four bands cover 6,946 of the 6,996 responses; the rest fall outside ages 20 to 59 or gave no usable age. The 50–59 row rests on 216 responses and is the least stable in the table. A pattern measured once across ages is not a path through time: it shows that younger respondents name AI more often, not that anyone’s habits changed.

Respondents in their 20s are least likely to name a doctor, at 50.7%.

The two columns are separate shares from the same multi-select question, not a trade-off inside one person. A response can name both. The age bands differ in the sources they name for health information. These figures do not show whom respondents consult when they need care.

A fluent answer is not necessarily a correct one. Research on AI in medicine also distinguishes accuracy on a test from performance in clinical care. A systematic review covering 761 studies and 1,534 evaluated model instances found accuracy to be the most commonly assessed property, and its authors flag that concentration as a weakness, warning that a focus on accuracy alone does not capture the complexity of clinical performance. Of those 1,534 instances, 1,435, or 93.55%, were general-domain models such as ChatGPT and GPT-4 rather than medically tuned ones (Shool et al., BMC Med Inform Decis Mak, 2025).

One property has been measured directly. When ChatGPT-3.5 was asked in April 2023 to produce 30 short medical papers with references, 47% of the 115 references it generated did not exist and a further 46% existed but were cited inaccurately; 7% were both real and correct. The authors state that they did not check whether the text above those references was right, only the citations themselves (Bhattacharyya et al., Cureus, 2023). That is a 2023 result about one model’s reference lists, not a measure of what a current assistant says about a symptom.

The highest income band names both AI and doctors more often#

The seven named bands hold 6,458 of the 6,996 responses. A further 537 responses answered “Prefer not to say” (AI 17.7%, doctor 58.5%) and one left the field blank. The AI column does not rise at every step: 25.7% at ₹15–25 lakh falls to 24.7% at ₹25–40 lakh before the top band.

Unlike the age comparison, the highest income band has higher shares for both sources than the lowest. The ₹40 lakh+ band names AI in 29.9% of its responses and a doctor in 69.3%, the highest figure in each column. The under-₹3-lakh band is the least likely on both counts, at 16.4% and 46.5%.

The contrast spans both sources: the highest income band is more likely to name a clinician as well as AI. But a table of separate shares does not show how many respondents name both. Nor does it establish whether cost or access shaped their choices. Income also varies alongside other characteristics, including age, which this table does not hold constant.

Workplace wellness programmes are named in 8.0% of responses; AI tools and Google search in about 20%#

Workplace wellness programmes are the only listed source that is an employer benefit by name. They are named in 8.0% of responses; only news and media is lower, at 7.4%.

AI tools are named in 19.8% of responses and Google search in 19.7%, each roughly two and a half times the wellness figure. Among responses from workers in their 20s, AI alone is named in 24.8%, roughly three times.

These are sources people name, not ownership categories. An employer may pay for a doctor someone names here, and the table does not show which source a given benefit provides.

Comparing future editions#

A later edition can compare these shares if it uses the same question and definitions, with any differences in the sample kept in view. That comparison alone would not establish the effect of a particular employer’s decisions.