What the research actually says about patients choosing a doctor by reviews

Sit through enough pitches in this category and a percentage in the seventies or eighties will appear on a slide. Some share of patients, it says, choose their doctor from online reviews. The slide rarely names a study. When it does, the trail usually ends at a vendor blog post citing another vendor blog post citing a press release nobody has read.

applaud sells review acquisition to medical practices. That makes us exactly the kind of company that profits when that number is large and unexamined, and I would rather say so at the top than have you work it out at the bottom. So this post does the dull thing instead. Three real sources, reported including the parts that cut against us, then the one check worth running on your own profile this week.

The good study is fourteen years old

The best-sampled evidence on whether Americans pick physicians from ratings is Hanauer and colleagues in JAMA, 2014. Fieldwork ran in September 2012: a national sample, n = 2,137, 60% response rate. A real denominator and a published response rate already put it ahead of nearly everything else in circulation.

What it found, straight:

• 65% of respondents were aware that websites rate and review physicians.
• Of those aware, 36% had sought physician ratings in the past year.
• Asked how important ratings are in choosing a physician: 19% said very important, 40% somewhat important, and 41% said not important.
• Among the people who had actually sought ratings, 35% selected a physician because of good ratings and 37% avoided one because of bad ratings.
• Awareness of physician rating sites trailed awareness of ratings for cars (87%), movies or books (82%) and restaurants (81%).

The third bullet is the one that never makes the slide. JAMA reported this the other way round: 59% said ratings were somewhat or very important. That is the sentence a vendor would quote, and it is a fair reading. But the same table says 41% chose the bottom option, and on a three-point scale that is the largest single group, ahead of “somewhat” by a point that sits well inside the margin of error. Both sentences are true. The industry only ever says the first one, and I am not going to only ever say the second. If someone in my industry tells you every patient is checking your stars before they book, they are selling you something, and the best-sampled data in the field does not back them.

Then the caveat that runs the other way, which I am not going to bury: this is 2012 fieldwork. It predates the smartphone becoming the default way people search for a clinician, and it predates most of what a patient now sees when they type a specialty into Maps. I will not pretend those figures are current. The point of citing it is that it is the good study, not the recent one, and those are separate virtues.

The recent number is about restaurants, not doctors

The current, well-run consumer survey in this space is BrightLocal's Local Consumer Review Survey 2026, published 11 February 2026 on a representative panel of 1,002 US adults. It is about local businesses in general. It is not a healthcare study, and it should not be quoted as one.

• 97% of consumers read reviews for local businesses.
• 41% say they “always” read reviews when browsing, up from 29% the year before.
• 31% will only use a business rated 4.5 stars or higher, up from 17% in 2025.

Those are large one-year moves and worth treating with some suspicion; a 17-to-31 jump on the star threshold is close to a doubling. BrightLocal reports it, I am attributing it to them, and I would want to see it hold for a second year before building a strategy on the slope rather than the level.

What the survey does establish is that the threshold habit exists and is hardening in the general population. The 4.5 floor is the most operationally useful line in it, because it describes a rule rather than a preference.

The one experiment that isolates the text

Surveys ask people what they think they do. A handful of experiments have tried to isolate what actually moves the decision; Han and colleagues, in the Journal of Medical Internet Researchin 2024, is the one whose design maps most directly onto what a practice controls: a 2×2×2 between-subjects decision-controlled design in which participants judged physicians whose textual reviews varied on three factors at once. The proportion of negative reviews, low or high. The kind of claim they made, evaluative or factual. And whether the physician had responded.

The design detail that matters most is easy to skim past: the physician's overall rating was held high throughout. Everyone in that experiment was looking at a well-rated doctor. The qualitative findings:

• Negative reviews decreased selection intention even with the overall rating held high.
• The proportion of negative reviews and the claim type both had a greater effect on selection intention than the physician's response did.
• A high proportion of negatives, factual negatives, and the absence of a response each reduced selection intention relative to their counterparts.
• The presence of a physician response decreased the influence of negative reviews, through both direct and moderating effects.

Its limits are the limits of any lab study. It measures stated intention in a controlled task, with participants recruited to study carefully what a real patient glances at on a phone in a car park between two other errands. I would not port the effect sizes into a spreadsheet. The direction and the ordering are what I take from it, and both are consistent enough with the other two sources to lean on.

Screening, not choosing

Put the three side by side and a defensible claim emerges, narrower than the one on the slide and considerably more useful. Reviews work as a screening filter rather than a choosing mechanism. They mostly eliminate. They rarely select.

Each source says a version of it. In Hanauer, among people who actually looked, 35% picked a physician on good ratings and 37% avoided one on bad ratings. I am not going to make anything of the two-point gap; both carry intervals eight points wide either side and they overlap almost entirely. What is worth noticing is that avoidance is at least as common as selection, on a question where you would expect people to under-report ruling someone out. In BrightLocal, the 4.5-star rule is a screen by construction: a floor you clear or fail, not a ranking you win. In Han, a proportion of factual negatives moved the decision while the average rating stayed high, which is the same thing said precisely. The number passed the filter. The text did not.

The asymmetry is what most practices have backwards. A hundred warm reviews and four specific complaints does not read to a patient as 96% satisfaction. It reads as four things that might happen to them. Praise clears the numeric gate and does close to nothing at the text gate, because praise is generic and complaints are concrete. If you have been running your review programme as an exercise in accumulating positive volume, the evidence says volume is only doing the first half of the job.

The recency check, which takes five minutes

BrightLocal 2026 also found that 74% of consumers look for reviews written in the last three months. That single figure converts into a check you can run on your own profile right now.

1. Pull your listing up in Google Maps. Search results will not let you re-sort; Maps will. Use the patient-facing card, not the merchant dashboard — you want to see what they see.
2. Open the reviews.
3. Change the sort from “Most relevant” to “Newest”.
4. Count down to the fifth review and read what Google says under it. Google gives you “two weeks ago” or “four months ago” rather than a date, which is imprecise in exactly the direction you care about. Treat anything reading four months or older as a fail.

That date is a floor, not a picture. If it falls inside the last 90 days you know at least five reviews sit in the window most people say they look in. If it does not, you know you have fewer than five — and the honest way to find out how many is to keep counting down from the top until you cross the 90-day line. Write that number down. It is the one that matters, and it is almost never the one on the card.

The reason this check is worth more than a count is that the count hides it completely. Our own 2026 dataset of 15,061,473 Google reviews across 142,417 US medical practices puts the median practice at 19 reviews, so most owners already know their number is small. What surprises people is that a practice holding several hundred can fail the same check. The lifetime total sits at the top of the card and looks like an asset. Five dates further down is what a patient with the recency habit actually experiences, and the two can disagree completely.

While you are sorted by newest, do one more pass. Read the ten most recent reviews and mark which of the critical ones make a factual claim — a wait time, a billing error, a specific process that broke — and which are evaluative, the ones that just register displeasure. That list is your work queue, and the order matters more than you would expect. Then put the check in whatever you already look at monthly. It is the only thing on that card that gets worse while you do nothing: your total can only go up, and your fifth date can only go stale.

Answer the factual ones first

Han's ordering has a direct operational consequence, and and it is the part of the paper I keep coming back to. Claim type moved selection intention more than the response did, and factual claims did more damage than evaluative ones. So the triage is not chronological and it is not by star rating.

A review saying “I waited ninety minutes past my appointment time and nobody at the desk told me why” is working harder against you than one saying “worst office I have ever been to”. The second reads as temperament, and readers discount it. The first reads as evidence, it is checkable against the reader's own specific fear, and it describes something they can picture happening on a Tuesday. Answer that one today. The all-caps one can wait until Friday.

Worth being precise about what a response buys you: in Han it decreased the influence of negative reviews, through direct and moderating effects. Decreased, not cancelled. A good reply softens a bad review; it does not delete it, and no amount of skilled writing turns a factual complaint into an asset. We have the scripts written up in how to respond to negative patient reviews and there is no sense repeating them here. The only thing I would add from this paper is the running order.

Why we ask everyone

Since I opened by admitting the incentive, here it is stated plainly and once. The last-90-days window stays populated only if the asking is systematic, and asking every patient after every visit is the mechanism applaud runs, in the practice's own name, rather than depending on whoever is at the desk remembering during a busy afternoon. Nobody is screened out of the ask based on what they said. The survey half is free, and if no review posts, no invoice exists.

The version of this argument I would be willing to defend in front of the 41% who told Hanauer that ratings do not matter to them is a narrow one. Reviews will not persuade an indifferent patient to pick you. What they will do is stop a checking patient from ruling you out — and the checking patient is the one who is shopping: the family that just moved, the person whose insurance changed in January, the patient whose doctor retired. That is a much smaller group than the pitch deck claims, it is the group worth the most to you, and for a meaningful share of them the good part of the decision turns on five dates on a page.

Sources

1. Hanauer DA, Zheng K, Singer DC, Gebremariam A, Davis MM. “Public Awareness, Perception, and Use of Online Physician Rating Sites.” JAMA. 2014;311(7):734–735. September 2012 fieldwork; n = 2,137; 60% response rate. doi:10.1001/jama.2013.283194. jamanetwork.com/journals/jama/fullarticle/1829975
2. BrightLocal. “Local Consumer Review Survey 2026.” Published 11 February 2026; representative panel of 1,002 US adults via SurveyMonkey. brightlocal.com/research/local-consumer-review-survey
3. Han X, Lin Y, Han W, Liao K, Mei K. “Effect of Negative Online Reviews and Physician Responses on Health Consumers' Choice: Experimental Study.” Journal of Medical Internet Research. 2024;26:e46713. doi:10.2196/46713. jmir.org/2024/1/e46713