The Reviews-First Patient: What Peer-Reviewed Evidence on Online Reviews and Patient Choice Actually Says

Two kinds of claim, one confident sentence

Search the phrase "how patients choose a provider" and you will meet two literatures wearing the same clothes. The first is academic: peer-reviewed studies, with disclosed samples and methods, published in indexed journals. The second is promotional: blog posts and sales decks that quote a precise-sounding percentage, credit it to a respected body, and move on. Both are delivered in the same authoritative register, and to a busy reader they look identical.

They are not identical. The difference is not tone but traceability. An established finding can be followed back to a document you can open, read, and criticize. A repeated marketing claim, when you try to follow it back, dissolves into a chain of sites all citing each other and none citing a primary source. The purpose of this piece is to hold the two apart in the specific domain where confusion is most costly, health-adjacent local services, and to model the tier-labeling that keeps a reader honest.

What the peer-reviewed evidence establishes about reviews and patient choice

Begin with what is genuinely settled. There is real, indexed research showing that reviews and the social signals around a provider change which one a patient selects. The evidence is strongest on the mechanism and weaker on the exact magnitude in any particular market, and that distinction should stay visible.

How much do reviews affect revenue, and where the effect concentrates

The nearest thing to a causal anchor in this literature does not come from healthcare at all. Michael Luca's study of Yelp used the platform's half-star rounding as a natural experiment and found that a one-star increase in a restaurant's Yelp rating produced a 5 to 9 percent increase in revenue. The design is what makes it quotable: because ratings near a rounding threshold are pushed up or down by essentially arbitrary rounding, the comparison approximates a controlled one, which is why the study is cited across platform-economics research.

Its most transferable result for high-consideration local services is the asymmetry. The revenue effect was concentrated in independent restaurants and was absent for chain-affiliated ones, plausibly because a recognizable brand already supplies the quality prior that an independent business has to earn through reviews. Extending that logic from restaurants to dental, med-spa, or home-services firms is reasonable synthesis, not proven fact: no equivalent large-sample US study was located for those verticals in the research behind this piece, and the extension should be labeled as an inference, not a citation.

The credential the buyer cannot see

Reviews carry extra weight in exactly the verticals where the buyer cannot verify quality directly. In aesthetic and cosmetic procedures, research in Annals of Plastic Surgery documents that patients routinely cannot verify the credentials that actually govern safety, and that non-board-certified practitioners performing aesthetic procedures are a recognized patient-safety concern. A prospective patient generally cannot tell, on sight or from a website, whether a med-spa runs under genuine physician supervision or a paper arrangement.

This is the economist's "credence good": a service whose quality the buyer cannot assess even after purchase, let alone before it. When the underlying quality signal is invisible, buyers lean harder on the visible proxies that remain, and reviews are the most available proxy of all. That is the structural reason reputation work matters more in dental, aesthetic, and legal services than in ordinary retail, and it is also the reason the folklore in these verticals is worth confronting directly: the reader has the least ability to check, and therefore the most exposure to a confident but unfounded number.

The marketing folklore: the uncited "X% of patients" statistic

Here is the claim this piece exists to isolate. Across dental and med-spa marketing content you will repeatedly meet figures like "70% of patients use the internet to find a new dentist" or "77% cite reviews as their first step," almost always attributed to the American Dental Association. They are precise, they are authoritative-sounding, and they are the kind of statistic a reader forwards without a second thought.

When you try to trace them to a primary source, they do not resolve. In the research behind this piece these specific figures could not be tied to any disclosed, locatable ADA study; they appear to originate in marketing content that credits the ADA without pointing to a document. A direct check of the ADA's own site found patient-survey methodology guidance, material on how to survey patients, rather than published national findings carrying those percentages. That does not prove the numbers are false. It proves they are unverifiable, which for a careful writer is disqualifying in its own right: a statistic you cannot source is a statistic you cannot responsibly repeat.

The correct move is not to invent a replacement number. It is to reach for the evidence that is sourced, the review-and-choice literature above, and to state the population and strength plainly. "Peer-reviewed work shows the content of reviews shifts patient choice, though the largest samples come from non-US platforms" is a weaker-sounding sentence than "77% of patients cite reviews first," and it is the one a publisher can stand behind.

Reading evidence in tiers: established, emerging, contested

The practical discipline is to attach a tier to every claim before it is published, and to keep the tier visible to the reader rather than hidden in a footnote. Three labels do most of the work.

Established means a disclosed, indexed, and ideally replicated or quasi-experimental result, such as the Yelp rounding study, read within its stated limits. Emerging means a real study whose design or population makes it directional rather than settled, such as a single cross-sectional survey of a non-US population. Contested, in this piece, is used for the uncited marketing figure: a claim in wide circulation that cannot be traced to a primary source and therefore should not be repeated as fact. Labeling is not hedging. It is the difference between reporting evidence and laundering a rumor.

Why honesty here is structural, not just polite

In these verticals the rules of the market already assume buyers cannot verify claims for themselves, and they penalize businesses that exploit the gap. The Federal Trade Commission's rule on consumer reviews and testimonials, effective October 21, 2024, makes fake and deceptive reviews a specified unfair-or-deceptive act, and it reaches every reviewed local business, not only the platforms. Attorney advertising is held to an affirmative truthfulness standard under ABA Model Rule 7.1, where a literally true statement can still be misleading if it omits what a reasonable person would need to avoid a false conclusion. The regulatory apparatus is best read as society substituting institutional guardrails for the verification the buyer cannot perform.

Buyer trust is moving in the same direction the rules point. Pew Research Center's 2026 survey found that even as 49 percent of US adults now use chatbots, up from 23 percent in 2023, only 29 percent of adult chatbot users trust the information they get from them "a lot" or "some." Trust in online reviews has likewise drifted down from its mid-2010s peak. In a market where buyers are increasingly skeptical of both reviews and answers produced by a machine, a confident but unsourced statistic is not a shortcut to credibility. It is the fastest way to lose it with exactly the reader who checks.

A short test for any reviews-and-choice claim

The separation this piece performs is repeatable. Before you trust, quote, or publish a "patients do X" statistic, run it through five questions.

• Does it name a specific author, title, year, and venue, or only an authority ("studies show", "according to the ADA")?
• Can you open the primary document and find the number inside it, rather than a chain of blogs citing one another?
• Does it declare the population and sample studied, and does that population resemble the buyers you actually serve?
• Is it a causal design (a natural experiment, an intervention) or a cross-sectional association, and is the claim being made only as strong as the design supports?
• If it fails the first two questions, is it being presented as an unverifiable claim to set aside, rather than repeated as fact?