Patient reviews: why the response rate matters

Before calling the front desk, a patient opens Google and reads what others wrote about the clinic. The location's average rating, its latest reviews and the way the organisation answers them have become, without anyone deciding it, the first interaction with the hospital. This article sets out what published surveys say about that behaviour, what responding does to ratings and trust, what a public reply must never contain, and how a network with many locations can answer every review without breaching confidentiality.

How patients choose: what the surveys say

The most cited academic study on the subject remains Hanauer et al. in JAMA (2014), a nationally representative survey of 2,137 adults in the United States. According to that study, 65% of adults were aware of physician rating sites and 23% had used them. For 59% of respondents the ratings were "somewhat" (40%) or "very" important (19%) when choosing a physician. Among those who had looked up ratings in the previous year, 35% had chosen a physician because of good ratings and 37% had avoided one because of bad ratings.

Industry surveys point the same way, with larger numbers. According to RepuGen's Patient Review Survey 2024 (1,426 US patients), 72.71% of patients consider online reviews a major influence on their choice of provider. Asked for the decisive factors, however, they still put referrals from family and friends (60.11%) and from another clinician (58.08%) first, with online reviews in fourth place (37.79%). According to rater8's 2026 Patient Choice Report (992 validated respondents, April 2026), 55% of patients have cancelled or avoided an appointment because of negative reviews, up from 40% a year earlier.

All of these surveys are American. For Romania, Eurostat reports that in 2022 only 29% of people aged 16 to 74 looked for health information online, the lowest share in the European Union against an average of 52%. For Moldova we found no comparable published figure. The behaviour exists here too, at a smaller scale, which means each review and each reply carries more weight because there are fewer of them.

The four-star threshold and what patients read

The average rating works as a filter. According to RepuGen (2024), 83.23% of patients require a minimum of 4 stars to consider a provider, and 95.94% require at least 3.5. According to rater8 (2026), 75% will not book a provider rated below 4.0 and 44% require at least 4.5. At the same time, 55.38% of patients would consider a provider with no reviews at all (RepuGen, 2024): having no reviews is less damaging than having a low rating.

After the rating filter, the patient reads. According to RepuGen (2024), 78.61% read at least five reviews, 86.12% check at least two platforms, and Google is the platform checked most often (75.37%). Asked what matters most inside a review, patients rank the tone of the text first, the rating second and the provider's response third.

The provider's response is the third factor, at 13.93%. The figure looks small, but it is the only item on the list that the clinic controls entirely. And its weight is growing: according to rater8 (2026), 66% of patients say a provider's responses to reviews influence their trust, up from 42% in 2025.

What happens when you respond: the evidence

The most rigorous study of the effect of responses comes from outside healthcare. Proserpio and Zervas (Marketing Science, 2017) analysed hotels that began responding to reviews on TripAdvisor and compared their trajectory with the same properties on a platform where responses were not visible. The result: a 0.12-star increase in ratings and a 12% increase in review volume for responding hotels. The authors observed that hotels typically start responding after a negative shock to their rating and that, once they respond, they receive fewer but longer negative reviews: dissatisfied customers avoid short, unfounded comments when they know these will be read and answered with arguments. The study is about hotels, not hospitals; the mechanism, however, does not depend on the industry.

On perception, the data are direct. According to BrightLocal's Local Consumer Review Survey 2022 (1,124 US consumers), 89% would use a business that responds to all reviews, positive and negative. If the business responds only to negative reviews, the share drops to 59%; only to positive ones, to 52%. And 57% would not use a business that does not respond at all.

Google states the same rules in its guidance for profile owners: a reply should be nice and professional, short, personalised and signed; mistakes should be admitted, but not responsibility for things outside the business's control; and, above all, the reviewer's private information must never be shared (Google Business Profile Help).

How rarely healthcare providers do this is visible in a European study. Emmert et al. (Journal of Medical Internet Research, 2017) analysed 1,052,347 physician ratings on the German platform jameda from 2010 to 2015. Physicians responded to 1.58% of them. The rate rose from 0.70% in 2010 to 1.88% in 2015 and was higher for middling ratings than for very good ones. In a landscape where almost nobody responds, a network that answers every review stands out immediately.

Patients set the average rating. Management sets the response rate. It is the only reputation indicator you can take to 100% by internal decision.

Why the response rate matters in a multi-location network

A network with ten or thirty locations does not have one Google profile; it has ten or thirty. Each has its own rating, its own reviews and, usually, a different person looking after them. The network's reputation is, in practice, the reputation of its weakest location.

The response rate is the indicator that exposes this inequality fastest. A location with 95% of reviews answered and one with 20% belong to the same organisation but send different messages. The difference is not goodwill but organisation: who receives the notification, who publishes, and who takes over when that person is on leave.

According to BrightLocal (2025), 63% of consumers expect a reply within two to three days up to a week.

For a network, the minimum indicators to track monthly, per location, are three: the response rate (reviews answered out of the total), the median time to reply, and the share of negative reviews that received a reply. The first shows whether the process exists, the second whether it works, the third whether the team avoids the hard conversations.

What a good reply contains, and what it must never contain

Regulation (EU) 2016/679 (the General Data Protection Regulation, GDPR) defines in Article 4(15) data concerning health as personal data related to the physical or mental health of a natural person, including the provision of health care services, which reveal information about their health status. Article 9(1) prohibits processing such data, subject to the exceptions listed in Article 9(2). A public reply that confirms the reviewer was a patient, or mentions the department, the date of the visit, the diagnosis or the treatment, is a public disclosure of health data. The fact that the patient wrote about their own case does not give the clinic a legal basis to add details.

In 2016 ProPublica analysed 1.7 million Yelp reviews in the United States and identified more than 3,500 one-star reviews in which patients mentioned privacy. The investigation documented doctors, dentists and chiropractors who, defending themselves against a critical review, publicly disclosed patients' diagnoses, treatments and habits.

A good reply has five parts and no medical information:

1. Thanks, addressed by the name the reviewer displays publicly, nothing more.
2. Acknowledgement of the experience described, without confirming or denying any facts about care.
3. What the unit does in general in the situation described (the procedure, not the case).
4. An invitation to a private channel: an e-mail address or phone number of a named role, for example the patient relations officer.
5. A person's signature, by name or initials, as Google recommends.

Common mistake	What happens	The correct version
Identical template on every review	Readers see that nobody read the review	Two sentences that pick up what the patient wrote, without medical data
Confirming the person was a patient, with department and date	Disclosure of health data (GDPR Article 9)	"Thank you for your message. Please write to us at ... so that we can look into the situation."
Publicly disputing the diagnosis or treatment	Public escalation, legal risk, longer negative reviews	Acknowledge the dissatisfaction and move the conversation to a private channel
Replying only to negative reviews	Only 59% would still use the business (BrightLocal, 2022)	Reply to all, including the five-star ones
Replying after weeks	The author no longer reads it; 63% expect a reply within 2-7 days	Internal deadline: every review, within a few working days
Offers or promotions in the reply	Google explicitly advises against it	No offers; thanks and a contact channel only

How to automate responsibly

Volume is the real problem for networks: dozens of reviews a week, in several languages, across many profiles. Full automation, where a program publishes on its own, solves the volume and creates the risk described above. Responsible automation has three layers, and the decision to publish stays with a person.

Layer 1: the draft. An artificial intelligence (AI) model reads the review, detects its language (Romanian, Russian or English, in the case of Moldova and Romania), identifies the themes (waiting time, courtesy, cleanliness, billing) and writes a proposed reply in the patient's language, following the rules above. The rules are written into the model's instructions: no confirmation of patient status, no medical data, no offers, an invitation to a private channel, a signature.

Layer 2: the check. A designated staff member for that location reads the draft, corrects it where needed and publishes it. Nothing appears publicly without that click. The system can help here too: it automatically flags drafts that contain sensitive words (names of conditions, medicines, calendar dates) and sends them for a second check. Every publication is logged with the name of the person who approved it.

Layer 3: the monthly report. The same system that read the reviews groups them by theme and delivers, per location, two lists: what patients praise and what needs fixing, together with the response rate and the median time.

Automation does not replace the staff member; it shortens their time: from composing a text from scratch in three languages to reading and approving one that is already written correctly.

What this means for a hospital in Romania or Moldova

Concrete steps, in the order worth taking them:

1. Take an inventory of the Google profiles of all locations and check that each is claimed and managed by the organisation, not by a former employee.
2. Measure the starting point: response rate and median time for each location over the last twelve months.
3. Write the reply policy on one page: what is allowed, what is prohibited (GDPR Article 9), who signs, within what time. Have it reviewed by the data protection officer.
4. Name an owner and a deputy per location. The response rate usually drops where the task has no name attached.
5. Set the target: 100% of reviews answered, within the few-day window patients expect.
6. Choose the tool: AI draft, human approval, monthly report. Test it for a month on two locations before rolling it out.
7. Close the loop: the themes from the report reach the management meeting, and the measures taken are communicated to the teams.

Questions to ask a software vendor before signing:

• Where are reviews and drafts processed and stored, and under which contractual basis (GDPR)?
• Can the system block publication without human approval? Can that block be switched off, and if so by whom?
• How does it detect health data in a draft, and what does it do when it finds it?
• In which languages does it write, and how well does it write in Romanian and Russian, tested on your own real reviews?
• How does it connect to Google: through the official interface for business profiles, or through unofficial methods that can be shut down at any time?
• Does it log who approved each reply and when, and what does the monthly report per location contain?

The working rhythm in a network

Daily, new reviews from all profiles enter a single queue with a draft reply attached, and the location's owner approves or corrects them the same day or the next. Weekly, the network coordinator picks up any review unanswered for more than three days. Monthly, the per-location report reaches leadership.

Consdinamic builds such systems to order, with its deepest specialisation in healthcare. Its own review platform has been running for more than 15 months in the private Gral Medical network in Romania, across 29 locations, where the review response rate rose from about 30% to 100%. This is the company's own result in that network, not a study, and we present it as such.

Conclusion

Patients read reviews before choosing, filter by the four-star threshold and notice whether the organisation replies. Studies show that responding lifts the rating, brings more reviews and increases trust, provided the reply is personal, prompt and free of any medical information. For a multi-location network, the response rate is the easiest reputation indicator to measure and the only one that depends entirely on organisation. Automation makes it reachable: AI writes the draft in the patient's language, a person approves it, and the monthly report turns reviews into a work list for every location.