Patient reviews do change whether an AI assistant recommends a surgeon, and the mechanism is review text rather than star average. When ChatGPT or Perplexity answers a question about which surgeon in a metro performs a procedure, it retrieves review platform and directory pages and reads what patients actually wrote. Volume, recency, and platform mix matter because they decide how much of that text exists and where it sits.
This is a different mechanism from the one in the ClinicAds post on local SEO versus paid search for surgical practices. There, reviews are a prominence input that holds Google map pack position, weighted as count, average, and recency. Here, reviews are retrievable source text, where the content of the sentence outranks the number attached to it. That is why a practice can hold a strong map pack position and still never be named in an AI answer.
- Patient reviews do change whether an AI assistant names a surgeon, but the working signal is review text rather than star average. A model retrieving an answer to a surgeon-selection query reads the words patients wrote.
- A review that names the procedure and the surgeon is retrievable. A review that says the staff was friendly and the results were great is not, because it contains no entity a model can bind to a procedure query.
- Star average is a weak separator in AI answers. Nearly every surgeon a model considers sits between 4.6 and 4.9, so the rating cannot break the tie the way it does inside Google's map pack ranking.
- Recency matters for a mechanical reason: review platform pages surface the newest reviews first, so the 10 to 20 most recent reviews are most of what a crawler stores and a model later retrieves.
- ClinicAds treats review text as a 3 to 6 month build, not a switch. Rewriting the request template changes what patients write within one post-operative cycle, and the accumulated text becomes retrievable a quarter or two later.
Do patient reviews change whether AI assistants recommend a surgeon?
Yes. Patient reviews change whether an AI assistant names a surgeon, and the deciding factor is what the reviews say rather than the rating they carry. A model answering a surgeon-selection query retrieves pages from review platforms, specialty directories, and the practice site, then assembles named recommendations from the text on them. A profile with 300 reviews that describe nothing and a profile with 90 that name procedures and surgeons are not equally useful to that process.
The confusion comes from treating reviews as a single score. Google's map pack does read reviews as a score. An AI assistant reads them as prose. ClinicAds sees practices that rank in the map pack and are absent from every AI answer for the same procedure, which is only possible if the two systems consume different parts of the same review. Reviews are also one surface among several, ranked in the ClinicAds post on which sources AI assistants cite for plastic surgeons.
Which review signals do models actually read?
A model reads the review text, the entity names inside it, and the recency ordering the platform imposes on the page. Star average and total count reach the model only indirectly, by changing how much text exists and how prominently a profile is displayed. The table separates the signals that move Google map pack ranking from the signals that move whether a surgeon gets named in a generated answer.
| Review signal | Moves map pack rank | Moves AI recommendation | Why |
|---|---|---|---|
| Star average (4.6 vs 4.9) | Yes, directly | Weakly | Google weights the rating; a model finds nearly every candidate surgeon in the same narrow band |
| Total review count | Yes, directly | Indirectly | More reviews means more text, and text is the retrievable part |
| Review recency | Yes | Yes | Platform pages sort newest first, so recent reviews are most of what a crawler stores |
| Procedure named in the review text | No | Yes, strongly | Supplies the procedure-to-surgeon association a model needs to answer a procedure query |
| Surgeon name in the review text | No | Yes, strongly | Binds the outcome to a named person rather than to a practice with three surgeons |
| Platform mix beyond Google | Partly | Yes | Specialty platforms are cited in surgeon-selection answers more often than a Google profile is |
| Owner responses to reviews | Slightly | Slightly | Adds text carrying the practice and surgeon names to the same crawled page |
How is this different from map pack ranking?
Google's map pack ranks a Google Business Profile on proximity, relevance, and prominence, and review count, average, and recency feed prominence as numbers. An AI assistant does not rank profiles. It retrieves documents, reads them, and writes a named answer, so a review helps only if its sentences contain the procedure and the surgeon a reader asked about. The inputs overlap; the units do not.
The divergence shows up at the review request. A one-line five-star review serves the map pack perfectly and is worth close to nothing to a model. Neither goal cancels the other, since the same review can carry five stars and three specific sentences, which is how ClinicAds writes the request.
- Map pack reads: count, average, recency, response rate, all as numbers
- AI assistant reads: procedure names, surgeon names, outcome descriptions, all as text
- Shared input: recency, which moves prominence and controls the top of a crawled page
- Shared cost: the review request itself, which is one message either way
Does a higher star average get a surgeon named?
Barely. Star average is a weak separator in AI answers because the candidate set is already compressed. Almost every board-certified surgeon a model would consider for a metro sits between 4.6 and 4.9, and a model cannot meaningfully distinguish 4.7 from 4.9 when the page describes both as excellent. The rating works as a filter at the low end rather than a ladder at the high end.
Where average does matter is exclusion. A profile below roughly 4.2 produces recent review text describing billing disputes and revision requests, and that text is as retrievable as the positive kind. So the rule is asymmetric: moving from 4.7 to 4.9 buys very little AI visibility, and moving from 4.1 to 4.6 buys a great deal, because it changes what the recent text says rather than what the number reads.
How much review text is enough, and how recent?
There is no threshold count, because the retrievable unit is the review that names a procedure rather than the review. A practice with 400 generic reviews and 6 specific ones has 6. ClinicAds targets a floor of roughly 15 to 25 procedure-naming reviews per primary procedure per surgeon, accumulated inside the last 12 months, before expecting that procedure to surface the surgeon in a generated answer.
Recency is mechanical rather than editorial. Review platform pages sort newest first and paginate, so the 10 to 20 reviews at the top are most of what a crawler stores and a model later retrieves. A profile whose last specific review is 14 months old is, for retrieval purposes, a profile with no specific reviews. That makes velocity the number to manage.
- Floor to aim for: 15 to 25 procedure-naming reviews per procedure per surgeon, inside 12 months
- Velocity to hold: 4 or more new reviews a month on the primary profile
- What a crawler mostly sees: the newest 10 to 20 reviews on the page
- Time to first measurable change: 3 to 6 months from rewriting the request
- Platforms to cover: the Google Business Profile plus at least two specialty or directory profiles
How should a practice rewrite its review request?
A practice should replace the open-ended request with a prompted one that asks for the procedure by name, the surgeon by name, and one specific detail. Patients write what they are asked to write. A message saying tell us about your experience produces the generic review that helps the map pack and nothing else. The five steps below need no new software and disclose nothing about the patient, since the practice is prompting a voluntary public review.
- 1. Merge the procedure name into the request from the chart, so each patient is asked about the operation they had
- 2. Name the surgeon in the request text, which is what gets the surgeon rather than the practice into the reply
- 3. Ask one specific question, such as what recovery was like, instead of asking for a general rating
- 4. Send at the 4 to 8 week follow-up, where the patient is satisfied and still remembers detail
- 5. Rotate the destination across the Google profile and at least two specialty profiles so the text is not concentrated on one page
What is this worth against paid consultation volume?
Review text work does not replace paid acquisition, and it should not be funded out of a media budget currently producing booked consultations at $80 to $150 each. These are agency averages, not guarantees. The rewrite costs the hour it takes to change an automated message, so the trade never has to be made. What it changes is whether the practice appears at all on surgeon-selection queries, which ClinicAds recommends baselining before the rewrite using the prompt-set method in the post on measuring AI visibility.
Do AI assistants read Google reviews directly?
Not as a live feed. An AI assistant retrieves crawled or browsed pages, and a Google Business Profile page is one of them, but specialty directories and review platforms are cited more often in surgeon-selection answers because their pages carry more procedure-specific text per profile. A practice should treat the Google profile as necessary and not sufficient.
Can a practice ask a patient to name the procedure in a review?
Yes. Asking a patient to describe their own experience in a public review they choose to write is a request, not a disclosure. The practice is not publishing anything about the patient. What a practice cannot do is write, edit, or incentivize the review, or reply in a way that confirms someone was a patient beyond what the reviewer already stated publicly.
How long until a rewritten review request shows up in AI answers?
Plan on 3 to 6 months. The first specific reviews arrive within one post-operative cycle, usually 4 to 8 weeks, but the text has to accumulate to the point where a procedure query retrieves the surgeon rather than one stray mention. ClinicAds re-runs the coverage prompt set quarterly rather than monthly for this reason.
Does responding to reviews help AI visibility?
Slightly, and mostly as a side effect. An owner response adds text containing the practice name and often the surgeon name to the same page a crawler reads, which strengthens the entity association. It is a small effect compared with what the patient wrote, and it is not a substitute for prompting better reviews in the first place.
Should a multi-surgeon group build reviews under the practice or the surgeon?
Both, with the surgeon named inside the text either way. A review that says the practice was excellent cannot answer a query about a specific surgeon, and a group with three surgeons and one undifferentiated review base gives a model nothing to allocate. The wider question of which entity to build is a separate decision that turns on succession risk.