Most operators track reviews as a number. Star average, count, maybe a trend line in a dashboard. The number is what shows up in listings, so it feels like the thing that matters.
When an assistant answers "is Marino's any good for a group dinner," it does not report your average. It writes a paragraph. That paragraph comes from sentences people wrote, and the sentences that get picked are the specific ones. "Great food but the service is slow on weekends" survives into a summary. "5 stars, love this place" does not, because it says nothing that can be turned into a claim.
This changes which reviews are worth caring about, and it changes what a good review response looks like.
Specific text outranks a general rating
Think about what a summarizer has to work with. It has a rating it can cite as a rating, and it has a pile of prose. Prose is where the answer to any real question lives, because real questions are specific. Is it good for kids. Do they have parking. Can you get in without a reservation. How is the gluten-free menu. Does the takeout travel well.
A restaurant with a 4.6 average and forty reviews that all say "amazing" will lose those answers to a 4.3 with forty reviews that describe the patio, the wait, the wings, and the fact that they answer the phone. The second restaurant supplied material. The first supplied a number.
This is why chasing the average as a target is less useful than it looks. The average moves slowly and says little. The text moves with every new review and says everything.
The sentences you do not want repeated
Certain complaints are unusually damaging in a summarized answer because they are concrete, repeatable, and about something a person can check. Slow service, dirty bathroom, and wrong order are the classics. So is "nobody ever answers the phone," which shows up in restaurant reviews far more often than owners think and reads as an operational fact rather than an opinion.
The reason that last one lands hard is that a reader has no way to discount it. A complaint about a dish is a matter of taste. A complaint that the phone rings out is a claim about your operation, and if three people say it, a summarizer will state it as characteristic. It is also, unlike most review complaints, entirely mechanical to fix. The math on what those unanswered calls are costing you before anyone writes about them is in the real cost of a missed restaurant phone call.
The pattern worth internalizing is that operational complaints compound in a way food complaints do not. One person disliking the carbonara is noise. Four people describing the same broken process is a finding.
Which suggests a use for your reviews that has nothing to do with marketing. Read them as an operations report. If the same friction appears in reviews across two platforms and six months, you have a documented problem that customers cared enough to write about, which is a stronger signal than most of what reaches you through staff or comment cards.
Responses are content, and most of them waste the slot
An owner response is published text sitting directly beneath a review. It gets read alongside it. Which means a response is your only chance to put your own sentence into the same material a model is reading.
Almost nobody uses it. The standard response is some version of "We're so sorry to hear about your experience. Please reach out to us directly so we can make it right." That is polite and it is empty. It adds no fact.
A response that adds a fact does real work:
- Naming what changed, as in "we added a second person to the phone on Friday and Saturday nights starting in March," which turns a complaint into a dated problem rather than a standing condition
- Correcting a factual error plainly, such as a review that says you do not take reservations when you do, without arguing about the rest of the review
- Confirming something the reviewer got right and adding the detail they missed, like the fact that the patio is heated through the winter
- Answering the question the reviewer implied but did not ask, since other readers have the same question
Do this on the reviews that describe a fixable operational thing. Skip it on the ones that are just unhappy. You are not going to convert the person who wrote it, and the audience you are writing for is everyone who reads it afterward, human or otherwise.
Where markup helps and where it does not
Marking up reviews on your own site with structured data makes the ratings and text machine-readable there. It does not change what a platform publishes about you, and it should never be used for reviews you collected selectively or wrote yourself. Review schema for restaurants covers what is legitimate and what gets a site penalized.
Generating the reviews you want without scripting them
You cannot dictate review content, and platforms are increasingly good at spotting coordinated language. What you can do is ask at the right moment.
The right moment is immediately after the specific experience you want described. A guest whose fifty-person catering order arrived on time and complete will write about catering if you ask them that afternoon. Ask them a week later by email blast and you get "great food, thanks." The ask has to be attached to the thing.
Volume of recent, specific reviews is the actual goal, because it dilutes old text and gives a summarizer current material. Ten detailed reviews in the last two months do more for how you get described than a hundred old ones.
Where this connects to the phone: if you use a voice agent, it can be the thing that follows up, and it can also be the reason a caller has a good experience worth writing about. It can also become the subject of a review, which is its own conversation covered in handling negative reviews about AI answering and telling customers you use AI.
Run this check once. Open your three most-used review platforms, sort by most recent, and read the last twenty reviews as prose rather than as ratings. Write down every operational claim someone made about your restaurant. That list is what an assistant has to work with when somebody asks about you, and unlike your star average, most of it is something you can go fix this month.