September 25, 2026 · 9 min read

Star Rating and Revenue: What the Research Shows

Two Yelp studies tie a half-star bump to real revenue and sell-outs. Here is what they measured, what they did not, and what it means for a Google profile.

Everyone quotes the number and few people read the study

Sooner or later a review-marketing page tells you that one more star is worth "5 to 9 percent more revenue." The figure is real. It comes from a single working paper about Seattle restaurants, and the way it is usually repeated makes it sound like a law of local business.

It is a narrower finding than that, and the narrowness is the useful part. This article reads the two best-known studies of how star ratings move demand, states exactly what each one measured, and then says what a small business can and cannot take from them. Both papers are about Yelp. They are used here as research evidence only. Yelp prohibits businesses from soliciting reviews, and we cover why you must not ask for them.

Study one: Luca, Seattle, revenue

Michael Luca's paper, "Reviews, Reputation, and Revenue: The Case of Yelp.com", is Harvard Business School Working Paper 12-016. It dates from 2011 and was revised in 2016. It is a working paper, meaning it is a draft circulated for comment rather than a peer-reviewed journal article.

The data is unusually good. Luca worked with the Washington State Department of Revenue to get actual revenue for restaurants in Seattle from January 2003 to October 2009. That is 3,582 restaurants in total, about 1,587 open in any given quarter, and 143 of them chain-affiliated. Yelp had, by 2009, more than 60,000 Seattle restaurant reviews covering about 70 percent of operating restaurants. The Seattle Times, by comparison, had reviewed roughly 5 percent.

There are two results, and people blur them together.

1. A correlation. With restaurant and quarter effects controlled for, a one-star increase in rating was associated with a 5.4 percent increase in revenue. Luca himself flags the weakness: a restaurant's rating can move because its real reputation moved, so this alone does not show the rating caused anything.
2. A causal estimate. Yelp shows a rating rounded to the nearest half star. A restaurant averaging 3.24 displays three stars. One averaging 3.26 displays three and a half. Those two restaurants are almost identical in quality, but one looks better. Luca compares restaurants within 0.1 stars of a rounding threshold, a method called a regression discontinuity. The effect is for a half-star display bump. He then rescales it to a full star, and the abstract reports "5-9 percent" depending on the specification.

The paper also reports that the effect is driven by independent restaurants. Ratings did not affect chain-affiliated restaurants, and chains lost market share as Yelp use grew. That fits a plain story: a diner who already knows what a chain serves does not need a stranger's opinion, and a diner choosing an unknown independent does.

Study two: Anderson and Magruder, San Francisco, sell-outs

Michael Anderson and Jeremy Magruder used the same rounding trick with a different outcome. Their paper, "Learning from the Crowd", appeared in The Economic Journal, volume 122, issue 563, in 2012.

Instead of revenue, they measured reservation availability. From July 21 to October 29, 2010, they recorded whether a table for four was open at 6, 7 and 8 pm on Thursday, Friday and Saturday, checking roughly 36 hours ahead. The full Yelp sample held 3,953 San Francisco restaurants. The subset that also had reservation data held 328.

Their headline: an extra half star made a restaurant sell out 19 percentage points (49 percent) more often. At the 7 pm slot, moving from three to three and a half stars went with a 21 percentage point rise in sold-out tables. Availability averaged 74 percent at 6 pm, 59 percent at 7 pm and 68 percent at 8 pm, so these are large shifts from a small change in a displayed number.

The authors also report that the effect was larger "when alternate information is more scarce." They tested this by setting aside 42 restaurants that held a Michelin star or appeared in the San Francisco Chronicle's Top 100, on the logic that diners already have other evidence about those places.

Side by side

What these studies do not show

The honest limits matter more than the headline, because the headline is the part that gets copied.

They measure a jump at a threshold, not a slope. The method compares restaurants sitting just either side of a rounding line. It tells you what a displayed half star did for a restaurant near the line. It does not tell you the value of moving from 3.9 to 4.0 as a smooth quantity, or the value of a rating far from any line.

They are restaurants. A restaurant is a fast, low-cost, repeatable choice made with a phone in hand. A plumber, a dentist or a wedding photographer is a different decision. Nothing in these papers speaks to those.

They are Yelp, with Yelp's rounding. The whole design depends on Yelp displaying a rounded half-star rating. Google's own page on review scores says "The review score is the average of all ratings published on Google for that place or business," and that a new review "may take up to 2 weeks" to change the score. It says nothing about rounding. The page carries no update date, and the only date on it is a 2026 copyright line. Treat this as what the page said when it was read on 2026-09-25. Any claim about Google "rounding to the nearest half star" is not supported by Google's documentation, so the threshold mechanism in these papers cannot simply be assumed to carry over.

The data is old. Seattle to 2009 and San Francisco in 2010 are a long time ago in online-review terms, and both the platforms and shopper habits have changed since. I could not find a 2025 or 2026 study that repeats the design for Google reviews or for non-restaurant businesses, so I have not cited one.

They say nothing about a single review. Neither paper prices one extra review. A business with 800 reviews barely moves its average with ten more. A business with 12 can swing a whole tenth of a point on a single one.

What consumer surveys add

The papers are about behaviour. A survey is about stated preferences, which is weaker evidence, but it covers today. BrightLocal's Local Consumer Review Survey 2026 used "a representative panel of 1,002 US adult consumers via SurveyMonkey." It reports that 68 percent of consumers want at least four stars, up from 55 percent the year before. It also reports that 31 percent will only use a business with 4.5 or more stars, up from 17 percent, and that 10 percent will only use five-star businesses. The page does not show a fieldwork date, so it is cited here by edition only.

A stated cutoff is not the same as a measured choice. But the direction agrees with the two papers: shoppers do use the number as a filter, and the filter has tightened.

The trap: the number can be gamed, and the studies knew it

If a half star is worth a fifth more sell-outs, some owners will cheat. The researchers say so themselves. Anderson and Magruder note the returns "suggest that restaurateurs face incentives to leave fake reviews," and then run a set of checks showing restaurants were not manipulating ratings around the threshold in a way that would spoil their result. Luca makes the same argument about his design.

Elsewhere the evidence on fraud is blunt. Luca and Zervas, in Management Science, volume 62, issue 12, December 2016, found that roughly 16 percent of restaurant reviews on Yelp are filtered as suspicious. They also found that a restaurant is more likely to commit review fraud "when its reputation is weak, i.e., when it has few reviews, or it has recently received bad reviews."

Read that alongside the revenue numbers and the pressure is obvious: the businesses with the most to gain are the ones most tempted to fake it. Faking the number is prohibited by platform policy and by consumer-protection law, and selective solicitation is a separate problem covered elsewhere on this site. The revenue effect these papers found belongs to an honest average. It is not a reason to manufacture one.

An illustrative scenario

The following is a made-up example to show arithmetic, not a real business. A café has 40 Google reviews averaging 4.2, so the ratings add up to 168 stars.

• Ten new five-star reviews bring the total to 218 across 50 reviews: an average of 4.36.
• Ten new four-star reviews bring it to 208 across 50: an average of 4.16.

The first case moves the average up by about 0.16. The second nudges it down. The lesson is not that four-star reviews hurt. It is that at small counts, a batch of honest reviews from ordinary customers moves the average by amounts that matter, and that the direction depends on how good the experience actually was.

What this scenario cannot tell you is what that 0.16 is worth in dollars. The studies cannot tell you either.

What this means for a small business

The findings that survive scrutiny are modest and practical.

• The displayed number matters most where shoppers know least about you. Both papers found the effect concentrated in independent or lesser-known restaurants. The pattern suggests a new local business has more to gain from a good average than a well-known one, though neither paper tested that directly.
• Small counts are volatile. With a few dozen reviews, each new one is a meaningful share of the average. That works in both directions.
• The only legitimate lever is asking every customer. A request that goes to everyone you served produces an average that reflects who you served. A request sent only to people you expect to be happy is the review gating practice Google prohibits. ReviewHero, for one, sends the same emailed request with the public review link to every customer you add.
• Do not turn a research range into a revenue forecast. "5 to 9 percent" describes Seattle restaurants near Yelp's rounding lines. It is not a number for your dentist's office.

What to do next

1. Write down your current Google average and your review count today. With the count in hand you can tell how much a single review would move the average, using the same arithmetic as the café above.
2. Look at the three or four businesses a customer would compare you with. Note their counts as well as their averages, since a high average on a handful of reviews reads differently from the same average on hundreds.
3. Ask every customer you serve for a review with the same message and the same link, whatever you expect them to say. Our guide to getting your first 50 reviews covers the practical steps.
4. Re-check your score in two weeks, not two days. Google's own page says an updated score can take up to two weeks to appear.
5. Leave any claimed revenue figure out of your own marketing unless you can trace it to a study and describe exactly what it measured.

This article is general information and not legal advice. Platform rules and consumer-protection law vary and change, so check with a lawyer about your own situation.