Google Shows You 1.84% of the Law Firm Reviews It Counted

What we measured

A law firm's Google Business Profile shows one number in large type and a column of reviews underneath it. Most people read the number, scroll a little and decide. This piece is about whether the column underneath is a fair sample of the number above it.

We hold two different facts for every firm here, and keeping them apart is the whole exercise. The first is the rating Google publishes on each listing, averaged across a firm's listings and weighted by how many reviews each one carries. That is Google's own average over every review it counted. The second is the set of individual reviews the Places API hands back when we ask, each with its own star rating and its own date.

The second is capped at five per business listing. A firm with one listing gets five. The firm here holding the most listings has 29 of them, so it can get up to 145. No firm in this directory came back with more than that cap allows, which we checked rather than assumed. Google's documentation describes the reviews it returns as its most relevant ones, not as a random draw and not as the most recent. There is no parameter to ask for either.

So across 1,386 firms in 19 markets and 13 states, Google counted 682,994 reviews and gave us 12,586. Everything below is what those 12,586 look like next to the 682,994 they came out of.

What comes back

11,936 of the 12,586 returned reviews rate five stars, which is 94.8% of them. 473 rate one star. Everything between those two poles, the two, three and four star reviews, comes to 177 reviews in total.

The obvious thing to say next is that no real population looks like that, so the selection must be inflating it. We checked, and that is not what the data says.

Pool every returned review and the mean is 4.83. Take the rating Google publishes for each firm, weight it by how many reviews that rating actually covers, and the mean is 4.83. The two agree to within 0.00 of a star.

So the returned reviews are not, in aggregate, better than the population they came from, and this page is not going to claim they are. A published average of 4.83 across 682,994 reviews is consistent with a population that genuinely is overwhelmingly five stars. The thin middle above may simply be what law firm reviews look like.

That check is what makes the rest of this piece worth reading. The problem with this sample is not that it flatters the industry. It is that it decides, one firm at a time.

Firm by firm, it decides

An average that is right across 1,386 firms can still be wrong about every one of them. So compare the two numbers per firm instead: the mean of the reviews Google returned for that firm, and the rating Google publishes over all of that firm's reviews.

A published listing rating carries one decimal, so a gap under a tenth of a star is inside the rounding and we count it as agreement. Past that, the split is not even.

The far tails run against the firms, not for them. 27 firms are flattered by half a star or more, while 95 firms are undersold by that much. A firm is more likely to be badly misrepresented downward than upward, which is the opposite of what a story about Google showing off its businesses would predict.

The sharpest version of this owes nothing to rounding. 1,075 of the 1,386 firms came back with a set in which every single returned review rates five stars. Only 441 firms here average exactly five. So for 646 firms, a caller asking Google for reviews is shown a perfect record that the same profile's own rating does not support.

The furthest case of that is Talmud Law, PLLC in Buffalo: every review Google returned for it rates five stars, while the rating printed on the same profile is 3.7 across 30 counted reviews.

It runs the other way too, and that matters for the argument. Sanabria & Associates is shown worse by its returned reviews than by its own rating: 1.80 against 4.27 across 577 counted reviews. An unrepresentative sample is not the same as a flattering one, and a firm can lose from it as easily as gain.

That is the finding in one line. Across the directory the sample is honest. About any particular firm it is not evidence. Neither of those sentences is much use to somebody reading one firm's profile, which is the situation every client is actually in.

Newer, too

Every returned review carries a date, and the dates skew hard. 8,243 of the 12,586 dated reviews Google returned, 65.5% of them, were published in 2026. The oldest in the set dates from 2012.

We cannot check this one the way we checked the ratings, because Google publishes no date alongside a firm's review count. So this is an inference and we will label it as one: a body of reviews accumulated since 2012 is unlikely to be 65.5% written in a single year, and the year is not over.

If that inference holds, the consequence is plain. A firm that collected nothing this year and a firm that collected two hundred can both come back looking current.

What this costs our own score

Here is the part that is about us rather than about Google.

Two sub-factors of the LFL Certification Score are computed from exactly this sample. C3 rewards review recency, worth 5 points, and it reads the share of returned reviews dated inside the last twelve months. C5 rewards a low complaint share, worth 4 points, and it reads how many returned reviews rate one or two stars. That is 9 points of the hundred, resting on 1.84% of the evidence, chosen by somebody else.

1,007 of 1,386 firms take full marks on recency. That looks like an industry keeping its reviews fresh until you set it beside the 65.5% figure above, which says the returned set is recent across the board. If that skew is Google's rather than the profession's, and we think it is, then C3 is measuring the selection at least as much as it is measuring the firm.

C5 has the same weakness, and it is the per-firm one rather than a claim about the industry. 1,123 firms take full marks because not one returned review rated one or two stars. For a single-office firm that verdict rests on five reviews, so a practice with hundreds of reviews and a real pattern of complaints inside them can return five clean ones and score exactly what a practice with no complaints scores.

We are not removing either sub-factor on the strength of one study, and we are not going to pretend the reading is stronger than it is. The evidence line on every firm's profile already says this measurement comes from a sample of five per listing rather than from every review, and it says so because the engine was written that way from the start. The full methodology sets out each sub-factor and the evidence behind it. This is a live limitation and it is now on a page a reader can find.

It is the same species of problem as the one we published about the denominator: two firms can share a score without sharing the evidence under it, and a sub-factor read from somebody else's selection is a thinner kind of evidence than one read from a register.

What this does not show

This is a study of a selection, not of honesty. Nothing here says a returned review is fake, bought or solicited. Every one of them may be a real client writing what they thought. The finding is about which real reviews get handed to whoever asks, and that is a different question from whether the reviews are true.

We also cannot see the reviews that were not returned. The 682,994 is Google's own count and we have 12,586 of them, so every statement above about the population is made through the published average rather than by reading the population directly. If Google's published average is itself filtered, this page cannot tell.

Nor is this peculiar to Google. Directories that carry lawyer ratings generally decide what is shown rather than showing everything: Martindale-Hubbell, whose peer rating is initiated from references the rated attorney submits, also lets a subscribing attorney choose which written reviews sit at the top of their profile. That is a stronger form of the same thing, and it is chosen by the firm rather than by an algorithm.

There is a second reason to weigh the count carefully. We measured the whole directory against how much search traffic each firm gets, and the review count rises steeply with visibility while the rating does not move at all. A large count is partly a measure of how many people a firm's marketing reached.

One practical consequence for anybody reading a firm's reviews. The five reviews on the screen are the weakest evidence on the page. The count is worth more than the rating, and the rating is worth more than the reviews you are shown, which is the reverse of the order most people read them in. We measured what the count and the rating are each worth across the New York injury market, and the short answer is that the count mostly tracks how many offices a firm runs.

Reviews are one pillar of five, and the score deliberately does not let any of them carry a firm on its own. You can see every sub-score and the evidence behind it on any profile, or compare firms directly on the personal injury and New York personal injury pages.

Review counts, published ratings and the returned review sample all come from the Google Places API. A firm whose listings we could not verify keeps these sub-factors pending rather than receiving an estimated score. If a firm believes a measurement is wrong, the profile links to the source we used, and a firm can add or correct its listing.