Review Count Is the Signal. Your 4.9 Is Not.

On September 10, Search Engine Journal ran a study from the listings platform Uberall built on 120,000-plus AI mentions across 3,793 locations and five engines — ChatGPT, Claude, Gemini, Grok, and Perplexity. Nine to eleven real prompt intents per vertical, each run 50 to 100 times, across dentists, restaurants, grocery stores, hotels, and banks. It is a vendor study, published as sponsored content, and the vendor sells the fix. Read it with that in mind.

Then read the finding anyway, because it is the kind of result that survives the discount: across all five verticals, review volume predicted whether an AI engine named a business. Star rating did not reach significance in four of the five.

That is a direct contradiction of how most practices have managed their reputation for a decade.

The Dental Numbers Are the Closest Analog

Immigration law was not in the study. Dentistry was, and it is the nearest thing in the dataset to a small professional-services practice — a licensed practitioner, a local catchment, a decision a person makes once and researches anxiously.

In the dental data, no star rating on any platform reached statistical significance. None. What separated practices that got named from practices that never did was how many reviews they had: an average of 643 for the named, 253 for the unnamed. Practices past 1,000 Google Business Profile reviews were named 92.9% of the time. Zocdoc volume was the only other individual review platform where the count significantly predicted mention probability.

The grocery numbers make the tradeoff even starker. Brands with high review volume and lower ratings were mentioned 94.3% of the time. Brands with high ratings and low volume: 60.6%. Hotels were the one exception, where Google star rating outperformed count — and hotels are also the vertical where a rating is the product.

Why Count Beats Rating

The mechanism is not mysterious once you stop thinking about the rating as a quality score and start thinking about what a model can do with it.

A 4.9 average is a single number with almost no variance across a market. Every established practice in your city is somewhere between 4.6 and 5.0. As a discriminator, it is nearly useless — it separates no one from anyone.

Review count does two things a rating cannot. It functions as a verification signal: a profile with hundreds of reviews has cleared spam filters, survived moderation, and accumulated activity over years, which tells the model the entity is real and currently operating. And the reviews themselves are text — the descriptive language a model draws on when it has to say something specific about you. A practice with 600 reviews has 600 unprompted descriptions of what it actually does. A practice with 40 five-star ratings and no substance has a number.

Uberall's own framing is the useful one: optimize volume for the models, keep the rating for the people. Humans still read the stars. Machines are reading the corpus.

The Immigration Problem Nobody Else Has

Here is where the study stops transferring cleanly, and where immigration practices need their own answer.

A dental patient will post a review without a second thought. An immigration client frequently will not — and often should not be asked to. A client in removal proceedings, a pending asylum applicant, a family whose adjustment of status is sitting at a service center: publicly attaching their name to your firm identifies them as someone with an immigration matter. For some clients that is merely uncomfortable. For others it is a real exposure.

That is why immigration firms systematically carry lower review counts than comparable practices in family law or personal injury, and why a firm that has been proud of its 4.9 across 38 reviews is not lazy — it is being careful. The study says that care has a cost in AI visibility, and pretending otherwise does not help anyone.

The practical middle ground is knowing who you can actually ask. Naturalization clients whose case has concluded. Employment-based clients — the H-1B beneficiary who got approved, the PERM case that cleared audit, the O-1 that came through without an RFE. Consular processing families whose visa is in hand. Corporate HR contacts who manage your firm's filings and have no personal exposure at all. That is a substantial share of most practices' closed files, and it is almost never asked.

And before any of it: check your state bar's advertising rule. ABA Formal Opinion 496 addresses responding to online criticism, and the short version is that a negative review does not open the door to disclosing anything about the representation. Model Rule 7.2(b) limits what you can give in exchange for a recommendation. Asking a satisfied client for an honest review is generally permissible; paying for one, scripting one, or answering a bad one with case facts is not.

Do This Week

1. Count, don't average. Pull your total review count on Google Business Profile, Avvo, and Yelp. Write down the raw numbers, not the stars. If you are under 100 on Google, that is your finding.
2. Sort your last two years of closed files into askable and not-askable. Concluded naturalizations, approved employment-based petitions, completed consular cases, and corporate contacts go in the first column. Anything pending, anything in proceedings, anything involving asylum or humanitarian relief stays out.
3. Confirm your state's rule before you send anything. Bar advertising provisions vary meaningfully, and a solicitation template that is fine in one jurisdiction is not automatically fine in another.
4. Build one request into your closing process. Not a campaign — a step. The approval email, the certificate-delivery appointment, the final invoice. A firm closing 15 askable matters a month that converts a third of them adds 60 reviews a year without a marketing budget.
5. Claim the profiles you have been ignoring. The study found nearly two in three banks had never claimed their TrustPilot page. The legal equivalents — Avvo, Justia, Martindale, your state bar's directory listing — are the same unclaimed surface, and they are read.
6. Log the count and check again in 90 days. Volume is a slope, not a number. One reading tells you nothing.

The Wider Point

The instinct with a study like this one is to find the reason it does not apply. It is sponsored. Legal was not a vertical. Dental patients are not asylum clients. All true.

But the argument underneath does not depend on the sample. A model recommending a practice has to justify the recommendation from evidence it can read, and a star average is the thinnest evidence on the profile — one number, identical across every competitor, attached to nothing. Hundreds of reviews written in human language about specific matters is the thickest. Whatever the exact coefficients turn out to be for legal services, that asymmetry is structural.

Which means the firm guarding a perfect rating across a handful of reviews has been optimizing the variable that does not move, and treating the one that does as an afterthought. The fix is not clever and it is not expensive. It is asking — carefully, of the right clients, every single month.

Clientory helps immigration law firms become visible to the families and individuals who need them most — turning AI search into consultations, and consultations into clients.

See Your AI Visibility Score → CLIENTORY.ORG