Do online reviews affect AI recommendations for law firms?

Do online reviews affect AI recommendations?

Reviews affect AI recommendations, and the clearest evidence is how few businesses clear the bar. SOCi’s 2026 Local Visibility Index, published on February 17, 2026 and covering 2,751 brands across roughly 350,000 locations, found ChatGPT recommends 1.2% of locations, Perplexity 7.4% and Gemini 11%, against 35.9% surfacing in Google’s local three-pack. The locations ChatGPT recommended averaged 4.3 stars.

The shape of that gap is what makes reviews a filter. Google’s three-pack ranks a large eligible set and shows the top three. An answer engine does the opposite: it assembles a very small set of businesses it considers safe to name, then names one or two. A firm below the review floor is not ranked lower, it is left out before ranking happens, which is why review work produces a step change rather than a gradual climb.

One caveat belongs here. The SOCi index measures multi-location consumer brands across verticals, not law firms, and no published study has set a legal-specific review threshold. Firm AEO treats 4.3 stars as the best available proxy for the floor rather than a measured figure for legal services.

How does a review on Google, Yelp or Avvo reach an AI engine?

A review reaches an answer engine by one of three routes, and the route decides which engines ever see it: the platform’s position in a search index, a licensing agreement between the platform and the engine, or the engine’s own crawl of a page carrying the rating. Most law firms work on one platform and assume all three routes are covered.

Google reviews take the first route and travel unevenly. Gemini and Google AI Overviews read Google’s index and Google Business Profile directly. Martindale-Avvo’s April 2026 analysis of its own legal query data found ChatGPT matching Google’s top ten results less than 25% of the time for legal queries, with Gemini near 50% and Perplexity and Claude near 75%.

Yelp takes the second route. SubscribePR’s July 22, 2026 analysis of reviews and AI recommendations points to Yelp’s data licensing with OpenAI and Perplexity as the reason Yelp content appears in engines that otherwise pass it over, which matters for a firm that never claimed its Yelp listing. Avvo, Martindale-Hubbell and Super Lawyers take the third route: the engine retrieves the directory profile and the star rating arrives as part of that page.

What rating, how many reviews, and how recent?

The published floor is 4.3 stars, recency inside the trailing 90 days, and a response rate at or above 5%. SOCi supplies the rating average and the response rate. SubscribePR’s July 2026 analysis argues recency inside a trailing 90-day window carries more weight than total review count, because an engine re-reading a profile sees a date before it sees a total. No engine publishes a threshold of its own, so every number here is a reading of third-party measurement.

Recency beats volume for a mechanical reason. Review count is a lifetime total a profile keeps whether the firm is still taking matters or not. A dated review stream is the only review signal that also says the business is currently operating and is still the business the profile describes. Two hundred reviews with nothing newer than 2024 reads as an archive. Forty reviews with six from the last quarter reads as a going concern. The targets below are illustrative starting points, not promised outcomes.

• •Rating at 4.3 or above on Google, the floor the SOCi data implies rather than a target to stop at.
• •At least one new review inside the trailing 90 days on the primary platform, every quarter.
• •A response rate above the 5% SOCi associates with recommended locations, with 100% as the practical standard.
• •Name, address and phone number identical across Google, Yelp, Avvo and the firm’s own site, since a mismatch splits one entity into two weak ones.
• •Review counts in the same order of magnitude as the firms currently named for that practice area and city.

Does replying to reviews change anything?

Replying changes two things. SOCi found recommended locations holding a response rate of at least 5%, which makes the reply part of the pattern engines read. The reply is also the only text on a review platform the law firm writes itself, so it is the one place a firm can restate its practice areas and jurisdiction in language an engine can lift.

Confidentiality is what makes legal replies different from other industries. North Carolina’s guidance on law firm review responses, reported by North Carolina Lawyers Weekly, holds that a reply must stay professional and restrained and must not disclose information relating to the representation, the duty ABA Model Rule 1.6 imposes. A reply confirming that the reviewer was a client can itself be a disclosure. The safe reply acknowledges the feedback and offers an offline channel, and the firm’s ethics counsel approves the standing template. ABA Formal Opinion 512 of July 29, 2024 extends the same competence, confidentiality and supervision duties to an AI-drafted reply.

Why a strong Google rating moves answers that never cite Google

Because reviews are consumed as an attribute of an entity, not as a quotable passage. An engine writing about a law firm needs sentences it can attribute, and a star rating is not a sentence. The rating is used earlier, when the engine decides which firms belong in the answer, and the citation then goes to whichever page supplied the prose. Citorian’s five-engine test of June 2026, covering 359 answers, put Google at 12% of cited sources, Reddit at 8% and Avvo at 5%, well below the directories that supply descriptive copy.

That is the trap in reading a citation list as a scorecard. A firm can improve its review position over a quarter, watch its appearance rate rise, and find no new review platform in its citation list, because the gain happened at candidate selection rather than at citation. Firm AEO measures appearance rate per engine across an agreed question set, run logged out and repeated, because review work shows up in that number and is invisible in the other one.

What can a law firm do about reviews without breaking a rule?

Everything above is permitted. Manufacturing the signal is not. The Federal Trade Commission rule banning fake and incentivised reviews took effect on October 21, 2024, with penalties reaching $51,744 per violation, and ABA Model Rule 7.1 separately prohibits a false or misleading communication about a lawyer’s services. Paying for reviews, gating negative ones before they post, or having staff write them is an ethics exposure before it is a marketing one.

The rules on asking differ by jurisdiction, so a Canadian practice cannot copy an American programme. Ontario, Alberta and British Columbia permit a genuine request to a consenting client. Quebec is the outlier: article 145 of the Code de déontologie des avocats bars testimonials outright, which extends to republishing Google reviews on the firm’s own website. A firm practising in more than one province lets the stricter rule govern the shared template, and its ethics counsel signs off before any request goes out.

The sequence that works is slow. Claim and correct every profile so the entity is one entity. Fire a review request at the same point in every closed matter. Reply to everything within a week, in the restrained language the confidentiality duty requires. A consumer firm starting from a few dozen reviews should expect two to three quarters before the profile reads differently to an engine, which is a typical pattern rather than a commitment.