Why ChatGPT, Gemini, and Google disagree about lawyers

The short answer: ask ChatGPT, Gemini, and Google AI Overviews the same lawyer question and you will usually get three different answers. In our Brooklyn personal injury tracking, each engine's most-named firm is a different firm, and that has held in every monthly window we have measured. It is not a glitch. The engines read different sources, weigh them differently, and have visibly different taste. There is no such thing as ranking well on "AI", only on each engine separately.

How we know

Ayo asks ChatGPT, Gemini, and Google AI Overviews the questions New Yorkers type when they need a lawyer, every day, many times per question, and records every firm each answer names and every page each answer cites. We ask the consumer products themselves, not the vendor APIs (why that distinction matters). Numbers below are from our July 2026 Brooklyn personal injury window unless stated; we re-checked the headline claims against the 30-day window ending August 11 before publishing, and they hold.

Three engines, three winners

Across the same set of tracked Brooklyn personal injury questions in July 2026, about 120 different firms were named at least once, and the three engines crowned three different leaders: ChatGPT most often named Silberstein & Miklos, Gemini most often named Rosenberg & Rodriguez, and Google AI Overviews most often named Shulman & Hill. Not one engine agreed with another about who the front-runner was. The leaderboard is public, with per-engine detail on every row.

Each engine has a taste

ChatGPT rewards the neighborhood

Hold the phrasing constant and swap only the place, and ChatGPT swaps the firm. In our first widened sampling day, the Williamsburg version of a personal injury question most often surfaced Stein Law Group, the Astoria version Marc S. Albert, and the Manhattan version named The Platta Law Firm in every sampled answer. ChatGPT reaches for whoever looks anchored to the specific place in the question, which is an opening a small firm with a strong neighborhood footprint can win.

Gemini rewards the household name

The same questions on Gemini kept returning large established brands regardless of place: Jacoby & Meyers appeared in 80 to 100 percent of sampled answers in four of the five boroughs we asked about. Gemini leans on brand-scale signals, so a firm can dominate its neighborhood on ChatGPT and barely register on Gemini at the same time. Both readings are stable patterns across our samples, not one-off quirks.

AI Overviews rewards the well-built website

Google AI Overviews behaved like neither chat engine: its most-named firms were SEO-strong local practices. Chopra & Nocerino, for example, was named in every sampled AI Overview answer for both Brooklyn and Queens on that first day. AI Overviews is assembled from Google's own retrieval, so it inherits the tastes of Google search rather than the tastes of a chat model.

They also trust different pages

The disagreement runs deeper than the names: the engines cite different parts of the web. In our data, all three engines cite law firm websites more than any other kind of source, legal directories are a low-single-digit slice on each (the full citation breakdown), and the specific pages behind the answers barely overlap from engine to engine. Independent research finds the same split inside Google itself: a SISTRIX study of 1.5 million answer snapshots found that AI Overviews and Google's AI Mode cite different domains 83 percent of the time for identical queries. Two AI products from the same company, reading the same web, disagree five times out of six about which sources matter.

Even one engine disagrees with itself

Phrasing splits answers too. When we ran the same Brooklyn car accident question in three phrasings, the most-named firm flipped between phrasings on ChatGPT. And whoever leads an engine today may not lead it next month: the leaders above are measured over stated windows because AI answers change, measurably, within weeks.

What a firm should do with this

• Stop optimizing for "AI" as one thing. A plan that treats the engines as one channel will read as working on one and failing on two.
• Play each engine's game. For ChatGPT, publish substantive neighborhood-level pages on your own site (how ChatGPT builds its recommendation). For AI Overviews, invest in the pages Google search already ranks, your own site first, with complete directory profiles as cheap insurance. Gemini's brand lean is the hardest to move quickly, which is exactly why the other two are the opportunity.
• Measure all three separately. A single spot-check on one engine tells you almost nothing about the other two, and per-engine share is the number that moves when your work lands.