One in three AI Overview citations lands on a law firm site with a chat widget

Ask an AI engine for a personal injury lawyer in New York, click the firm it cites, and there is a fair chance the first thing you meet is a recorded attorney, in a box in the corner of the screen, saying hello and asking what kind of matter you have. On some sites the box is a plain chat window with an avatar. On a smaller number it is a video that starts playing on arrival. Both are intake widgets, installed to turn a visitor into a phone call, and both are invisible to the engines that sent the visitor. We wanted to know how much of the AI referral traffic in our market lands on one.

We counted the widgets

In August we fetched the homepages of 2,626 New York law firm websites and could read 2,365 of them, the same census behind our AI-readiness study. Every homepage that embeds a chat vendor leaves a script tag in the page, so we fingerprinted nineteen vendors and found a widget on 257 homepages, 11 in 100. Two vendors dominate the legal market and both sell a video mode: Intaker, on 73 sites, and Juvo Leads, on 28.

A script tag says the widget is installed, not what it looks like. So we loaded all 101 Intaker and Juvo Leads homepages in a real browser and checked whether the video mode was on. It usually was not. Of the 64 Intaker widgets that loaded, 32 opened with a video and 32 with a text chat. Of the 27 Juvo Leads widgets, 3 opened with a video and 24 with a chat bubble. The video greeter, the thing that prompted this article, sits on 35 homepages in the whole market, 1.5 in 100.

Rosenberg & Rodriguez, the most-cited video greeter site in our window, on Intaker

Hill & Moin, the most-cited Juvo Leads video greeter, which hands off to a human chat agent

Screenshots of the public homepages as a desktop visitor sees them, September 2026. Both firms are named because both lead their group in AI citations.

Where the AI referrals land

We then matched every census site against seven weeks of AI answers to the questions New Yorkers type when they need a lawyer, on ChatGPT, Gemini, and Google AI Overviews. That gives each site a citation count per engine. Summed across the market, a large share of the citations point at a site with a chat widget:

Eleven in 100 homepages carry a widget, but 34 in 100 AI Overview citations and 31 in 100 Gemini citations land on one. The video greeter specifically catches 4 to 7 in 100, depending on the engine, with ChatGPT the most likely to send you to a talking head. Read as an experience: a client who follows an AI Overview to a firm will meet a chat window one time in three, and a recorded attorney one time in twenty-five.

Widget sites get cited twice as often, and the widget is not why

The site-level version of the same fact is the one that tempts a marketer. Of the 35 video greeter sites, 19 were cited at least once in the window, 54 in 100. Of the 222 sites with any other chat widget, 95 were cited, 43 in 100. Of the 2,108 sites with no widget, 466 were cited, 22 in 100. Both gaps clear any reasonable significance bar. If the story ended there, the advice would write itself, and it would be wrong. The engines never execute the widget script. They read the page's text and markup. Whatever is going on happens somewhere else on the page.

The first clue is site size. Split the market by page count and the widget advantage lives only among large sites:

Among sites under 100 pages, a widget buys nothing: 4 of 35 widget sites were cited against 135 of 934 without. The gap opens past 100 pages and is widest past 1,000. The widget is riding on something else, and the portrait of a widget site says what:

A video greeter site has six times the pages of a no-widget site, two and a half times the homepage text, twice the rate of legal structured data, and nine times the rate of call tracking. It is a site somebody pays a marketing agency to run, and the widget is one line item on that invoice. When we fit a model that accounts for site size, homepage text, structured data, call tracking, and practice area at the same time, neither the video greeter nor the plain chat widget predicts a citation anymore. The gap shrinks to well under two to one and stops clearing significance. Text volume and legal structured data carry the prediction on their own, each roughly doubling the odds, which is the same pair that led our cited-sites portrait. The widget is a marker of the kind of firm that gets cited, and nothing in the data says it is a lever.

What this means for your firm

• Do not buy a chat widget for AI visibility. The engines never see it. Buy it, if you buy it, for what it does after the visitor arrives: 57 in 100 video greeter sites also run call tracking, which is the tell that these firms measure the phone.
• Your AI referrals are landing on somebody's widget. A third of AI Overview citations in our market point at a site that opens a chat window on arrival. If yours does not, the comparison a client makes in the next tab is a firm that answered before they finished reading.
• The video greeter is rarer than it feels. 35 homepages in a market of 2,365, and half the firms paying for a vendor that offers it leave it off. It reads as everywhere because it sits on the big, cited, personal injury sites you land on most: among personal injury firms with a video greeter, 14 of 19 were cited, against 182 of 730 without.
• Spend on what the model actually rewards. More pages that address the problems your clients bring, and legal-service markup on them. That is the pair that survives every control we ran, in this study and the last two.

One more thing about arrival. A widget greets whoever shows up, and increasingly that will not be a person. When a client asks an assistant to find and contact a lawyer, the assistant lands on the same homepage, meets the same video, and looks for a form it can fill. In this census, only 30 in 100 homepages had a contact form with no captcha in front of it, and widget sites were slightly more likely to run one. The door is narrow for software too.

Usual caveats: one metro, one seven-week window (July 8 to August 21), homepage-only fingerprints so every widget share is a floor, and the answers to our tracked searches rather than every question anyone could ask. Video mode was verified in a headless browser for Intaker and from the firm's own tag script for Juvo Leads; nine Intaker widgets did not load during our check and are counted as installed, not as video. The other seventeen vendors were not checked for a video mode. All of this is correlation measured carefully, and none of it says that adding or removing a widget changes a single citation.