Research

We traced 9,049 AI answers to New York car accident questions back to their sources. ChatGPT reads the state and the directories, Gemini reads firm websites, Google AI Overviews reads YouTube and social. Our client William Mattar Law Offices gets 55%, 57% and 21% share of voice on the three.

The full question set we run daily for a New York motor vehicle firm: 51 ask AI to find a lawyer, 26 ask how a claim works, 10 ask whether a lawyer is worth it, 8 compare firms, 16 name one firm. Grouped by intent, with the high-volume ones marked.

ChatGPT reads the state, then the directories, then the firm. What it retrieves in 3,330 answers and what a firm changes to be named.

Google AI Overviews read YouTube, firm sites and social, not the state or the directories. What 2,389 answers show and what a firm changes.

Gemini reads law firm websites and cites them heavily. What 3,330 answers show about its source diet and what a firm changes to be named.

Every ranking of law firm AEO and GEO agencies is written by an agency in it. So instead of a ranking: five tests to apply before you sign, what First Page Sage, AEO Engine, Rankings.io and the others say about themselves, and how SourceNode holds up against the same tests.

More people ask AI first, and fewer searches end in a click. The visitors who still click convert at 4.4 times the rate of those from a search result. What the number measures, why AI search converts better, and how we turn those visitors into customers.

We traced 4,222 AI answers in one category back to the sources behind them. Models repeat the claims that agree across several sources and drop the rest. This is what the data shows, and what it means for a business that wants to be named.

Three numbers describe a business’s standing in AI answers. They measure different things, move for different reasons and are easy to mix up. These are the definitions we use, with worked numbers from a live engagement.

Every answer about a business is assembled from six kinds of source. This is the checklist we use to audit them, the measure we track for each, and the order we work them in.

An engagement log for one client in the legal category: what was changed on each source, month by month, and how share of voice and visibility responded. Same 60 prompts, same 14 competitors, measured daily.

We run the same loop on ourselves that we run for clients. This is what AI models should be able to say about SourceNode, where each fact lives, and how we keep it consistent. Published so you can check it.

Buying research now starts with one question to an AI model, and the two or three businesses in that first answer tend to be the only ones compared afterwards. Here is what our data shows about the first query, and how a business gets into it.

The category we work in, defined: what AI search identity management covers, what it is not, and the measures that tell you whether it is working.

A business’s identity is the set of facts its sources agree on. How that identity shapes an AI answer, where a business can influence it, and how customer intent changes the answer at each stage.

The models name businesses whose facts agree across every source they read. What our tracking of 4,222 answers shows about how that shortlist forms, and the order to work in.

Three numbers describe your standing in AI answers, and none of them is traffic. What to measure, how to build a prompt set that means something, and how to get from mentions to revenue.

There are three reasons AI search leaves a business out, and they need different fixes. How to tell which one is yours, from the sources behind the answers rather than from guesswork.

Across 4,222 answers in one category, 46 third-party domains carried most of the citations and the client’s own site carried a fifth. What that concentration means for where the work goes.

Visibility, share of voice and position describe standing, not value. How to trace an AI answer through to a lead, and why the traffic behaves differently from search.