There is a lot of confusion among law firms trying to measure their AI visibility. One platform may name your firm, another may skip it, and a third may list firms you have never heard of. It looks like noise.

The spread is real. Across nearly 350 firm mentions on four AI platforms in Atlanta, 69% of the firms named showed up on just one platform. But it is less random than it looks. Across the two markets studied so far, Nashville family law and Atlanta personal injury, two things decided which firms got named far more than any single platform's quirks. The first is whether the platform shows a map. The second is how specific the question is. A map closes the shortlist to a handful of firms. A specific question opens it back up, and the platforms start agreeing with each other. Both patterns are worth checking in your market, whatever you practice, because both change where your firm should spend its effort.

Google's local pack (the three map listings that sit above the regular results) looked almost identical in both markets. In Nashville family law, six searches produced a pack, filling 18 slots with just five firms. In Atlanta personal injury, six searches filled the same 18 slots with six firms. In both cities, one firm appeared in every single pack.

Two cities, two practice areas, the same shape.

ChatGPT showed the same behavior once it drew a map. In Atlanta it attached a map to every hiring search, and across ten differently worded questions (car accident lawyer, settlement lawyers, free consultation, highly recommended) it named only 12 firms in total. The same firm came first in the written answer all ten times, and three firms took 65% of all its mentions.

Rewording did not reshuffle anything. That matters for how your firm measures itself. Tracking twenty phrasings of "best lawyer near me" mostly tells you the same thing twenty times. The useful question is simpler. Is your firm on the map roster (the short list of firms a platform keeps showing on its map) at all?

The likeliest reason, and this is a reading of the pattern rather than something the platforms disclose, is that map results are built from business listings, not from the words in the question. Change "car accident lawyer" to "settlement lawyers" and the platform still pulls the same kind of listing in the same city. The wording picks the shelf. The listings on that shelf stay put.

That shifts where the effort goes. If the shelf barely moves, the details of your firm's listing (its categories, service area, photos and reviews) do more work than any single phrase you might try to rank for.

The rosters were not shared evenly across platforms.

In Atlanta, every firm Google's AI Overview (the AI-written summary at the top of Google results) recommended on a hiring search also appeared on Claude's map, and the AI Overview's picks matched the local pack beneath it in nine of ten slots. Google's pack, Google's AI Overview and Claude's map behaved like one list seen through three windows.

ChatGPT's map was a different list. Of the eight firms it placed on its maps, one also appeared on Claude's map and two in Google's pack.

So checking four platforms is not four independent checks. Finding out which rosters your firm sits on takes a couple of searches, not a week of monitoring.

That split can make a firm look strong in one place and absent in another without anything having changed on the firm's side. The two lists appear to draw from different places, so a gap on one of them is a listing question before it is a reputation question.

One Atlanta search showed what happens when the question narrows. Asked for the best truck accident lawyer, Claude dropped its map and answered from the web. It named seven firms, and none of them had appeared on any of its maps.

I expected that to scatter the results further. It did the opposite.

Four of those seven firms also surfaced in Google's organic results (the regular listings below the map and ads) or in Perplexity's answer to its own truck question. Every firm Perplexity named for trucks appeared on at least one other platform. One firm showed up on all three platforms that ran a truck search. Meanwhile none of the firms from the local pack, and none of the map regulars from the generic searches, made a single truck list.

The likeliest reading, and it is a reading rather than a proof, is that a specific, high-stakes question pushes the platforms off the roster and onto web pages. Once they are reading pages, they read many of the same ones, and those pages belong to firms with a visible record in that exact kind of case.

This was one case type in one market, and the next markets will test it. But every practice area has its own version of the truck question. In family law it might be a custody relocation or a high-asset divorce. In criminal defense it might be a federal charge or a DUI with injury. These are the searches where the generic roster stops mattering.

A useful way to spot them is to ask which cases a client would never trust to the first name on a map. Serious injuries, large financial stakes, a charge with real exposure, anything where the person searching wants proof of experience in that exact situation. In this data, the truck question was exactly that kind of search, and it was where the answers moved off the map and onto pages.

Page-based answers spread across far more firms. Google's organic results in Atlanta named 62 firms in 107 mentions, and no firm appeared in more than half the searches. Perplexity never showed a map in either market and stayed wide both times, naming 24 firms in Atlanta and 29 across just six answers in Nashville.

Put the two sides next to each other and the gap is stark. Google's regular results named ten times as many firms as its local pack, 62 against 6. On Google's local pack in both markets, and on ChatGPT's and Claude's maps in Atlanta, the most-named firm appeared in 100% of the searches that showed a map. On Atlanta's page-based answers, the most-named firm reached 50% at best.

That is where most of the fragmentation in this research comes from. It is crowded, but nobody owns it, and classic SEO is still the way in. A page that ranks for a specific question is a page an AI platform can find, read and cite.

The open field also had room for people the maps left out. Google's organic results and Perplexity both named individual attorneys by name, not just firms. None of the maps named an individual lawyer at all. For a solo or a small firm without a deep review count, the page side may be the only side where the name gets said out loud.

Everything here comes from a single collection window in each market. AI answers move. The same search can return a different list a week later, or a different list for someone searching from across town, and the platforms update their models without announcing it.

So treat these patterns as a record of how the platforms behaved, not a promise of how they will behave in your market next quarter. The closed roster held across two cities and two practice areas, which makes it the first thing worth checking. The truck result rests on one search and deserves more caution.

That is also why a single look at your own visibility is thin evidence. One search tells you what happened once.

Start with the map. Run your most common hiring search on Google and on ChatGPT, then run two or three rewordings. If the same handful of firms keeps coming back, your market has a closed roster, and you know exactly who is on it.

Then check which family your firm belongs to. Being on Google's pack and Claude's map while missing from ChatGPT's map was the most common split in Atlanta, and it tells you which listing work is actually unfinished.

Next, write down the three most specific, high-stakes questions a client in your practice area would ask. Run them on Claude, Perplexity and Google, and look at the firms that come back. If competitors with deep pages on those exact issues keep appearing across platforms, that is the open market, and it is where page work has room to compete.

Keep score the simple way. For each platform, write down how many of your searches named your firm, out of how many you ran. Four of four on a map and one of six on the web is a very different position from the reverse, and the gap points to which kind of work comes first.

Then run the same searches again in a month. A count that holds steady looks like a roster seat. A count that swings looks like the open field, where steady page work matters more than any one result.

Markets can differ in degree. ChatGPT's shortlist in Nashville was looser than in Atlanta, naming 15 firms across six answers. The direction was the same, though, and the only way to know your own numbers is to run the searches.

The map rewards whoever is already on it. The page still rewards whoever answers the question best.

Go find out which one your firm is playing.