Research question: of the on-site fixes the AI-visibility industry sells, which ones correlate with a firm actually being named by AI?
The short version lives on the study hub: speed tracked visibility, most of the checklist did not. This page is the long version. How we measured it, the full ranking of every signal we tested, the firms behind the numbers, and the one firm that did everything the industry recommends and stayed invisible anyway.
For every one of the 30 firms we computed a single visibility rate: the share of its own AI queries in which it was named. Because each firm was tested on queries drawn from its own detected practice areas, this asks the fair question, does this firm show up for the cases it actually handles, rather than penalizing a dog-bite firm for missing trucking answers.
We then took each on-site signal our audit records and asked whether it moves with that visibility rate, using a rank correlation (Spearman’s rho). A positive number means firms with more of the signal tended to be more visible. A number near zero means the signal did not separate the visible firms from the invisible ones. Because the sample is 30 firms, we treat only the strongest relationships as significant and everything else as directional.
Why rank correlation, and why the caution. Rank correlation is robust to outliers and does not assume a straight-line relationship, which suits a small, messy real-world sample. At n=30, a correlation needs to reach roughly ±0.36 to be statistically significant at the usual threshold. Two signals cleared that bar. The rest are reported honestly as leans, not proof.
The hub shows the headline eight. Here is the complete set, including the signals that landed dead in the middle. The pattern is stark: the two bars that reach furthest are both speed, pointing in opposite directions. Nearly everything the industry markets as an AI-visibility fix clusters at the center line.
A correlation coefficient is abstract, so here are the actual firms. Sorted by visibility, with the signals the industry sells most alongside the one that mattered. Watch the speed column separate the two groups, and watch the citability and statute columns fail to.
| Firm | Visibility | Speed | Load (LCP) | Citability | Statutes |
|---|---|---|---|---|---|
| Most visible | |||||
| Bachus & Schanker | 38% | 77 | 4.4s | 0 | 10 |
| Ramos Law | 38% | 52 | 4.6s | 52 | 17 |
| Parker Lipman | 33% | 93 | 1.8s | 27 | 1 |
| Denver Trial Lawyers | 27% | 75 | 1.8s | 18 | 4 |
| Burg Simpson | 23% | 96 | 2.3s | 12 | 0 |
| Least visible | |||||
| Whalen Injury Lawyers | 4% | 45 | 6.6s | 14 | 0 |
| Jordan Law | 4% | 34 | 19.7s | 42 | 52 |
| Amy G Injury Firm | 2% | 35 | 14.2s | 18 | 8 |
| Kanner & Pintaluga | 0% | 55 | 20.9s | 22 | 0 |
| Daniel R. Rosen | 0% | 33 | 13.4s | 38 | 1 |
By the AEO checklist, Jordan Law is a model firm. It carries the highest statute-citation count of all 30 firms and one of the highest citability scores. If citations and content depth drove AI visibility, it would be everywhere. Instead its pages take almost 20 seconds to load in Google’s mobile test, and it is among the least-named firms in the study. The content was there. The speed was not, and the visibility followed the speed.
A signal landing near zero does not automatically mean it is worthless. There are three different reasons a signal can look flat, and they carry different lessons.
It may genuinely not drive visibility. The “content citability” score is the clearest case. It varies widely across firms, from 0 to 52, so the sample had plenty of range to work with, and it still showed no relationship with who AI named. That is a real null, and it is why we treat citability as a diagnostic aid rather than a growth lever.
It may be a floor everyone already clears. Some signals could not be tested at all, because almost every firm already has them. That is a finding of its own, and it is the subject of the technical-readiness study. A signal with no variation cannot correlate with anything, even if it is genuinely required to be eligible.
It may be necessary but not sufficient. Schema markup may help a firm become readable and eligible without, by itself, pushing that firm up the list. Our data cannot separate “useless” from “necessary but not enough.” What it can say plainly is this: on a real market, adding schema did not, on its own, correlate with being named. Speed did.
It is the one on-site signal that tracked visibility, and it is measurable, fixable, and cheap relative to a rebuild. Target a mobile Largest Contentful Paint under 2.5 seconds. Most of the invisible firms in this study were sitting between 13 and 21 seconds.
The negative lean on script load is the same story as speed. Every analytics tag, chat widget, and pixel has a cost. Justify each one against the performance budget rather than stacking them by habit.
Have the basics in place because they make a site readable and eligible. Do not pay a premium for schema as your AI-visibility plan, and be skeptical of anyone who sells it that way. This data does not support the claim.
Statute presence leaned mildly positive; raw statute count did not. Jordan Law’s 52 citations bought it nothing. The value is in being a genuinely useful, specific answer to a real question, not in stacking code sections on a slow page.
Every Denver firm in the study has its own numbers: load time, the signals it has and does not, and the exact queries where AI named someone else. We are happy to walk any firm through its own results.
See your firm’s results