When a good firm cannot understand why it is invisible online, the cause is almost never what the partners suspect. It is not the content budget, the brand, or the agency's effort. Across more than 1,500 audited personal injury websites, invisibility decomposes into four structural mistakes, all boring, all measurable, all fixable. Here they are, with the data and the mechanisms.
Read a typical PI homepage the way a machine reads it and you cannot cleanly tell that this is a law firm, in these specific cities, handling these specific case types. A human squints and infers it from the logo and the gavel photos. Machines do not squint. They move on to a site that made it explicit.
The scale of this one still surprises me: in my 1,005-firm audit of page-one sites, only 35.3 percent declared themselves a legal service in machine-readable form. Two thirds of winners leave their own business identity to guesswork, and in the companion profile audit, 43 percent compounded it by sitting under the wrong Google Business category.
The mechanism is corroboration. Google patented cross-checking facts about an entity across sources, US 8682913, over a decade ago, and the AI answer layer inherited the habit. A firm whose site, profile, and directories all agree is a confident entity. A firm whose own website will not testify is a guess, and machines do not cite guesses.
One thin page per practice area. A menu of case types, each behind a paragraph and a call button. It looks like coverage on a sitemap and reads as emptiness to both audiences that matter.
The person searching has twenty real questions. The firm that answers all twenty earns the trust and the call, even when it is the smaller firm. The machine matching concepts, per Google's semantic patents, US 8577907 among them, maps your pages against the whole topic and finds the gaps immediately.
The median firm in my audit sits at Level 2 of 5 on structural maturity; only 1.3 percent reach Level 4. Thinness is not a laggard problem. It is the market's default state, top to bottom, which is why one genuinely complete case type is still a differentiator in every metro I have measured.
Median mobile render time across the 1,005 page-one firms: 5.5 seconds.
Hold that against the actual user: in pain, on cellular data, in a parking lot, typing with one hand. They are gone before the median winner's hero image paints. In a vertical where the click that just bounced can cost $100 to $300 at auction, the median firm is running one of the most expensive latency problems in commercial existence, invisibly, all day.
And the cost is not only the bounce. Google patented predicting site quality from the site itself, before any user data exists, US 9767157. Heavy and slow sets a low floor before a single visitor arrives, and can suppress how much of the site gets indexed at all. Speed is not polish. It is admission.
The name is spelled two ways. The old office address survives on a directory. The site says one set of practice areas, the profile another. Each contradiction is small; together they tell every corroborating machine not to be confident about who you are.
Confidence is the currency of both surfaces that now matter: the map pack and the AI answer. In my SERP Authority Report, 663 of 1,005 firms were structurally absent and only 11.1 percent of AI citations on PI queries were on-topic. The vacuum is real, and self-contradiction is how firms volunteer to stay out of it.
A note on how mistake four happens to careful firms: records drift. Profiles accept public edit suggestions. Directories import stale data from each other. An office move updates the website and nothing else. Consistency is not a project you finish; it is a record you keep, and the firms that stay clean simply assigned someone to keep it, checking quarterly what every major surface says about the firm. Unowned records decay toward contradiction on their own schedule.
None of these mistakes survives contact with attention. They persist because they live in the gaps between vendors: the profile belongs to someone gone, the site to a designer, the content to a retainer, and no dashboard anywhere flags that the pieces disagree. Every failing firm in my dataset had a dashboard that looked fine.