Publishing a practice-area page does not, on its own, get your firm named in an AI answer. In our own testing across fourteen near-identical law-firm pages, the ones an assistant actually cited shared two things the rest lacked — and neither of them was the page itself. This is a directional read from a single engine, not a guarantee, but the pattern is clean enough to change how a firm should spend its content budget.
Most GEO advice still treats content like a coin-operated machine: publish a page for a practice area, and the AI will start naming you for it. The engines don’t work that way, and it’s worth understanding why before you fund another round of pages that won’t move anything.
Does publishing a page get your firm cited by AI?
No. A page has to exist for an engine to quote it, so existence is necessary — but it is table stakes, not the differentiator. When an assistant answers “who’s a good [practice] lawyer in [metro],” it retrieves a set of candidate sources and assembles a short, named answer from the ones it can quote and corroborate. The question is which pages clear that bar, and page existence isn’t it.
We were able to isolate this because we ran the test with almost everything held constant. Across fourteen practice-area pages on one site — same template, same schema, the same depth band, the same internal linking — most were not cited at all. When the thing you’re testing is held identical and the result still varies, that thing can’t be the cause. The page is the control.
What we held constant — so we know it isn’t the cause
On every one of the fourteen pages, these were effectively identical:
- The page itself — all live, indexed, and crawlable. If crawl access were the problem, none would be cited; some were.
- Schema and entity markup — the same structured-data stack on every page. Not a differentiator here.
- Depth — all within a tight word-count band. The page cited most was barely longer than pages that were never cited. It isn’t a word-count story.
- Internal linking — every page linked from the same practice-area hub, navigation, and footer.
Hold all of that constant and the citation results still split sharply. So the cause has to be one of the few things that varied.
The two variables that actually tracked citation
Two things separated the pages an assistant named from the ones it ignored.
1. A dated, verifiable result in that specific lane. The single page cited as the named answer for its practice area was the only one with a concrete, dated client result attached to it — and when the engine explained its answer, it quoted that result back as the reason. Not a claim of being “the best”; a specific, checkable fact the model could repeat.
2. Topical cluster depth. The cited lane was surrounded by a cluster of supporting pages on the same topic; the ignored lanes had one page and little else. Depth alone nudged a couple of lanes into a weak fallback tier — cited only through a generic, site-wide FAQ, never as the named answer — but depth without a dated result never earned the top spot.
Read the two together and a ladder appears.
The selection ladder
Across the set, pages sorted into four tiers by what was attached to them — not by how good the page was in isolation:
The jump that matters — from “cited through a generic fallback” to “named on its own page” — was gated by the dated result, not by more words or more markup.
Why this kills the “just publish more pages” shortcut
The uncomfortable, useful conclusion: there is no publish-a-page-and-get-named play. You cannot content-farm your way into AI citations, because the signal that separated the cited pages wasn’t content volume — it was a real, dated, checkable result in the lane, the kind you can only earn by actually doing the work for a client in it, or by equivalent dated, verifiable corroboration. Content is necessary scaffolding; it is not the thing that gets quoted.
This is directional, not gospel — it’s one engine, a small sample, and AI answers shift by engine, phrasing, and day. But it lines up with how retrieval works, which we walk through in how AI engines decide which lawyers to recommend. The mechanism isn’t mysterious: models quote specific, corroborated facts, and “we published a page about it” isn’t one.
What this means for how your firm spends
Three practical implications follow.
- Don’t fund content-only expansion into a new practice area expecting AI visibility to move. On its own, it won’t — a page in a lane with no dated result behind it tends to sit indexed and unquoted.
- Win a lane by earning something quotable in it first, then build the cluster around it; the dated result and the supporting content reinforce each other. Choose the niche where you can realistically earn that.
- Measure per engine, not once. Because the reads vary, a single screenshot proves nothing; the honest baseline is each major engine, asked several times, before and after.
None of this is a promise that a page, or a result, will get any particular firm named — no honest agency can promise that, and AI answers move on their own. What the testing shows is narrower and more useful: where the visibility actually comes from, so a firm can stop paying for the pages that don’t move it.
If you want to see which of your practice-area lanes are cited today and which are invisible, that is what the free AI Visibility Audit measures across the major engines. For the fuller picture of how citation works and what it costs, start with GEO for law firms.
Written by Shawn Lai, founder of Legal Torch. Legal Torch is a generative-engine-optimization agency for law firms, not a law firm; what it measures and works to improve is a firm’s visibility in AI answers, never a guaranteed ranking, lead, or outcome.
