Mentions, citations and recommendations
An AI answer can involve a law firm in three different ways, and the three are often reported as if they were one. Joe Ashta, a GEO / AEO consultant for law firms, keeps them on separate lines. This page takes one answer apart to show a citation, a mention and a recommendation side by side, explains why they are never summed, and states which of them a buyer acts on.
One answer, three outcomes
A buyer asks an AI assistant: "Who should I hire for a DUI in Vancouver?" The answer, in outline, runs like this.
It opens by explaining that a first impaired driving offence in British Columbia is usually handled under the immediate roadside prohibition scheme, that the review window is short, and that a lawyer should be contacted within days. Under that paragraph sits a source link to an explainer page on the website of Firm A.
It then notes that several Vancouver firms handle these matters, and that Firm B, among others, publishes guidance on the review process. Firm B's name appears in the sentence. There is no link and no advice to call it.
It closes by saying that for a driving prohibition review in Vancouver the buyer should consider contacting Firm C, which focuses on impaired driving defence and lists recent review outcomes on its site. Firm C is named as the firm to call.
| Firm | What happened | Outcome |
|---|---|---|
| Firm A | Its page was linked as a source for the background paragraph | Citation |
| Firm B | Its name appeared in the answer text, in passing | Mention |
| Firm C | It was named as the firm the buyer should contact | Recommendation |
Firm A did the most writing. Firm C gets the phone call. Firm B is in between, and its position is the hardest to interpret, because a mention proves the system knows the firm exists and does not prove it prefers the firm.
Why the three are counted separately
They are different events, produced by different evidence, and they carry different weight with the buyer.
- A citation is a property of a page. It follows from the page being specific, quotable and attributable. The AI citations page covers what earns one.
- A mention is a property of an entity. It follows from the system being confident about who the firm is, which depends on consistent naming and records that agree.
- A recommendation is a property of an entity's standing against its competitors for one question. It follows from the system preferring that firm over the others it knows about, and that preference is built from corroboration by sources the firm does not control.
Adding them together produces a number with no meaning. Two citations plus one mention plus one recommendation is not four of anything. Worse, the sum hides the outcome that matters: a firm with forty citations and no recommendations looks better on a combined score than a firm with two recommendations and no citations, and the second firm is the one being hired. For the same reason a citation count is never compared against a mention rate. One is a count of links across a sample of answers a vendor chose. The other is a share of answers in which a firm was named. They have different units.
Which one a buyer acts on
The buyer acts on the recommendation. In the example, the buyer reads the background, sees a firm named as the one to contact, and contacts it. Firm A's source link may be clicked by a careful buyer who wants to read the explainer; that click is a visit, and it is the same kind of visit a search result produces. Firm B's name may register. Firm C's name is the instruction.
This is why AEO targets mentions and recommendations, and why the Brand in AI answers page describes the goal as a firm becoming the default answer. A firm that only counts citations is measuring how often its writing is borrowed. A firm that counts recommendations is measuring how often it is chosen.
How each is reported
Citations are vendor-reported. A data vendor, here Ahrefs, crawls AI answers at scale and counts source links per site. The BigLaw GEO Leaderboard is built from that count and is labelled as a citation instrument.
Mentions and recommendations are measured. They come from observing the platforms' own answers to the questions a firm's prospective clients ask and recording whether the firm was named, and whether it was named as the one to hire. Both are reported as a mention rate: the share of answers in which the firm appears, with recommendations tracked as their own share within that. The method behind measured work is not published on this site. The entity record states that boundary, and every number on the site carries the label that says which of the two kinds it is.
A firm reading its own visibility report should expect to see three lines, each labelled, each with its own unit, and no total.
Related
AI citations goes deeper on the first outcome. How AI chooses which law firms to name explains what moves a firm from mentioned to recommended. AEO vs SEO sets the three outcomes against the click, the unit search work has always counted.