Being named isn't being recommended. AI knows the difference.

Ask an AI assistant who the best law firm in South Africa is and a shortlist comes back — three or four names, stated as fact. Being on that list is the whole game, and most brands never find out whether they made it. But there is a second question hiding inside the first, and it decides who actually wins the buyer: once you are named, are you recommended — or just listed alongside the ones that are?

Those are not the same thing, and a mention count cannot tell them apart. We built a reading that can. Across the 1,296 grounded answers in Edition 1 of the SA AI Visibility Index, we recorded — for every brand in every answer — whether it was merely named in passing or actively presented as a recommended choice. Aggregated, that gives conviction: of the answers that name you, how many recommend you. It is the difference between being in the room and being the one pointed at.

AIV Index research — the SA AI Visibility Index, Edition 1 — 1,296 grounded answers across South African property, law and accounting, measured June–July 2026 and analysed at the answer level for both mention and recommendation.

Share-of-mention asks if you're named. Conviction asks if you're chosen.

Share-of-mention is the headline number: of all the answers to a category's questions, how often is your brand named against the field. It is the right first question, and for most brands it is a sobering one — plenty of famous names sit near zero. But it has a ceiling. It counts the door you got through; it does not count what happened once you were inside.

Conviction is the second question. Two firms can be named in roughly the same share of answers and be in completely different positions — one recommended nearly every time it comes up, the other listed as an also-ran, a name the assistant mentions on the way to endorsing someone else. To the buyer reading the answer, that gap is everything. To a mention count, it is invisible.

Being named puts you in the answer. Being recommended is what the buyer acts on.

In a close market, conviction is what separates the field

Property is the clearest case of the two questions agreeing. Pam Golding Properties is named in 45% of all property answers — a monarchy, a wide margin over any rival — and when it is named, it is presented as a recommended choice 91% of the time. Volume and conviction point the same way: it owns the category on both readings. That is what total dominance looks like on the instrument.

Law is where the two questions come apart. No firm dominates the volume race — Bowmans leads at just 15%, with five more firms clustered within three points behind it. When mention counts are that close, conviction does the separating. Webber Wentzel is recommended in 96% of the answers that name it — the highest conviction we measured anywhere in Edition 1. It doesn't win the head-count; it wins the answers it's in. A firm reading only its share-of-mention would see a tight pack and miss that one of them is being endorsed far more decisively than the rest.

AIV Index research — Edition 1 property and legal league tables (432 grounded answers each), scored for both mention rate and recommendation rate per named brand.

A brand named-but-not-recommended is losing in a way it can't see

Here is the trap. A firm checks its AI visibility, sees itself named in a healthy share of answers, and concludes it is doing fine. It is not looking at the reading that matters. If it is named often but recommended rarely, the assistants are using it as furniture — a name that lends the answer credibility on the way to endorsing a competitor. The buyer's eye slides past it to the one the machine actually stands behind.

That is a fixable position, but only if you can see it. The lever is different from the one that moves share-of-mention. Getting named is a corroboration problem — being present on the directories, reviews and guides the engines read. Getting recommended is a conviction problem — being the brand those same sources describe in the language of a clear choice, not a hedge. You cannot work on the second until you have measured it apart from the first.

If the engines name you but recommend someone else, your visibility number is lying to you.

And like everything on the instrument, it is four races, not one. A brand can be recommended decisively on Claude and merely listed on ChatGPT. Conviction is scored per engine for exactly that reason — the answer is never one number.

What to actually do about it

1. Measure recommendation, not just mention. Your share-of-mention is the first rung, not the whole ladder. Ask, per engine: of the answers that name us, how many recommend us?
2. Read the gap against your rivals. A competitor with lower volume but higher conviction is beating you where it counts. That's the firm to study.
3. Fix the floor first. If you're barely named, conviction is premature — get corroborated on the sources the engines read before you worry about the language they use. Check the engines can even read you; it's free.
4. Then work the language. The pages that earn a recommendation answer a buyer's question with a clear stance, not a brochure hedge.

Conviction is a Pro-plan reading on AIV Index, scored per engine on your own brand and your tracked competitors. Where your firm sits — named-and-recommended, or named-and-passed-over — is specific to you, and you can measure it directly. The published SA AI Visibility Index shows the same reading run across whole industries.

A sibling piece goes wider on the same edition: every industry has a different AI shortlist.

Sources: AIV Index research — the SA AI Visibility Index, Edition 1 (1,296 grounded answers across South African property, law and accounting; read it here). AI answers shift over time; this is a point-in-time read of patterns we measured, not a guarantee.