A market is a configuration object, not a locale string.
Most trackers localise by changing a two-letter code, or worse, by appending “in South Africa” to the prompt. A market is actually a dozen decisions — which engines matter here, what geography to pass, which languages people ask in, which surfaces even exist yet, and what a local-intent query is supposed to return. We made all of them explicit.
What a market carries.
Brands attach to a market and inherit all of it. Nobody configures geography per brand, which is why adding the eighth client takes minutes rather than a checklist.
| Field | South Africa | What goes wrong without it |
|---|---|---|
| Engine mix | Six engines, weighted to local usage | A flat average across engines nobody here uses, presented as market visibility |
| Geo parameters | Country ZA, city, timezone, hl=en | An unlocated answer, or a location injected into the prompt text and hoped for |
| Languages | English, Afrikaans; isiZulu following | A third of the variance in answers goes unmeasured |
| AI Overview profile | Tri-state presence; availability tracked | Vendor detection misses recorded as brand absences, permanently |
| Device | Desktop and mobile measured separately | Two surfaces that disagree, averaged into one that describes neither |
| Local-intent rules | Local intent expects a map pack, not an AI Overview | A map-pack query scored as an AI miss, dragging a trendline down for no reason |
| Cadence defaults | Weekly; daily as a lever | Cost that scales seven-fold with no measurable gain in a slow-moving metric |
Why this market, first.
Not sentiment. The conditions here are unusually good and unusually unmeasured at the same time, which is the rarer combination.
- Google’s share in Africa is around 96% — so the Google AI surfaces are not one channel among many, they are effectively the channel.
- Roughly 70% of South African adults have used an AI chatbot, and the country leads the continent on AI diffusion.
- AI Mode has been live here since August 2025, and four South African languages joined it and AI Overviews in March 2026.
- The agency market already sells GEO retainers — the buyer exists and is already having the conversation.
- And there is no credible published South African measurement of any of it. No local AI Overview trigger rates, no engine-mix data, nothing.
That last absence is not a gap in our research. It is the reason we publish a benchmark.
The SA AI Search BenchmarkLanguages, measured separately
Illustrative of the pattern we see: same intent, same engine, same week, materially different answers. Language explains roughly 32% of answer variance — more than the model, more than the brand.
Language cannot be bolted on later. It has to be a dimension of the prompt from the moment the prompt is stored, or every historical figure is an unlabelled mixture. A tool that adds language support afterwards can only tell you about the future.
English and Afrikaans day one. isiZulu next.
| Language | Status | Live in Google AI surfaces | Notes |
|---|---|---|---|
| English | Launch | Yes | The default phrasing for most commercial prompt sets |
| Afrikaans | Launch | Yes — since March 2026 | First-class prompt language, not a translation applied at report time |
| isiZulu | Fast-follow | Yes — since March 2026 | Shipping once prompt authoring in it is genuinely good rather than merely possible |
| Sesotho, Setswana | Under evaluation | Yes — since March 2026 | Surface support exists; the question is whether commercial prompt sets in them do |
We are not going to claim eleven languages because the country has eleven official ones. A language is supported when someone can author a real commercial prompt set in it and the results are worth putting in front of a client — not when a locale code accepts it.
Adding a country is configuration plus proof.
This is the whole point of the abstraction, and it is also where most platforms quietly cheat — a new market that is really the old market with a different flag on it.
- Define the market: engine mix from local usage, geo parameters, languages, device treatment.
- **Validate that every adapter resolves for it.** Not assume — run located and unlocated control prompts and compare.
- Establish AI Overview availability and the local trigger profile by intent.
- Write the local-intent rules, so map-pack queries are not scored as AI misses.
- Seed a benchmark. If we cannot publish credible measurement of a market, we do not understand it well enough to sell it.
The second market is a customer-led decision rather than a speculative build. If your agency operates across a border, that conversation is a short one.
Candidate markets, and why
| Market | Why it is interesting |
|---|---|
| Nigeria | Scale, English-first, almost entirely unmeasured |
| Kenya | Mature digital agency sector; strong mobile-first search behaviour |
| Namibia, Botswana | Shared language and engine profile with ZA; small config delta |
| UAE | Multilingual by default — the hardest and most instructive test of the language model |
None of these are built. They are listed so you can tell us which one matters to you.
Tell us which market you actually need
South Africa is built and measured. If your client list crosses a border, the second market is a configuration exercise — and we would rather scope it with you than guess.