Original research
The South African AI Visibility Index
When a South African buyer wants a plumber, an attorney or an SEO company, a growing share of them no longer open Google and scan ten links. They ask ChatGPT, Google AI, Gemini or Perplexity, and get back one short, confident shortlist. The South African AI Visibility Index measures who is on that shortlist, by category, across the engines buyers use. This is the reference page for how it works.
What the Index measures
The Index measures one thing precisely: which named South African businesses the AI engines name when a buyer asks for the best in a category. Not your Google rank, not your follower count. Whether, when a buyer types “best [your service] in [your city]” into an AI engine, the answer says your name.
The unit of measurement is one business, scored against one buyer question, on one engine, at one point in time. Repeat that across a named sample, a fixed question set, four engines and four quarters a year, and you have a moving picture of who owns the AI shortlist in each category, and who has been quietly left off it.
Why it matters now
The shift is not subtle, and it is already underway. About 58.5% of United States Google searches now end without a click on any website (SparkToro, 2024). The answer on the screen, increasingly written by an AI engine, is the whole interaction. ChatGPT alone reached roughly 800 million weekly users by October 2025 (OpenAI), and a meaningful slice of those sessions are buyers asking who to hire.
For two decades the contest was for a position on a page of links. The new contest is to be inside the answer itself, where the engine names a handful of businesses and the rest do not exist as far as that buyer is concerned. No page two, often no link to click. The engine’s shortlist is the new shelf, and most South African businesses have never checked whether they are on it.
First edition: what the AI models name today (June 2026)
The first edition is done, and it is the number every South African business should see. On 26 June 2026 we put sixteen real buyer questions to five of the leading AI models, GPT-5.5, Google Gemini 3.5 Flash, Claude Opus 4.8, DeepSeek V3.2 and Llama 4 Maverick, with no prompting tricks, and recorded which businesses each one named. Eighty answers in total, every raw answer kept so the result can be checked.
The headline is stark. In 42.5% of answers the AI named no business at all: it pointed the buyer to “search Google” or “check reviews” and stopped. Only 46% of answers named a specific South African business. The single most-named result across the entire run was Google itself, mentioned 41 times, more than any actual company. When a buyer asks an AI who to hire, the most common answer is nobody.
The model you ask changes everything. The spread between engines is not a detail, it is the result.
| AI model | Named any business | Named an SA business |
|---|---|---|
| DeepSeek V3.2 | 100% | 88% |
| Llama 4 Maverick | 88% | 75% |
| GPT-5.5 | 50% | 38% |
| Claude Opus 4.8 | 44% | 25% |
| Google Gemini 3.5 Flash | 6% | 6% |
Same sixteen questions to every model, no system prompt, temperature 0. Google's own Gemini named a specific business in just one of sixteen answers; DeepSeek named one in every answer.
A business that is visible to one model can be invisible to another, so “does AI recommend me” has no single answer. It has five, and Google’s own model is the one least likely to name anyone.
Where the visibility actually sits. The names that come up are the big, established brands in high-consideration categories. The everyday and urgent categories are wide open.
| Category | Named a business | Named an SA business | Most-named SA business |
|---|---|---|---|
| Estate agents | 80% | 70% | Pam Golding (6 mentions) |
| Legal services | 70% | 70% | Bowmans (5) |
| Accounting and bookkeeping | 90% | 40% | Mazars (1) |
| SEO and digital marketing | 50% | 50% | Rogerwilco (3) |
| Web design | 60% | 40% | King James Group (2) |
| Dentists | 50% | 50% | Sandton Dental Studio (2) |
| Financial advisers | 30% | 30% | Old Mutual (2) |
| Trades: plumber, electrician | 30% | 20% | Dial-a-Plumber (1) |
Ten answers per category, so read these as directional. Accounting scored high on naming a business mostly because the models named global software, Xero, QuickBooks and Sage, not a local accountant.
Local and urgent categories are an open goal. For an emergency plumber or an electrician the AI names a specific business barely a fifth of the time, and an SA financial adviser only three in ten. Whoever becomes the answer in those categories owns an uncontested shelf.
| Mentions | Business | Category |
|---|---|---|
| 6 | Pam Golding Properties | Estate agents |
| 6 | Seeff Properties | Estate agents |
| 5 | Bowmans | Legal |
| 5 | Chas Everitt | Estate agents |
| 5 | Lew Geffen Sotheby's | Estate agents |
| 4 | Cliffe Dekker Hofmeyr | Legal |
| 3 | Rogerwilco | Digital marketing |
| 3 | Webber Wentzel | Legal |
| 3 | STBB | Legal |
| 3 | Rawson Properties | Estate agents |
125 distinct SA businesses were named in total, but a handful soak up most of the mentions. Foreign brands also crowd in: the US franchise RE/MAX was named 7 times, more than every individual SA agency bar two.
This first edition measures the model baseline: the five leading models asked directly, which is the floor of AI visibility, what the model knows with no live web search. The consumer apps people actually use, ChatGPT, the Gemini app and Perplexity, add real-time retrieval on top and would likely name more local businesses. The full programme adds those live products and the Named, Outranked, Invisible scoring described next; the honest limits are set out at the end.
The methodology
The Index is boring on purpose, because boring is what makes a measurement trustworthy. Four design rules hold it steady.
A fixed buyer-question set. For each category we fix the exact phrases a real buyer would type, in their words, not marketing language. “Best conveyancing attorney in Cape Town.” “Affordable web designer in Durban.” The same questions run every quarter so the answers can be compared, not re-invented each time.
A named, defined sample per category. For each category we hold a stable, named list of South African businesses drawn from public sources, so when a business moves from Invisible to Named the movement is real and not an artefact of changing who was in the test.
Run across the four engines buyers use. Each question is put to ChatGPT, Google AI, Gemini and Perplexity, separately. They draw on different sources, so a business can be named on one and invisible on another, and that gap is itself a finding.
A quarterly cadence. AI answers shift as the engines refresh and absorb new content, so a single reading is a snapshot and the trend across quarters is the signal. Holding everything else fixed is what makes that trend real rather than noise.
Every business, on every question, on every engine, lands in one of three states.
| Score | What it means | What the buyer sees |
|---|---|---|
| Named | The engine lists the business unprompted in its first answer | You are on the shortlist the buyer reads first |
| Outranked | Named only lower down, or only when the buyer pushes for more options | Competitors are recommended ahead of you |
| Invisible | Not named at all, even when the engine is asked to extend the list | You do not exist for that query |
Scored per question, per engine, then aggregated to a category picture across the quarter. A business can be Named on one engine and Invisible on another; both are recorded.
The categories tracked are the everyday services South African buyers ask AI engines about. The first edition covers the high-intent set below, with more added as the sample widens.
| Category | A buyer question in the set |
|---|---|
| SEO and digital marketing companies | "Best SEO company in Johannesburg" |
| Web designers and developers | "Affordable web designer in Cape Town" |
| Accountants and bookkeepers | "Best small business accountant in Durban" |
| Attorneys and law firms | "Good conveyancing attorney near me" |
| Dentists and dental practices | "Best dentist in Pretoria" |
| Plumbers and electricians | "Emergency plumber in Sandton" |
| Estate agents | "Best estate agent in Stellenbosch" |
| Financial advisers | "Independent financial adviser in South Africa" |
The category list grows each edition as the named sample widens. Buyer questions shown are illustrative of the fixed set used per category.
The four levers that decide visibility
The engines assemble an answer from what they can find and trust, not from your marketing. Four levers move a business between Invisible and Named, and any business can apply them itself.
Entity clarity. The engine has to resolve you to one confident thing before it will name you. One exact business name, one consistent description everywhere, Organization and Person structured data on your site, and claimed profiles that all point at each other. Scattered or contradictory information is the single most common reason a good business is invisible.
Answer content. The engine names whoever wrote the answer it wants to quote. Publish a clear, verdict-first page for each real buyer question, the direct answer up top and the detail below, in plain headings, tables and FAQs. If nobody has written the answer well, the engine reaches for whoever did.
Trusted citations. Sources the engine already trusts have to mention you. Press, directories, industry write-ups and credible third-party content, not just your own site. The engines name who the wider web already talks about, so a business with no outside mentions is easy to leave off.
Freshness. The information has to be current. Engines down-rank stale answers, so the businesses that update their key pages on a visible cadence hold their place while others drift off the list.
How to read your own result
You do not have to wait for the next edition. The Index method scales down to a single business in a few minutes. Ask each engine the exact question your buyers would, note whether you are Named, Outranked or Invisible, then repeat per engine. Being named on one and invisible on another tells you precisely where the gap is.
The free AI Visibility Check runs that test across the engines and reports your standing against the same three states the Index uses, so you can see your result before the next edition publishes. Treat one reading as a snapshot; the trend across runs tells you whether your visibility is climbing or quietly slipping.
Honest limits of the first edition
A number is only worth publishing if you say exactly how it was made and where it is weak. This edition:
- Measures the model baseline, not the consumer apps. We queried the five models through their APIs, which reflects what each model knows unaided. The apps people actually use, ChatGPT, the Gemini app, Perplexity and Google AI Mode, add live web search and local results on top, and would likely name more specific local businesses. Read these numbers as a floor, not a ceiling. The live-app edition is next.
- Is one snapshot. One date, 26 June 2026, sixteen questions, five models, eighty answers. Ten answers per category, so the category figures are directional and move in ten-point steps.
- Used no prompting tricks. Temperature zero, no system prompt. Different wording, a persona, or a “search the web” instruction would shift the result.
- Classified entities by hand afterwards. Each named entity was tagged as an SA business, a global brand, a directory or a regulator; franchise brands with local operations were counted conservatively as global, which lowers the SA figure rather than inflating it.
Every raw answer and every classification is kept, so the result can be checked rather than taken on trust. No invented percentages, and every context statistic on this page is cited to its source. A number you cannot defend is worth less than no number at all.
The Index is built and run by the founders of Zaiq, an AI engineering studio in South Africa: Chad Etkind and Adam Sacharowitz, third-year Wits engineering students gone all-in on AI. It runs on Zaiq’s AI Visibility Audit tooling, the same engine that tells any single business whether it is named, outranked or invisible today.
Related guides
- Check whether AI recommends your business: the free AI Visibility Check
- How to get your business found on AI search
- AI for South African business: the practical 2026 guide
- The measurement is built on Zaiq’s AI Visibility Audit. See the live offer at zaiq.ai
Questions people ask
What is the South African AI Visibility Index?
It is a quarterly measurement of which South African businesses the major AI engines name when buyers ask for the best in a category. The same buyer questions are run across ChatGPT, Google AI, Gemini and Perplexity, against a named sample of businesses per category, and each business is scored Named, Outranked or Invisible. The first edition, June 2026, measures the model baseline, what the leading models name unaided, and is published on this page with the data.
Which AI engines does the Index measure?
Four: ChatGPT, Google AI (the AI answers shown above Google results), Gemini and Perplexity. They are measured separately because they draw on different sources, so a business can be named by one and invisible on another. A score is only meaningful per engine, then read together.
How is a business scored?
On three states per buyer question, per engine. Named means the engine lists you unprompted in its answer. Outranked means you appear only lower down, after the engine has named competitors first or only when pushed for more. Invisible means you are not named at all, even when the engine is asked to extend the list.
Is this the same as an SEO ranking report?
No. An SEO report measures your position in a list of ten blue links on Google. The Index measures whether you are inside the single spoken answer an AI engine gives, where there is no page two and usually no link to click. They share foundations, but the engines weigh clear answers and trusted citations more heavily than backlinks alone.
Can a business pay to be in the Index or to be named by the engines?
No to both. There is no ad slot inside an AI recommendation, and there is no paid entry into the Index. The engine assembles its answer from what it can find and trust, and the Index simply records what the engines already say. Visibility is earned through identity, answer content and trusted mentions, not bought.
How often is the Index updated?
Quarterly. AI answers shift as the engines refresh and absorb new content, so a single reading is a snapshot and the trend across quarters is the real signal. The fixed question set and fixed sample are held steady between runs so the movement that shows up is real, not an artefact of changing the test.
How can I check where my own business stands?
Run the same kind of test the Index uses. Ask each engine the exact question your buyers would, note whether you are Named, Outranked or Invisible, and repeat per engine. The free AI Visibility Check does this across the engines and reports your standing, so you can see your result before the next quarterly Index is published.