AI Search in South Africa: How ChatGPT, Claude, Gemini and Perplexity Cite SA Brands
A South African shopper, CFO or journalist now asks an AI engine a question and gets an answer instead of a list of blue links. The engine names three or four brands. Those brands get evaluated. The rest are not in the room. Discovery Health is in the answer or it is not. Capitec is in the answer or it is not. The shortlist is short, and the composition of it has very little to do with who paid for which Google ad.
This is the category guide to AI search in South Africa. It explains what the four engines actually do, which SA sources they read, where they disagree, and what an SA brand can do about any of it. It is written by the team behind SA-AEO-Bench — the pre-registered, quarterly measurement of how AI cites 100 SA brands across 12 industries. Everything below is anchored to that data.
SA-AEO-Bench — pre-registered on OSF before data collection. 19,020 successful AI responses across GPT-5, Claude Sonnet 4.5 and Gemini 2.5 Pro, with Perplexity in the newer cycles. 232,254 cited URLs classified against a published source taxonomy. 100 South African brands, 12 industries, ~US$1,805 per quarterly cycle — transparent on cost because the methodology is open. Pre-registered at osf.io/w4az2. The category-wide read is in the Index Report; the brand-specific read is in the free Scorecard.
What AI search actually is.
Google gives links. AI gives answers.
For two decades, search meant a list. You typed a question, Google returned ten blue links, and you decided which to click. The work of choosing was yours. The work of writing the answer belonged to the brand that wrote the page you clicked.
AI search collapses that. You ask ChatGPT, Claude, Gemini or Perplexity for the best medical aid in South Africa and it composes a paragraph. The paragraph names three or four brands. It cites three or four sources. The shopper reads the paragraph and forms a shortlist. The shortlist is short. If your brand is not in it, you do not get evaluated. There is no second page.
That is the shift. Not "AI is the new SEO". Not "AI changes the funnel". The funnel still exists. What changes is who composes the consideration set, and on what basis. In South Africa, the answer is: four engines, four sets of SA sources, four shortlists. For the working vocabulary, we keep a glossary. For the underlying measurement, see our research.
Why South Africa is different.
Most of what gets written about AI search assumes a US market and a US source base. The engines were trained on a lot of CNBC and the New York Times. They were not trained on an equivalent depth of BusinessTech and Moneyweb. That mismatch shows up everywhere in SA answers. International sources leak in where SA sources should anchor. Hellopeter is read by Gemini and not by ChatGPT. The SA-source share of the citation pool runs from over 70% in short-term insurance to about 3% in safari lodges, depending on category. The spread is wide. The implications are category-specific.
SA also reads its own information differently. Hellopeter is the SA pre-purchase reflex — the place a shopper checks before they switch banks, sign up for fibre, book a car service. MyBroadband forums settle the spec arguments. Daily Maverick and News24 shape the political read. The engines that pick up these signals answer SA questions differently from the engines that do not. Gemini does. ChatGPT and Claude largely do not. That single asymmetry shapes more of the SA answer set than any other variable we measure.
And SA brands have less press behind them than their international peers. Discovery has a story. Capitec has a story. Most SA brands carry a tenth of the editorial tail that a US peer of comparable size carries. That makes the SA sites that do cover them disproportionately important. Three to five sites carry most of the citation weight in every category we measure. Five sites is a smaller fight than it sounds — and a more winnable one.
Who is using AI search in SA right now.
The CFO uses Claude. The graduate uses ChatGPT. The journalist uses Perplexity. The everyone-else, doing the everyday Google search, sees Gemini in Google AI Overviews without knowing they are using an AI engine at all. The user base is not "tech early adopters". The user base is the South African consumer, fronted by an answer engine they did not choose.
The questions that have moved fastest to AI are the high-intent ones. Which bank should I switch to. Which medical aid covers chronic. Which SUV holds value. Where to eat in Stellenbosch. Where to stay in the Kruger. The everyday Google search is still alive. The high-intent purchase question has already largely moved. That is the bit that matters for brand consideration, and that is the bit we measure.
Where you sit on the four engines, for the questions that matter to your category, is what the free Scorecard tells you. The category-level pattern is in the Index Report.
The four engines and how they treat SA brands.
ChatGPT, Claude, Gemini and Perplexity are not interchangeable. They read different sources, weight different signals and answer the same SA question with materially different shortlists. A brand needs a posture in all four. Below is what each engine actually does when the question is South African.
Plays it safe. Defaults to the names with the longest press tail.
ChatGPT is the most widely-used standalone AI assistant in SA, and the most conservative of the four engines on SA brand questions. It defaults to the names with the longest editorial history behind them — Discovery, Standard Bank, FNB, Vodacom, MTN, Toyota, Woolworths, Shoprite, Pam Golding. It hedges on newer entrants. It reads BusinessTech, Moneyweb, the SA category specialists and a smattering of international press. It largely does not read Hellopeter, Reddit or MyBroadband forums.
Best engine if you are a heritage SA brand with decades of press behind you. Worst engine if you are a newer entrant or a digital challenger that built its reputation on forums and reviews. The per-engine view of your brand is in the Scorecard.
Reads the journalism. Hedges on the hype. The analyst engine.
Claude is the engine of choice for the SA analyst class — the CFO, the consultant, the journalist. It leans more on editorial than on listings or forums, and it is the engine most willing to flag a complication in the same answer it makes a recommendation. Moneyweb, Daily Maverick, Daily Investor and TechCentral show up heavily in Claude answers. So does the long-form national press.
Best engine for brand-positioning and reputation questions. Weakest for "where do I actually buy this" intent. Claude is the engine where a clean Wikipedia entry and a consistent press tail pay back hardest. The brand-specific Claude posture is in the Scorecard.
Reads the forums. Reads Hellopeter. Powers Google AI Overviews.
Gemini is the single most consequential engine for SA brands, because it powers Google AI Overviews — the AI-generated answer most South Africans see when they search Google. Gemini reads Reddit, MyBroadband forums and Hellopeter heavily. It weights freshness more than ChatGPT or Claude. It elevates digital challengers — TymeBank, Capitec, Discovery Bank, Naked, the Chinese-import car wave — more readily than the other engines.
Most dangerous engine if you have active complaint volume on Hellopeter. Most rewarding engine if you have built credibility through community signal. The Gemini posture is the one most SA marketing teams under-monitor — see the Index Report for the full picture.
Whatever is freshest wins. The current-news engine.
Perplexity is the freshness engine. It privileges recent press over historical authority. A brand that has been in the SA editorial conversation in the last fortnight reads as the current authority here. A brand that has gone quiet drops out of Perplexity fastest. It cites News24, IOL, Daily Maverick, BusinessTech and specialist outlets with date stamps prominently.
Best engine for a brand running a press cadence. Worst engine for a brand that spends its budget on production and not on the SA editorial relationship. The freshness decay curve for your brand is in the Scorecard.
For category-specific reads on how each engine handles a sector, see the per-industry pages — for example banking, automotive, medical aid and telecom.
The SA citation landscape — who actually shapes the answers.
A short list of South African sites does most of the citation work across all four engines. Below are the names that turn up in answer after answer. None of them is a surprise to a journalist. All of them get under-credited by marketing teams that still think of "SEO" and "PR" as separate budgets.
The SA business-news default — every engine reaches for it first on finance, telecom, retail and policy questions.
Where Claude and ChatGPT go for the analyst read — markets, banking, retirement, the CFO question.
The SA tech and consumer-pricing authority. Forums leak into Gemini. The editorial leaks into all four engines.
The SA telecom and enterprise-tech reference. Heaviest weight in B2B answers.
The freshness layer for general news. AI reaches here when the question is news-shaped — launches, recalls, leadership changes.
Broad SA news authority. Carries weight across categories but rarely the deciding source on a brand question.
The investigative and long-form weight. ChatGPT and Claude lean on it for the "what is really going on" framing.
The SA pre-purchase reflex. Gemini reads it. ChatGPT and Claude largely do not. The single biggest engine asymmetry we measure.
The default authority on every car question. The site every engine reaches for when the answer is about a SA nameplate.
The default authority on every property question — buy, rent, suburb, semigration.
The SA restaurant reference. The category where it most clearly outranks international alternatives.
The "what does it actually cost" authority for banking and short-term insurance. A specialist that punches above its size.
The Hellopeter signal — the asymmetry no one talks about.
Hellopeter is the SA pre-purchase reflex. Before South Africans switch banks, change insurers, sign a fibre contract or book a hotel, a meaningful share of them check Hellopeter. The platform is read deeply by Gemini and largely ignored by ChatGPT and Claude. That one asymmetry produces some of the largest engine-to-engine disagreements we measure.
A brand with active complaint volume can read clean in ChatGPT and rough in Google AI Overviews — different shoppers, different answers, same question. The team that manages Hellopeter response in your business is unknowingly managing your Gemini answer too. The per-brand Hellopeter posture is part of the free Scorecard.
The international sources that leak in.
For some SA categories, international sources carry more of the answer than SA editorial does. Safari lodges leak heavily to Condé Nast Traveler, TripAdvisor and Travel + Leisure. Hotels lean on Booking.com and TripAdvisor. High-end property leaks to Knight Frank and Savills, especially around the semigration conversation. Streaming sits with international entertainment press as much as it does with Showmax-specific SA coverage.
Leak is not a flaw. It is a measurable pattern. The category-by-category SA-source share runs from over 70% in short-term insurance to about 3% in safari lodges. Knowing which side of that range your category sits on is the first lever in any AI-visibility strategy. The full table is in the Q2 Index Report.
The wider source taxonomy.
Beyond the dozen sites above, each industry has its own specialist layer. Automotive has TopAuto and IOL Motoring. Banking has Daily Investor and RateWeb. Real estate has Bizcommunity and Private Property. Restaurants has Eat Out and TheJuice. The specialists matter because they are where the long-tail wins live — categories where a well-placed piece can move the engine answer faster than a national press release ever does. Our full source taxonomy and per-domain pages are in the bench.
For the SA-source share specific to your industry, see the relevant industry page or the research index.
AI search by industry — what is at stake.
Twelve South African consumer industries. Twelve different patterns. The engine behaviour is category-specific — what works for a bank does not work for a safari lodge — and the SA-source share varies more than most teams expect. Each row links to the full industry read.
These are not deep dives. Each industry has its own page with the per-engine posture, the source map, the SA brands ahead and behind, and the actions to take. Start here. Click through to the category that matters to you.
Banking
ChatGPT plays it safe with the Big 5. Gemini elevates Capitec and TymeBank far more readily. Hellopeter changes the Google AI Overviews answer entirely.
Telecom
Vodacom and MTN dominate every engine. The interesting movement is in Rain, Telkom and Cell C — each named cleanly by one engine and missing from another.
Medical aid
Discovery Health surfaces first in ChatGPT and Claude. Gemini pulls toward Bonitas, Momentum Health and the Hellopeter-shaped comparisons. Same question, two shortlists.
Automotive
Cars.co.za and AutoTrader do the heavy lifting. ChatGPT defaults to heritage marques. Gemini and Perplexity reward the Chinese-import wave faster than the others.
Retail
Shoprite, Pick n Pay, Woolworths and Checkers anchor every engine. Where it breaks: discount and value framing, where AI confuses brands with their parent groups.
E-commerce
Takealot is the default. Then a long tail — Superbalist, Loot, Yuppiechef, Bash — that each engine surfaces in a different order. Amazon SA changes the whole conversation.
Streaming
Showmax leads the SA-specific answers. Netflix and Disney+ are named by every engine but framed as international. The "local content" question is where engines diverge most.
Short-term insurance
OUTsurance, Discovery Insure and Naked surface fastest. SA editorial coverage carries more of the citation weight here than in any other category we measure.
Real estate
Property24 and Private Property are the default authorities. International leak is heavy — Knight Frank, Savills — especially on the high-end and the semigration story.
Restaurants
Eat Out runs the show. TripAdvisor leaks in. Engines disagree heavily on the "best" question — by city, by cuisine, by price band. A category that rewards fresh editorial.
Hotels
Booking.com and TripAdvisor dominate. The SA editorial layer — Travel + Leisure SA, Getaway — appears thinner than the actual SA hospitality press warrants.
Safari lodges
The most internationally-leaked category we measure. Condé Nast Traveler, TripAdvisor, Travel + Leisure US — the SA editorial share runs in the low single digits.
For the cross-industry comparison — which categories carry the highest AI-visibility risk this quarter, which have the most room to move — see the Q2 Index Report. For the brand-specific position inside your category, run the free Scorecard.
Five patterns no one is talking about.
Five findings from SA-AEO-Bench that come up in every CMO conversation we have, and that almost no SA marketing team has on the dashboard yet. Each one is a real lever. Each one is in the data.
01 · The Reddit asymmetry.
Gemini cites Reddit hundreds of times on SA queries. ChatGPT and Claude cite it zero times in the same prompt set. That single difference shapes the engine where most SA shoppers actually meet AI — Google AI Overviews. A SA brand with an active SA-specific subreddit thread is being represented in Gemini answers without anyone on the marketing team knowing. A brand with no Reddit footprint at all is being represented by the SA consumer of last resort — usually a complaint, sometimes a comparison, occasionally a recommendation. The Reddit asymmetry is the first place we look when a SA brand asks why ChatGPT and Google AI Overviews are answering the same question completely differently.
02 · Hellopeter as the SA pre-purchase reflex.
The Hellopeter signal is what the SA shopper checks before they sign. The engines that read it answer SA questions like a SA shopper. The engines that do not, answer them like an offshore analyst. There is no SA category we measure where the Hellopeter posture is irrelevant — banks, insurers, telecoms, hotels, e-commerce, even car brands. The CFO conversation has not caught up with this yet. The customer-service team that owns Hellopeter response is materially shaping the Google AI Overviews answer about the brand. They almost certainly do not know that.
03 · Industry SA-source share varies by 25x.
From over 70% in short-term insurance to about 3% in safari lodges, the share of AI citations going to SA-domiciled sources varies more by category than anything else we measure. Short-term insurance is a SA-editorial fight — RateWeb, BusinessTech, Moneyweb, Daily Investor decide it. Safari lodges are an international-editorial fight — Condé Nast, TripAdvisor, Travel + Leisure decide it. The implication is structural: your AI-visibility playbook depends on which side of the spread your category sits on. The lodges in the Sabi Sand and the Kruger that win Gemini answers do so via international press placement, not Eat Out.
04 · The B2C-vs-B2B AI visibility gap.
SA B2C brands have a richer citation surface than SA B2B brands by a wide margin. A consumer bank gets named by AI more readily than a corporate bank. A retail brand more readily than its supplier. A medical aid more readily than an underwriter. The engines read where the consumer reads. B2B brands have, by and large, not built a parallel presence in the press the engines actually trust. The "we do not need AI visibility, our buyers are sophisticated" position is the position most likely to be wrong by the next quarterly cycle.
05 · Multilingual citations — Afrikaans surfaces, isiZulu does not.
The engines read Netwerk24 and Maroela Media. Afrikaans coverage of a SA brand surfaces in answers when the prompt is Afrikaans, and sometimes when it is not. isiZulu and Xhosa coverage is thinner in the citation pool than the speaker base warrants. That is a measurable underrepresentation, and a gap a SA brand with the right press strategy can move in. Cited Brands tracks multilingual share as part of the bench. See the research index for the latest cut. The CMO conversation about this is starting; the data is ahead of it.
Each of the five above is a lever. Each one is brand-specific in how it lands. The Q2 Index Report names the category-level posture for each. The free Scorecard names which lever has the highest payoff for your specific brand. The Demo walks the action plan through with our team.
How to win AI search in South Africa.
There is no universal play. There is a framework. Three pillars. Each one matters differently per category and per engine. The brand-specific weighting is in the free Scorecard. The framework is below.
Pillar one — source placement.
Get covered on the SA sites the engines actually read. Three to five sites carry most of the citation weight in every category. The names are not a mystery. In banking, they are BusinessTech, Moneyweb, Daily Investor and RateWeb. In automotive, Cars.co.za, AutoTrader, IOL Motoring and TopAuto. In retail, BusinessTech, MyBroadband and the consumer-press long tail. The brief is simple: get reviewed, get profiled, get quoted, and do not stop. The press relationship is the AI-visibility relationship now, and the brands that treat them as separate budgets get out-cited by the ones that do not.
The lever beneath the lever: Wikipedia. A clean, well-cited Wikipedia entry is the single highest-payback entity signal across all four engines. Most SA brands either do not have one, or have one that has not been touched in three years. Both states are fixable. The category-level Wikipedia posture for your industry is part of the measurement methodology.
Pillar two — narrative control.
The engine names you. What it says about you is the next question. A brand named with "the digital challenger" framing reads completely differently from one named with "the legacy player" framing, in the same answer. That framing comes from the editorial language in the sources the engines read. The brand that has not done the work to seed that language gets whatever the SA press happened to land on, sometimes accurate and sometimes not.
The most common SA failure here is entity confusion. AI confuses one SA brand with a foreign namesake. Conflates a sub-brand with its parent. Mis-attributes a product to a competitor. Each one of these breaks individually. Together they shape a brand’s entire AI-search presence. The free Scorecard names the entity confusions currently active against your brand, and the consideration impact of each one.
Pillar three — freshness.
Perplexity drops you in weeks. Gemini drops you in months. ChatGPT and Claude drop you over quarters. A brand with a current press cadence reads as the current authority to the engines that update fastest. A brand that went quiet six months ago has already half-decayed out of Perplexity and is on the slope in Gemini. Recovery is slower than the loss — the engines need months of fresh signal to rebuild a default answer.
The freshness lever is the cheapest to pull and the easiest to neglect. It is also the one most exposed to load-shedding-recovery cycles, election cycles, results cycles — the SA news rhythm reshuffles the editorial calendar in ways that the engines pick up within fortnights. A brand that publishes results into a quiet news fortnight gets a disproportionate share of citations. A brand that publishes into a noisy one gets buried. Timing is leverage. The freshness state for your brand is in the Scorecard.
Each of the three pillars carries different weight for different SA brands. A heritage retailer needs almost no source-placement work and a lot of freshness work. A digital challenger is the inverse. The free Scorecard returns the brand-specific weighting and the three highest-payback actions to take in the next quarter.
Questions South African CMOs actually ask.
What is AI search?
AI search is what happens when a shopper, a CFO or a journalist asks ChatGPT, Claude, Gemini or Perplexity a question and gets an answer instead of a list of links. The engine reads the open web, picks three or four sources it trusts, and composes a short paragraph naming a handful of brands. Those brands get evaluated. Everything else does not. In South Africa, the four engines disagree more than most marketing teams realise — same question, four different SA shortlists.
What is AEO?
AEO — Answer Engine Optimisation — is the work of getting an SA brand named, accurately, in answers produced by ChatGPT, Claude, Gemini and Perplexity. It is not SEO with a new sticker. The signals that move AI answers are different: editorial placement on the SA sites the engines actually read, complaint posture on Hellopeter, freshness in BusinessTech and Moneyweb, structured presence on Wikipedia and category-specific specialist sites. See our AEO primer for SA brands for the longer read.
How do I rank in ChatGPT for South African queries?
You do not rank in ChatGPT — you show up in its answers, or you do not. The lever is editorial coverage on the SA sites ChatGPT actually reaches for (BusinessTech, Moneyweb, MyBroadband, Cars.co.za, the SA category specialists), plus a clean Wikipedia presence and consistent entity signal. ChatGPT is the most heritage-leaning of the four engines in South Africa — it favours brands with a long press tail. Your starting picture is in the free Scorecard.
Does Google AI Overviews use Gemini?
Yes. Google AI Overviews — the AI-generated answer that now sits above the blue links for many SA Google queries — is powered by the Gemini family. That makes Gemini the single most consequential engine for SA brands, because most South Africans still start a search on Google. Gemini reads forums (Reddit, MyBroadband), reads Hellopeter, and weights freshness more than ChatGPT or Claude. The full read on Gemini in SA is in the Q2 Index Report.
How do I check if my brand is cited by AI?
The fastest way is the free Scorecard. It runs a structured set of category-relevant prompts against the engines that matter in SA and tells you where your brand currently surfaces, where it does not, and which SA sources are shaping the answer about you. A deeper per-brand audit is available via the Demo.
What is the Hellopeter signal in AI search?
Hellopeter is the SA pre-purchase reflex — a complaint and review site that South Africans check before they buy. Gemini reads it. ChatGPT and Claude largely do not. That asymmetry means a brand with active complaint volume can read clean in ChatGPT and rough in Google AI Overviews — different answers for different shoppers from the same question. The category-level pattern is in the research index.
Is AI search bigger than Google in South Africa?
Not yet, and not by share of total queries. But the queries that have moved to AI are disproportionately the high-intent ones — the "best medical aid", "which bank should I switch to", "is this car worth the money" questions where the answer determines the purchase. Google AI Overviews has also brought generative answers into the Google experience itself. The practical position: AI search is already shaping the SA shortlists shoppers consider, and the gap will widen.
Which AI engines matter most for SA brands?
Four. ChatGPT (GPT-5) is the most-used standalone assistant in SA. Gemini powers Google AI Overviews — so it shapes what most SA Google users see. Claude (Sonnet 4.5) is the engine of choice for the analyst class — CFOs, consultants, journalists. Perplexity is the freshness engine and the one most likely to cite a recent SA news story. A brand needs a posture in all four. The per-engine portrait is in the Index Report.
What South African sources do AI engines cite most?
A short list does most of the work. BusinessTech, Moneyweb, MyBroadband, TechCentral, IOL, News24, Daily Maverick, Hellopeter, Wikipedia, and a handful of category specialists (Cars.co.za, Property24, Eat Out, RateWeb). The exact mix varies by industry — short-term insurance is dominated by SA editorial; safari lodges leak heavily to international sources like TripAdvisor and Condé Nast. The breakdown per industry is in the Q2 Index Report.
Does AI search read content in Afrikaans, isiZulu or Xhosa?
It reads them — unevenly. English-language SA sources do most of the citation work across all four engines. Afrikaans coverage on Netwerk24 and Maroela Media surfaces in specific categories. isiZulu and Xhosa coverage is thinner in the engines today than it should be given the speaker base. Cited Brands tracks multilingual citation share as part of the SA-AEO-Bench measurement.
How often does AI search update its answer about my brand?
Per engine, very differently. Perplexity refreshes within days — a recent press cycle can move the answer in a fortnight, and a silent quarter can drop you out of it. Gemini updates over weeks. ChatGPT and Claude decay over months, which means heritage brands carry inertia, and challengers carry catch-up. The freshness decay curve for your brand is in the free Scorecard.
What does Cited Brands actually measure?
SA-AEO-Bench — a pre-registered quarterly measurement of how ChatGPT, Claude, Gemini and Perplexity cite 100 South African brands across 12 industries. 19,020 successful AI responses last cycle. 232,254 cited URLs classified against a published source taxonomy. Pre-registered on OSF before data collection at osf.io/w4az2. The methodology page has the full design.
Where to go from here.
Q2 SA AI Visibility Index Report
The category-level read across all 12 industries. Source share, engine posture, what changed this quarter.
The free Scorecard
Where your brand sits on each engine. Which SA sources shape your answer. The three actions with the highest payoff this quarter.
Where SA brands need to get cited in AI search
The cross-industry SA-domain priority list. The sites that earn their citations. The ones that do not.
What is AEO, for a SA brand?
Answer Engine Optimisation, explained without the jargon. What it is, what it is not, and what a SA brand should be doing about it this quarter.
SA-AEO-Bench methodology
Pre-registered design, prompt set, source taxonomy. The full transparency stack for the SA-AEO-Bench measurement.
The knowledge library
Every guide, primer, glossary entry and quarterly write-up we have published on AI search in South Africa, in one place.
Brands we have written about in depth.
Quarterly per-brand reads on the SA names most asked about in AI search: Discovery, Capitec, MTN, Vodacom and Shoprite.
For working vocabulary, see the glossary. For the ongoing research, see the research index.
See where your brand stands on each of the four engines.
The free Scorecard runs the SA-AEO-Bench prompt set against your brand and returns the per-engine posture, the SA sources shaping your answer, and the three actions with the highest payoff. Twelve minutes to set up. No card needed.