South African AI Visibility Audit 2026: Which Brands Are Cited by ChatGPT, Perplexity, and Google AI Overviews?

If you sell anything in South Africa and you have not yet thought about how AI search works locally, this report is for you.

We spent the last six days running a study. The question was simple: when a South African business owner asks ChatGPT, Perplexity, or Google "what is the best X in South Africa," which brands actually show up in the answer? We ran 79 queries across the three platforms, then ran the whole thing again six days later to test whether the answers held steady. They mostly did.

Here is what we found.

The short version

Discovery is the most AI-visible brand in South Africa. It got cited in roughly one in four queries we tested. Three times the average for the 25 brands we tracked. The dominance held across both runs of the study.

Google's AI Overviews showed up on 93% of the South African commercial category queries we ran. That is a long way above the 25 to 40% rate Google has reported globally. If you sell to South Africans on Google, the AI Overview is not a side feature any more. It is the answer.

FNB is the AI-default South African bank. Capitec sits second, and the gap between them widened slightly between our two runs. TymeBank is the consistent third pick.

The telecoms category came back as a near-perfect three-way tie. MTN, Vodacom, and Cell C are interchangeable as far as the AI engines are concerned. None of them stand out.

Travel and hospitality is the lowest-visibility category in our data. FlySafair appeared a handful of times. Tsogo Sun got three citations across 79 queries. Most of the South African travel brands you would expect to see did not show up at all.

The most-cited source domains across all three AI platforms are Facebook, BusinessTech, hippo.co.za, YouTube, and Wikipedia. A quietly surprising mix, and worth thinking about if you are planning your own AI visibility strategy. We will come back to it.

Why we ran this

Honey Whale built one of South Africa's first dedicated Answer Engine Optimisation service line in 2025. We needed our own data. There was no published study of how SA brands appear in AI search, and the US and UK studies are not useful here because South African brands do not have the same press footprint as their global equivalents. So we built the study.

The audit is meant to be reproducible. The full methodology brief, prompt list, and raw dataset are available on request to researchers and journalists. Email info@honeywhale.co.za. We will refresh this study annually. This is the 2026 edition.

How we did it

We picked 25 South African brands across 9 commercial categories: ecommerce, banking, insurance, telecoms, property, food, logistics, travel, and healthcare. The brands are well-known names like Discovery, FNB, MTN, Vodacom, Takealot, Nando's, The Courier Guy, FlySafair, and Netcare. We added Honey Whale as the 26th, partly as a check on our own AI visibility (zero, as it turned out, which is honest data).

For each category we wrote three commercial prompts a real South African buyer might type. Things like "What is the best bank in South Africa?" and "Best courier services in South Africa." Twenty-seven prompts in total.

We ran every prompt three times: once through ChatGPT (model: gpt-5) via the OpenAI API, once through Perplexity (model: sonar-pro) via the Perplexity API, and once through Google's AI Overview via SerpAPI with South African localisation. That gave us 81 queries on Day 1. Two of the Google searches did not return an AI Overview, so the usable AIO dataset was 25 of 27. We recorded the full text of every answer, the cited sources where the platform provided them, and the brand mentions inside the response.

Then, six days later, we ran the whole thing again. Same prompts, same models, same scripts. We compared the two datasets brand by brand to see what shifted.

The stability check matters because AI answers are not static. They are influenced by whatever fresh content the underlying models have indexed since the last query. If results swing wildly between runs, the leaderboard is noise. If results hold, you are looking at signal.

Headline findings

Across both runs, Discovery was cited in 18 queries on Day 1 and 16 queries on Day 6. Roughly a quarter of every query we ran. The next closest brand was FNB at 13 citations both days.

Discovery wins for three reasons we could identify from the data. The brand name reinforces across all its subsidiaries (Discovery Health, Discovery Bank, Discovery Insure), so every mention of any product strengthens the parent. The Wikipedia article is substantial and Wikipedia turned up as a top-five citation source across our dataset. And Discovery has heavy editorial coverage in the publications AI assistants treat as authoritative, particularly BusinessTech and Moneyweb.

If you are trying to build AI visibility for your own brand, Discovery is the case study. The work is editorial presence, encyclopedic content, and consistent naming across product lines. Not paid links, not promotional content.

FNB sat at 13 citations on both Day 1 and Day 6. Capitec moved from 8 to 11 across the two runs. TymeBank held at 7 and 8. Together those three brands account for the vast majority of banking-related AI answers we saw.

Within the three banks we measured (FNB, Capitec, TymeBank), AI citations clustered heavily on the digital-first names. Our scoping brand set did not include Standard Bank, Nedbank, or ABSA, so we cannot report measured volumes for them. Raw AI response text shows the traditional retail banks do appear in AI banking answers at meaningful levels; the 2027 edition of this study will expand the brand set to capture them systematically.

The category-level signal worth flagging is that AI assistants are confidently picking a small set of names as canonical answers to SA banking questions. For SA banks competing for that visibility, the levers are editorial content, Wikipedia depth, structured data, and consistent answer-first content on the brand's own site.

MTN, Vodacom, and Cell C came back at 9, 9, and 8 citations on Day 1. On Day 6 the numbers were 9, 8, and 6. Almost statistically equivalent. The AI assistants are treating them as interchangeable answers to South African telecom queries.

This is unusual. In most categories the AI engines pick a winner. Telecoms is the one place where they refuse to.

Our brand set did not include Telkom, Rain, or Vodafone Business in this scoping cycle. Raw AI response text shows Telkom and Rain do appear in AI telecom answers; we will measure them systematically in the 2027 edition. If you work in telecoms and want a shot at category leadership in AI answers, the opening is not against MTN, Vodacom, or Cell C directly. It is against the assumption that they are interchangeable. A consistent, well-structured "Why X is the best South African mobile network" answer page on your own site has a real shot at being cited as the breaker.

Discovery Health appeared in 10 healthcare queries on Day 1 and 9 on Day 6. Netcare appeared 5 times both days. Our brand set did not include Mediclinic or Life Healthcare in this scoping cycle; raw AI response text confirms both appear in AI healthcare answers at volumes comparable to Netcare. The 2027 edition will expand the healthcare brand set.

The travel category had the sparsest measured presence. FlySafair, the only major SA airline in our tracked brand set, was the only travel brand to consistently show up, with 5 citations on Day 1 and 6 on Day 6. Tsogo Sun got 3 citations both days. We did not include South African Airways or hotel groups in this cycle; raw text confirms SAA appears in AI travel answers occasionally.

Within the brands we measured, travel and hospitality has the sparsest AI citation footprint. A South African travel business that moves first on Answer Engine Optimisation, with structured FAQ content and citation-ready source material, has a meaningful opportunity to establish category authority while the rest of the industry is still optimising for Booking.com referrals.

This is the most consequential finding for South African digital marketers.

Independent SEO research firms (Conductor, BrightEdge, Semrush) have reported AI Overview appearance rates between roughly 25% and 48% globally across general search queries. Our SA category-level commercial data shows 93%. Commercial queries trigger AI Overviews more often than informational queries, so the two numbers are not perfectly comparable, but the size of the gap suggests SA commercial categories are seeing AI Overviews at substantially higher rates than the global all-query average. Out of 27 category queries we tested, 25 triggered an AI Overview. The two that did not were "Cheapest flights within South Africa" and "Best hospital group in South Africa." Every banking query, every retail query, every insurance, telecoms, property, food, and logistics query came back with an AI-generated answer at the top of the Google search results page.

The implication for SA businesses is direct. If your customers find you through Google, the AI Overview is now the answer they see first. If you are not cited in it, you are below the fold for an increasing share of your category's search demand.

This is the practical case for taking AEO seriously in South Africa, in 2026, not in two years. The infrastructure shift has already happened locally. The work is to be cited.

This is the surprise of the dataset.

Across both runs combined, Facebook was the single most-cited source domain, with 52 citations out of 1,032 total source citations in our dataset. BusinessTech sat second with 37. hippo.co.za third with 34. YouTube is fourth with 29. Wikipedia is fifth with 24. Reddit sixth with 21.

Three quick observations.

Social platforms account for roughly 11% of all source citations in our combined dataset (Facebook 5.0%, YouTube 2.8%, Reddit 2.0%, Instagram 1.2%). This breaks the assumption that authoritative business publishing dominates AI answers. In South Africa specifically, where the editorial publishing footprint is thinner than in the US or UK, social platforms have stepped into the citation gap.

Brand-owned domains earn surprising amounts of AI citation. King Price's own .co.za site was a top-20 source. Outsurance.co.za appeared 13 times across both days. The Courier Guy and Aramex's own sites appeared multiple times. AI citations are not exclusively a third-party publisher game. Your own site, structured properly, gets cited.

We included Honey Whale in the brand set as a check on our own visibility. The result was zero citations across all 158 queries, both runs combined. This is consistent with how AI assistants treat agency queries. They list consumer-facing brands when asked about commercial categories, not the agencies behind those brands.

The implication for South African agencies is that the path to AI visibility is to build category-defining content that gets cited on behalf of clients, not to optimise our own agency page for direct citation. The asset is the work, not the brand.

How stable are the findings?

We re-ran the entire 79-query suite six days after the Day 1 baseline. Same prompts, same models, same scripts. The point of the re-run is to test whether AI answers swing week to week. If they do, the leaderboard is noise. If they hold, you are looking at signal.

The signal held.

Total brand mentions on Day 1: 182. Total brand mentions on Day 6: 183. Net change of less than 1%. Nine of 26 tracked brands held exactly the same citation count across both runs. Seventeen brands moved, but mostly by one or two mentions on small bases. The top five most-cited South African brands in AI search held their positions from Day 1 to Day 6. Discovery remained the most-cited brand across all three platforms. FNB held its position as the AI-default bank. The telecoms three-way tie persisted. The Courier Guy continued to dominate logistics citations.

A few movers are worth noting.

Capitec gained three citations between Day 1 and Day 6 (8 to 11). The digital-banking trio of Capitec and TymeBank both ticked up at the expense of the more traditional names. If the trend holds across future runs of this study, digital banks are eating AI visibility share at the same rate they are eating account share in the real world.

Aramex SA also gained three (3 to 6), narrowing The Courier Guy's lead in logistics citations slightly.

Spur lost two citations between runs (4 to 2). Discovery itself shed two (18 to 16) but remained the runaway leader.

Source domain composition was also broadly stable. Facebook was the top source on both days. BusinessTech and hippo.co.za held their top-five positions. The domain mix is not random. The AI assistants are pulling from a consistent set of sources, and that set is dominated by social media platforms and a small group of established South African business publications.

The takeaway from the stability check is that the Day 1 findings are durable signal, not snapshot noise. South African brands that show up in AI answers tend to keep showing up. South African brands that do not appear at all on Day 1 still do not appear on Day 6. The leaderboard holds.

Per-platform notes

ChatGPT, Perplexity, and Google AI Overview did not always agree. The disagreements are themselves interesting.

ChatGPT mentioned at least one tracked brand in 24 of 27 queries (89%). The other two platforms hit 100%. ChatGPT also tended to produce longer, more historical brand mentions, while Perplexity and AI Overview tended to converge on a tighter set of currently-relevant brands.

For South African marketers planning Answer Engine Optimisation, the practical recommendation is to optimise for Perplexity and Google AI Overview first. They agree more, they pull from cleaner source sets, and they are where the citation gain is most measurable. ChatGPT visibility tends to follow once the underlying entity signals (Wikipedia, schema, content depth) are in place.

What this means for South African businesses

If your business depends on commercial search visibility in South Africa, the strategic question has shifted. It is no longer "do we rank?" It is "do we get cited?"

Four practical priorities for the next 12 months.

1. Audit your own AI visibility this week. Run five to ten category queries on ChatGPT, Perplexity, and Google for the questions your customers actually ask. Note which brands appear instead of yours, and which source domains the AI is citing. The gap tells you the work.
2. Build category-level answer pages on your own site. AI assistants need a citation-ready source. A page titled "What is the best [your category] in South Africa?" with structured data, FAQ schema, and a clear answer in the first 80 words is the foundation.
3. Update your Wikipedia entry if one exists. Commission a draft if it does not. Wikipedia turned up as a top-five citation source across our dataset. It is the cheapest credibility floor available for AI visibility.
4. Earn editorial coverage in BusinessTech, Moneyweb, and category-specific SA publications. A single article in BusinessTech is materially more valuable for AI visibility than ten syndicated press releases. The publication that AI assistants cite most often is the one to pitch first.

A note on methodology

This study is reproducible. The full prompt list, brand list, scoring rubric, Python scripts, and raw CSVs are available on request to researchers and journalists. Email info@honeywhale.co.za. Any independent observer can re-run the queries, extend the brand set, or critique the framing. One data integrity disclosure: the Day 1 raw ChatGPT response CSV was inadvertently overwritten in a local file system event before this report was published. The per-brand Day 1 ChatGPT citation counts used in this report were preserved via the brand visibility matrix produced during the Day 1 analysis on 10 June 2026, and that matrix is available on request. Day 1 Perplexity and Day 1 Google AI Overview raw responses are intact and available, as are all Day 6 raw responses.

A few transparency points.

We used the OpenAI API for ChatGPT rather than the web interface. API responses can differ from web responses, particularly because Perplexity's web product runs additional citation orchestration on top of the model. API results are the most consistent across runs, which is why we chose them, but they may slightly understate the citation diversity a typical user would see in the web interface.

Google AI Overviews were captured via SerpAPI with South African localisation. SerpAPI scrapes the live Google search results page from a SA-localised IP and returns the AI Overview structured data when present. This is the same data a SA-based user would see in their browser.

Honey Whale included itself as the 26th brand. None of the other 25 brands commissioned, sponsored, or were aware of this study. We have no commercial relationship with any of the brands in the leaderboard.

We plan to refresh this study annually. The 2027 edition will expand to 50+ brands, add Claude as a fourth AI platform, include longitudinal comparison, and bring in expert review from independent South African SEO practitioners.

Frequently asked questions

What is Answer Engine Optimisation (AEO)?

Answer Engine Optimisation, also known internationally as Generative Engine Optimisation (GEO), is the practice of structuring content so AI assistants like ChatGPT, Perplexity, Claude, and Google AI Overviews cite your brand as the answer to relevant buyer questions. It builds on traditional SEO but optimises specifically for AI-generated answers, not just blue-link search rankings.

Why is this audit specific to South Africa?

AI visibility patterns vary materially by country. South African brands and South African publishers have different presence in AI training data than US or UK equivalents. A US-focused AI visibility study would tell a South African business almost nothing useful. This is the first SA-specific study we are aware of.

Why only 25 brands?

The 25 brands cover the most commercially significant South African brands across 9 categories that represent meaningful search demand. We plan to expand to 50+ brands in the 2027 edition. The current scope is the minimum viable dataset for category-level conclusions.

How is this different from a Google search ranking study?

Traditional Google search rankings measure which URLs appear in the blue-link results. AI visibility measures which brands appear in the AI-generated answer that increasingly sits above or replaces those blue links. They are correlated but different. A brand can rank position 1 in traditional Google and still be invisible in AI Overviews if the AI cites a different source.

Can a small South African business compete for AI visibility against Discovery and FNB?

In their categories, no. In smaller niches, yes. AI assistants give smaller brands real visibility on category-specific or geographic queries. The AEO playbook for small SA businesses is to win the long-tail category questions rather than fight for the head terms that incumbents dominate.

Will AI search replace Google?

No. AI Overviews and AI assistants are layered on top of Google search, not replacing it. Both interfaces will coexist for the foreseeable future. The shift for South African businesses is that you now need to be visible in both AI answers and traditional rankings to capture full commercial search demand.

Can I download the raw data?

Yes, on request. Email info@honeywhale.co.za and we will share the full methodology brief, prompt list, and dataset. We just ask that you credit the study if you cite or extend any of it.