We measured 170,720 citations across 39,258 ChatGPT, Perplexity and Claude responses to South African questions. The engines agree with each other far less than the industry assumes, and the sources they trust most are not the ones anyone is optimising for.
87.8%
of cited domains appear on only one AI engine
1.7%
appear on all three engines
170,720
domain-citations measured
39,258
AI responses analysed
Source: Algorithm Lighthouse GEO Visibility, production data as of 2 August 2026. Market: South Africa. Methodology and caveats published in full below.
Finding 01 · The overlap collapse
The assumption underneath most AI visibility strategy is that authority is portable: build it once and every engine will find you. We tested that against 170,720 citations. It does not hold.
Between May and August 2026 the dataset roughly quadrupled. More data would normally pull a measurement toward the mean. Instead the central finding got stronger. Cross-platform agreement fell.
| Metric | May 2026 | Aug 2026 |
|---|---|---|
| SA client projects | 43 | 99 |
| Completed Lighthouse runs | 50 | 178 |
| LLM provider responses | 8,242 | 39,258 |
| Total domain-citations | 46,315 | 170,720 |
| Platform-specific citations | 83.4% | 87.8% |
| Cited by all three engines | 3.1% | 1.7% |
| Cross-platform overlap | 16.6% | 12.2% |
Red indicates the metric moved against the “build once, appear everywhere” assumption. Source: Algorithm Lighthouse GEO, May and August 2026 snapshots.
Qwairy's four-platform analysis of 118,000 responses found 89% of citations were platform-specific. We measure 87.8% across three platforms in South Africa. Effectively identical, and the residual gap is plausibly the missing fourth platform. The fragmentation is not a South African quirk. It is how these systems work.
Finding 02 · The credibility hierarchy
These are the ten most-cited domains when AI engines answer South African questions. The four at the top are not South African, and not one of them is a publisher in the traditional sense.
Combined citation counts across ChatGPT, Perplexity and Claude. Green marks South African domains.
The finding that matters most: the four most-cited sources answering South African questions are YouTube, Facebook, Reddit and Wikipedia. Video, two social networks and an encyclopedia. For most brands, three of those four sit outside the SEO remit entirely, which is exactly why the citations are still available.
Finding 03 · Three credibility models
Each engine has a distinct posture toward sourcing, and all three became more idiosyncratic between May and August. These are not three views of one truth. They are three different definitions of what counts as credible.
The encyclopedic and community web
78.5%
unique to it
6.89
sources / reply
97.4%
cite anything
Top 5 SA sources
Wikipedia and Reddit are its first and second most-quoted South African sources, by a wide margin, before any South African institution appears.
Effectively a social engine
74.5%
unique to it
8.98
sources / reply
99.8%
cite anything
Top 5 SA sources
Its top four South African sources are all social platforms. It cites 30.3% more sources per response than ChatGPT and returns a source almost every time.
The most local of the three
61.8%
unique to it
3.96
sources / reply
69.1%
cite anything
Top 5 SA sources
Every one of its top five is a South African or Africa-focused site, with no global aggregators at all. But it returns no source whatsoever in 30.9% of responses.
Claude now returns no source at all in 30.9%of South African responses, against 0.2% for Perplexity. That is not a volume difference, it is a fundamentally different posture toward sourcing. If your AI visibility reporting treats the three engines as interchangeable, roughly a third of Claude's answers are invisible to it by definition.
Finding 04 · Sector variation
Overlap varies by category, but not by enough to change the conclusion anywhere. There is no sector in which optimising once and appearing everywhere holds.
| Sector | Prompts | Citations | Overlap |
|---|---|---|---|
| Financial Services | 2,016 | 34,047 | 15.7% |
| Education & Conferences | 161 | 2,880 | 14.9% |
| Telecoms & Connectivity | 127 | 1,983 | 14.8% |
| Marketing, Data & Technology | 520 | 8,892 | 13.5% |
| Travel & Hospitality | 570 | 10,186 | 13.4% |
| Industrial & Construction | 635 | 9,909 | 12.6% |
| Other / unmapped | 2,827 | 45,154 | 12.1% |
| Health & Wellness | 120 | 2,097 | 11.5% |
| Retail & Ecommerce | 2,679 | 41,761 | 10.4% |
| Food & Beverage | 179 | 2,859 | 10.1% |
| Facilities & Business Services | 676 | 10,952 | 6.2% |
Highest consensus
Financial Services, 15.7%
Authoritative South African financial sources such as SARS, the big four banks and established asset managers are recognised across engines. Even so, 84% of its citations remain platform-specific.
Most fragmented
Facilities & Business Services, 6.2%
In B2B services there is effectively no agreed-upon source of truth on the South African web. 93.8% of citations appear on a single engine. For challengers, that is an open field.
What to do about it
Research is only worth publishing if it changes a decision. These are the four decisions we think it should change.
With 87.8% of citations appearing on a single engine, a strategy that optimises for ChatGPT and assumes Perplexity and Claude will follow is measuring one third of the picture. Each engine needs its own baseline and its own target set.
YouTube, Facebook and Reddit are the three most-cited sources in South African AI answers. For most brands these sit outside the SEO remit entirely, which is precisely why the citations are available.
Claude rewards South African sources heavily and cites nothing at all a third of the time. Perplexity rewards social presence. The same investment produces very different returns depending on which engine your buyers use.
Facilities and business services shows 6.2% cross-platform overlap, the lowest of any sector. There is effectively no agreed source of truth on the South African web for these categories. That is a gap, not a dead end.
Methodology and caveats
Every prompt in the set was issued to ChatGPT, Perplexity and Claude within the same time window, so this is a controlled head-to-head comparison rather than three separate datasets stitched together. Responses were parsed for cited domains, and each unique domain within a unique prompt context was counted once.
Data comes from Algorithm Lighthouse GEO Visibility production as of 2 August 2026, filtered to market = South Africa. All figures on this page are reproducible from that system.
99
SA client projects
178
completed runs
10,510
prompts with 2+ providers
39,258
provider responses
Three platforms, not four. Gemini and Google AI Mode are not yet integrated into Lighthouse, so this measures ChatGPT, Perplexity and Claude only.
The sample is weighted to Algorithm's client base, with heavy representation in retail and ecommerce and in financial services.
A domain-citation counts a unique domain within a unique prompt context. Domains recur across prompts.
Each prompt was issued to all three providers within the same time window, so this is a controlled head-to-head comparison rather than a scrape of separate datasets.
The Other sector bucket still accounts for 27% of prompts. The v2 taxonomy is a substantial improvement on v1 but is not exhaustive across 418 entity categories.
All figures are cleared for external use. Journalists, analysts and other agencies are welcome to reference them. Please cite as:
Algorithm, “Who AI Engines Cite in South Africa”, August 2026. https://www.algorithm.agency/guides/who-ai-engines-cite-in-south-africa
For the underlying methodology, a sector-level cut, or commentary for a story, contact us via the contact page.
Frequently asked questions
It measures which websites ChatGPT, Perplexity and Claude cite when answering questions about South African topics, and how much those three engines agree with each other. Every prompt was issued to all three engines inside the same time window, so the comparison is controlled rather than assembled from separate datasets. The dataset covers 170,720 domain-citations across 10,510 South African prompts and 39,258 provider responses.
Because it disproves the assumption most AI visibility strategies are built on. The common belief is that if you build authority, you will surface everywhere. The data says the opposite: the engines largely cite different sources, and agreement fell from 16.6% in May 2026 to 12.2% in August 2026 as the dataset quadrupled. Only 1.7% of cited domains appear on all three engines. Optimising once and expecting to appear everywhere does not work.
Because AI engines weight platforms they can parse, verify and trust at scale, and those platforms carry enormous volumes of South African content. The first South African organisation in the combined ranking appears fifth, behind YouTube, Facebook, Reddit and Wikipedia. It is a structural feature of how these systems source information, not a judgement on South African publishers.
It closes the gap to the global benchmark almost exactly. Qwairy's four-platform analysis of 118,000 responses found 89% of citations were platform-specific. Algorithm measures 87.8% across three platforms in South Africa. The residual difference is plausibly the missing fourth platform. Separately, Ahrefs found globally that YouTube mentions are the single strongest predictor of AI visibility, which this South African data independently confirms.
Yes. All figures on this page are cleared for external use and reproducible from Algorithm Lighthouse production as of 2 August 2026. Please cite as: Algorithm, Who AI Engines Cite in South Africa, August 2026. The methodology and caveats are published in full on this page so the numbers can be assessed honestly.
Lighthouse GEO is Algorithm's AI visibility platform, the first built in Africa. It issues controlled prompt sets to the major AI engines and measures which brands and domains get cited, how often, and against which competitors. This report is one output of that production system. See the Lighthouse GEO page for how it works, or our GEO and AEO agency page for how we apply it.
Continue reading
How Algorithm applies this data to get brands cited by AI engines.
The complete South African guide to GEO and AI search visibility.
Nine providers compared against six checkable criteria, including us.
We run the same measurement on individual brands. You get your citation share across ChatGPT, Perplexity and Claude, against the competitors in your category, on the prompts your buyers actually use.