SA Language Search: What South Africans Actually Search In

In March 2026 Google added Afrikaans, isiZulu, Sesotho and Setswana to AI Overviews and AI Mode, as part of an expansion to thirteen Sub-Saharan African languages. Google stated the languages were selected on the basis of existing search activity, which is the detail worth pausing on: this was not a goodwill gesture ahead of demand. It was a response to search volume that already exists. What it changes for a South African business is smaller than the announcement suggests and more interesting than it first appears, because the shift it points at is not translation. It is voice.

This article is about search behaviour and visibility. Whether your website should be built in more than one language is a separate decision with its own economics, covered in a companion piece.

What Google Actually Did

The expansion covers two products. AI Overviews, the summary that appears above search results, and AI Mode, the conversational interface that supports follow-up questions through text, voice or images. Both now operate in Afrikaans, isiZulu, Sesotho and Setswana alongside English.

The wider release covers thirteen languages across Sub-Saharan Africa, including Kiswahili, Hausa, Yoruba, Amharic and Wolof. Google’s stated selection criterion was strong and growing usage in Google Search across the continent.

The detail deserving more attention than the announcement itself is what the expansion was built on. Google describes it as drawing on its Waxal language project, and Waxal is not a translation effort. It is a large-scale open speech resource covering 27 Sub-Saharan African languages, carrying both automatic speech recognition and text-to-speech data, with the collection work led by African academic and community organisations and the dataset published under an open licence. The name means to speak in Wolof.

That is a substantial investment in machines understanding these languages when they are spoken rather than typed. Hold onto that, because it is the thread running through the rest of this article.

Kabelo Makwane, Google’s country director for South Africa, framed it in terms of exclusion: when technology only speaks a dominant international language, it marginalises people whose first languages reflect a different culture and identity.

Two things follow that are commercially relevant. Google has invested in understanding these languages well enough to generate answers in them, which is a much higher bar than merely matching keywords. And Google’s own usage data indicated the demand was already there.

The Gap Between Home Language and Search Language

Here is the tension that makes this genuinely interesting rather than straightforward.

Census 2022 records isiZulu as the language most often spoken at home by 24.4 percent of the population, isiXhosa at 16.3 percent, Afrikaans at 10.6 percent and Sepedi at 10 percent, with English fifth at 8.7 percent, Setswana at 8.3 percent and Sesotho at 7.8 percent.

Commercial search in South Africa does not remotely reflect that distribution. It is overwhelmingly English, and any agency that has looked at Search Console data for South African businesses across sectors will tell you the same.

The reasons are structural rather than attitudinal, and they are worth understanding because they explain which parts of the gap will close and which will not.

The vocabulary of commerce is English. Products, services, professions and technical categories are named in English in South African trade. There is often no widely used isiZulu equivalent for a specific industrial component or a professional specialisation, because the term entered the language in English and stayed there. You cannot search in a language that lacks the noun.

Search is a learned behaviour and it was learned in English. People type what has historically worked. A decade of English-only results trained an entire market that English is what you search in, and behaviour adapts to systems more readily than systems adapt to behaviour.

Code-switching is the actual norm. South Africans routinely mix languages in a single sentence, and search queries reflect this. A query may carry an English product noun inside an isiZulu question structure. That is neither an English query nor an isiZulu one, and it is far more common than either.

Written and spoken usage diverge. Many South Africans who speak isiZulu daily do most of their reading and writing in English, because that is the language they were schooled and employed in. Search has historically been a written activity.

That last point is the hinge, and it is where the situation is genuinely changing.

Voice Is Where the Shift Actually Happens

If the barrier is that people speak one language and type in another, then the thing that changes the picture is not translated web pages. It is search that does not require typing.

This is why AI Mode supporting voice matters more than AI Overviews supporting text, and it is why the Waxal detail earlier is the most telling part of the announcement. Google did not invest three years in collecting speech recognition and text-to-speech data across 27 African languages in order to serve better written results. It built the capability to hear these languages.

Speaking a query in isiZulu carries none of the friction of typing one. There is no spelling to negotiate, no keyboard layout to fight, no learned habit of switching to English at the search bar. People speak in the language they think in, and they always have.

Layer the South African context onto that. This is a mobile-first market on metered data, where voice input is faster than typing on a phone, and where a substantial part of the population is more comfortable speaking a language than writing it. Every one of those conditions favours voice, and voice favours home languages.

So the reasonable expectation is not that written isiZulu search queries suddenly surge. It is that spoken queries in South African languages grow, that they arrive through AI Mode rather than the traditional results page, and that they look conversational rather than keyword-shaped, because that is how people speak.

That is a meaningful change in how questions are asked. It is a much smaller change in what businesses should do about it, for a reason covered next.

What This Means for Your Website, Honestly

The tempting conclusion is that you should translate your website to capture this. For most businesses that is the wrong inference, and it is worth setting out why carefully.

These are answer-generation systems, not page-matching systems. AI Overviews and AI Mode generate a response from sources. Google’s models operate across languages, and a system capable of generating a fluent isiZulu answer is capable of drawing on English source material to do it. The requirement to have the answer written in the language of the query is far weaker than it was in traditional search.

The queries shifting to home languages are mostly informational. Someone asking a question conversationally in Sesotho is usually learning something, not commissioning a supplier. Commercial and transactional intent, where the buyer knows the product name and is comparing options, is exactly where English vocabulary is most entrenched.

The local pack has not changed. For a South African business, the highest-value searches are location-anchored: a service plus a suburb. Those results are decided by proximity, Google Business Profile completeness, review volume and citation consistency. None of that is language-dependent in a way translation addresses.

The practical position for a typical Johannesburg business is therefore that this development is worth understanding and is not worth acting on yet. That is an unsatisfying answer and it is the accurate one.

What Is Actually Worth Doing

Three things follow that are useful, cheap, and do not require a multilingual build.

Write for conversational questions. The strongest overlap between the language shift and ordinary good practice is that voice queries are longer and phrased as questions. A website structured around real questions with direct answers underneath them performs better for voice search, better in AI Overviews, and better in ordinary search, regardless of language. This is the highest-return action on the list and it is language-agnostic.

Use the vocabulary your market actually uses, including local terms. South African English carries words with no international equivalent, and businesses routinely write around them in favour of formal alternatives nobody searches for. Bakkie, geyser, robot, stoep, braai, load shedding, spaza, tekkies. If your customers say bakkie, write bakkie. This is not a language project, it is accuracy, and it is one of the more reliable local search advantages available because international competitors cannot replicate it.

Record what appears for your own terms. If you want evidence rather than speculation, search your main service terms in Afrikaans and in isiZulu and see what Google returns. It takes fifteen minutes, it is specific to your sector, and it will tell you more than any general article including this one. Repeat it in six months and you will have your own trend line, which is worth more than a national statistic.

Who Should Watch This More Closely

The general answer does not apply evenly. Four categories should be paying more attention than the rest.

Consumer businesses in Afrikaans-dominant regions. Afrikaans has the strongest existing written search behaviour of the four languages added, a long history as a language of commerce and print, and mature machine translation support. Census 2022 puts it at 41.2 percent of the Western Cape population and 54.6 percent in the Northern Cape. Afrikaans search is a present reality rather than a projection.

Health, legal, financial and educational services aimed at the public. High-stakes informational queries are where people revert to their home language, particularly when anxious. This is also where AI Overviews are most heavily used, because people are looking for an explanation rather than a supplier.

Government-adjacent and public access services. Reaching the general population rather than a business audience means the language distribution is closer to the census distribution than to the commercial search distribution.

Anyone publishing genuinely educational content. If your content strategy is built on explaining things rather than listing services, you are in the part of search where this shift is happening first.

The Longer View

The reason to track this rather than dismiss it is that the constraint holding South African language search down was never demand. It was capability.

People did not search in isiZulu because searching in isiZulu did not work well. That produced low volume, which justified low investment, which kept it working poorly. Google’s expansion breaks that loop from the supply side, and Google’s own statement that the languages were chosen on existing search activity indicates the demand side was already moving.

Whether that becomes commercially significant depends on things nobody can currently forecast: how quickly voice search adoption grows in South Africa, whether AI Mode becomes a primary interface or stays a feature, and whether the commercial vocabulary itself shifts, which is a generational question rather than a technical one.

What can be said with confidence is that the direction is one way. There will not be less African language capability in search next year than there is now. Businesses that build websites structured around real questions, written in the vocabulary their customers actually use, are positioned for that shift without having bet anything on its timing. That is the sensible position: build for how people ask questions, and let the language question resolve itself when the evidence arrives.

Frequently Asked Questions

Which South African languages does Google AI Overviews support?

Afrikaans, isiZulu, Sesotho and Setswana were added in March 2026, as part of an expansion covering thirteen Sub-Saharan African languages across both AI Overviews and AI Mode. Google stated the languages were selected based on existing search activity rather than in anticipation of future demand.

Do South Africans search in English or their home language?

Commercial search in South Africa is overwhelmingly English, despite English being only the fifth most spoken home language at 8.7 percent in Census 2022. The reasons are structural: the vocabulary of commerce is English, search behaviour was learned in English, code-switching within a single query is common, and many South Africans read and write in English while speaking another language at home.

Should I translate my website into isiZulu to rank better?

For most businesses, no. AI Overviews and AI Mode generate answers rather than matching pages, and Google’s models can draw on English sources to answer a query asked in another language. Home language queries also skew informational rather than commercial, and location-based local searches are decided by proximity, Google Business Profile completeness and reviews rather than by language.

Why does voice search matter more than translated pages here?

The main barrier to home language search was never willingness, it was that many South Africans speak one language and write in another. Voice removes the typing barrier entirely, along with spelling and keyboard friction. In a mobile-first market on metered data, that makes voice the mechanism through which home language search is most likely to grow.

What can I do about this without building a multilingual website?

Structure your content around real questions with direct answers, which suits voice queries, AI Overviews and ordinary search equally. Use the vocabulary your customers actually use, including South African terms like bakkie, geyser and robot rather than formal alternatives. Then search your own service terms in Afrikaans and isiZulu to see what Google currently returns in your sector.

Which businesses should be watching language search most closely?

Consumer businesses in Afrikaans-dominant regions, where Census 2022 records Afrikaans at 41.2 percent of the Western Cape population and 54.6 percent of the Northern Cape. Also health, legal, financial and educational services aimed at the public, government-adjacent and public access services, and any business whose content strategy is built on explaining things rather than listing services.