AI Search Optimisation South Africa: What Google Says

Illustration of a person at a reading desk as the pages of an open book drift upward into a network of connected dots

AI SEO is sold as a new discipline with new rules. Google’s own documentation says otherwise, and reading it carefully is the fastest way for a South African business to stop wasting money. This sets out what genuinely changes when an AI system answers instead of listing links, what stays exactly as it was, and how to measure any of it without fooling yourself.

What is AI SEO, and what actually changes?

AI SEO is optimising to be found and quoted by systems that answer rather than list. What changes is the surface: an answer sits where ten blue links used to. What does not change, according to Google, is the work. Its guidance states there are no additional requirements and no special optimisations necessary. You will also see this sold as generative engine optimisation, or GEO, and as answer engine optimisation. They describe the same work under different names. Nothing in Google’s guidance changes depending on which label an agency uses.

The confusion comes from treating one word, visibility, as if it meant one thing. It is really five different events, and you control each one differently.

Stage What happens What you actually control
Crawling A bot fetches the page Your robots file, your host, your server response
Indexing The page is stored and understood Whether the content is in text, and whether you block indexing
Retrieval The page is pulled in as a candidate for one specific question Whether the page answers a real question in plain words
Citation The answer names or links your page Very little directly. You influence it, you do not set it
Referral Somebody actually clicks through Whether the snippet earns the click

Keeping those five apart is the single most useful discipline in this whole subject, because almost every bad claim collapses the moment you ask which stage it refers to. An agency promising you citations is promising an outcome at stage four while selling work at stages one and two. Google is unusually direct about the limit here: to be eligible as a supporting link in AI Overviews or AI Mode a page must be indexed and eligible to be shown in Google Search with a snippet, there are no additional technical requirements, and even then, as Google puts it, indexing and serving isn’t guaranteed. KM Digital Solutions quotes that sentence to clients before any AI-search work starts, because a programme built on an outcome nobody can promise is a programme built to disappoint. The same caution applies to every agency pitch a South African business is receiving on this subject right now.

What still works, and what can an AI system actually read?

Nearly all of it, which is the part the market least wants to hear. Google’s guidance on AI features lists its recommendations plainly: allow crawling, make content findable through internal links, give people a good page experience, keep important content in text, and make structured data match the visible text.

That list is worth reading twice, because of what is absent from it. There is no new file to publish, no new markup to add and no separate AI channel to buy into. That is as true for a firm in Cape Town as it is for one in London.

Google states it outright: you do not need to create new machine readable files, AI text files or markup to appear in these features, and there is no special schema.org structured data you need to add. The same documentation set records that llms.txt files are not needed for Google Search and will not affect visibility or rankings either way. It also records that the FAQ rich result was removed because that feature is no longer shown in search results. So three of the things most commonly sold as AI SEO in South Africa right now, an AI text file, extra schema and FAQ markup for rich results, are each documented by Google as either unnecessary or discontinued. None of that makes schema worthless. It makes schema a tool for Google rich results and entity clarity, which is a real job, rather than a lever on AI citations.

As for what a machine can use, the answer is the visible words first and foremost. The best available evidence says adding structured data to a page that is already being cited does not lift citations any further, which points the same way Google’s own advice does.

The strongest study on this is worth quoting precisely, including its limits. Ahrefs tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026 and compared them against 4,000 matched control pages. Citations in AI Overviews fell 4.6%, a small but statistically significant decline relative to the matched controls, while AI Mode and ChatGPT changes were indistinguishable from zero. The limitation matters as much as the result: every page in that dataset already had over a hundred AI Overview citations before any schema was added, so the study says nothing about pages that are not being cited at all. Read honestly, it tells you that schema will not push an already-cited page higher. It does not tell you to remove your markup.

What that leaves is prose. Answer the question in the first two sentences under a heading. Write self-contained sections, because a system chunking your page will not carry context across from three paragraphs earlier. Name your sources and date your figures. Use tables and lists where the content is genuinely list-shaped. KM Digital Solutions writes to that shape by default now, for readers first and parsers second.

What should a South African business do first?

Fix the boring plumbing, then earn mentions. The plumbing is a one-off afternoon. The mentions are the long game, and the evidence suggests they matter more off your site than on it. Neither requires a new budget line called AI, and both help ordinary search at the same time.

  1. Confirm the pages are crawlable and indexable. Nothing downstream can happen if a bot cannot fetch the page or is told not to index it. Our technical SEO checklist orders those checks by consequence.
  2. Choose your crawler settings deliberately, using each operator’s published controls rather than a copied robots file.
  3. Move key facts into visible text. Prices, locations, services and credentials belong in prose a reader can see, not only in markup.
  4. Answer real questions in self-contained sections, because a system chunking your page will not carry context across from earlier paragraphs.
  5. Go and get mentioned. Directories, roundups, professional profiles and genuine press all count, and they help ordinary search too. A complete Google Business Profile is the first of those profiles.
CORRELATION WITH AI OVERVIEW BRAND VISIBILITY 75,000 brands. Spearman correlation only. Neither bar shows that one thing causes the other. 0.664 0.218 web mentions backlinks A coefficient is a pattern across many brands, not a promise about yours.
The gap is real and it is only a correlation. It is a reason to shift effort, not a formula.

The plumbing list is short. Make sure your pages are crawlable and indexable, because nothing downstream happens otherwise. Decide deliberately which AI crawlers you allow, using the controls their operators publish rather than guesswork. Put your key facts in visible text on the page rather than only in markup. Then turn outward. In a study of 75,000 brands, brand web mentions showed a Spearman correlation of 0.664 with brand visibility in AI Overviews while backlinks came in at 0.218. That is a correlation across a large sample, not proof that mentions cause visibility, and it is emphatically not a reason to stop earning links. It is a reason to treat being talked about, in directories, roundups, professional profiles and genuine press, as at least as deserving of effort as link building, which is a rebalancing our SEO work and our content marketing now reflect. For how an SEO programme is actually put together, and what to ask before you buy one, see our guide to choosing an SEO agency in South Africa.

How do you measure AI search visibility honestly?

Start with the reporting that already exists, then be candid about what it cannot tell you. Google reports AI Overviews and AI Mode inside Search Console, within the overall search traffic in the Performance report. That is measurement you already own and did not have to buy.

What Search Console will not give you is a clean count of how often an AI answer mentioned you without a click, and no tool on the market can give you that with certainty either. Third-party AI visibility trackers work by running prompts and recording what comes back, which is sampling, not census, and the same prompt can return different answers to different people on different days. Treat those tools as a directional signal and never as a scoreboard. The two honest numbers to watch are the ones your own systems record: whether qualified enquiries are going up, and what people say when you ask them how they found you. KM Digital Solutions adds that question to enquiry forms for exactly this reason, because it catches the referral that analytics quietly loses. It is the cheapest instrument in the whole programme and the only one that reports in plain language.

Get an honest read on your AI search readiness

We will check what the crawlers can actually reach, what your pages say in visible text, and where your brand is mentioned across the web. No promises about citations, because nobody can make them.

Book a free strategy call

Which AI crawlers should you allow, and how?

Decide crawler by crawler, using each operator’s own documentation, because the settings are independent and they do different jobs. Blocking everything makes you invisible to AI answers. Allowing everything means your content also trains models. Those are separate choices and you are entitled to make them differently.

OpenAI documents two robots directives that control different things. OAI-SearchBot is the search crawler: in OpenAI’s words it surfaces websites in search results in ChatGPT’s search features, and sites opted out of it will not be shown in ChatGPT search answers. GPTBot is the training crawler, and the two settings are explicitly independent, so a site can allow search while disallowing training. A third agent, ChatGPT-User, fetches pages when a person asks for something specific; because it is user-initiated, robots rules may not apply, and OpenAI states it is not used to decide whether content appears in search. Google’s position is different again: it treats AI as built into Search, so Googlebot’s robots rules are the control, with nosnippet, data-nosnippet, max-snippet and noindex limiting what is shown, and Google-Extended covering training and grounding in its other systems. Allow about a day for a robots change to take effect on the OpenAI side.

One caution on that table below. These are the operators who publish their agents and honour a robots file. Publishing a rule is not a technical barrier, and it will not stop a crawler that ignores the convention, so treat this as a stated preference rather than a lock.

Agent What it is for What allowing it means
Googlebot Search, including AI Overviews and AI Mode Eligibility for classic results and AI features together. They are not separable
Google-Extended Training and grounding in Google’s other systems A separate decision from Search. Blocking it does not remove you from Search
OAI-SearchBot Search results in ChatGPT’s search features Opting out means you will not be shown in ChatGPT search answers
GPTBot Training foundation models Independent of the search setting. You can allow one and refuse the other
ChatGPT-User Fetches triggered by a person’s request Not used to decide search appearance, and robots rules may not apply

Frequently asked questions

Do I need to add special schema to appear in AI Overviews?

No. Google states there is no special schema.org structured data you need to add, and that you do not need to create new machine readable files or AI text files to appear in AI Overviews or AI Mode. Keep the schema you have for Google rich results and for making your entity unambiguous, and make sure it matches the visible text on the page, which Google does list as a recommendation. Just do not expect markup alone to earn a citation.

Is llms.txt worth publishing?

Not for Google. Google’s documentation states these files are not needed for Google Search and will not affect your visibility or rankings either positively or negatively. It also says it is fine to keep one if some other service you use reads it. So the honest answer is that publishing one is harmless and optimising one is a waste of an afternoon that your visible page copy would have repaid.

Are FAQ sections still worth writing?

Yes, for readers and for quotable text, but not for rich results. Google removed the FAQ rich result feature from 7 May 2026 and then removed its documentation, on the stated basis that the feature is no longer shown in search results. A well-written question and a direct, self-contained answer are still among the easiest things for any system to lift and quote. Write them for the person reading, and never let anyone report FAQ rich results to you as a deliverable.

Do backlinks still matter for AI visibility?

Yes. One large study found brand web mentions correlated more strongly with AI Overview visibility than backlinks did, at 0.664 against 0.218 across 75,000 brands. That is a correlation, not a cause, and a lower coefficient is not zero. The sensible reading is to stop treating links as the only off-site work worth doing, not to abandon them. Links and mentions are usually earned by the same activity anyway.

How long before AI search work shows results?

Nobody can give you a date, and the documentation explains why. Google says a page must be indexed and eligible to be shown with a snippet before it can be used at all, and that indexing and serving isn’t guaranteed even when everything is done correctly. On the OpenAI side, robots changes alone take roughly a day to register. Judge the work by whether your pages are reachable, readable and being mentioned, not by a promised week, and by who wrote it and what they can show, which is the subject of our guide to E-E-A-T for South African sites.

Google’s position on AI features, eligibility, recommendations and reporting is set out in AI features and your website; the FAQ rich result removal from 7 May 2026 and the llms.txt clarification are recorded in the Search Central changelog. The crawler descriptions are from OpenAI’s overview of OpenAI crawlers. The schema study is Ahrefs, We Tracked 1,885 Pages Adding Schema, and the Spearman correlation figures are from Ahrefs, AI Overview Brand Visibility Factors. Last reviewed and all sources accessed on 16 September 2026.

← Previous Next →
Start here

Start with a 30 minute call. No pitch deck, no pressure.

Pick whichever way suits you. Enquiries sent during business hours get a reply the same business day.

Book a 30 minute discovery call

Google Meet, Microsoft Teams or a normal phone call, whichever you prefer.

Tell us when suits you and we will set the call up personally.

What would you like to talk about?

Prefer to type instead? Send us your details or call +27 72 686 3338.

Speak to a founder directly

No call centre, no gatekeeper. Milan is the primary contact for new enquiries.

Tell us about your project

Straight to Milan and Kyle.

What do you need? *