Your customers are asking ChatGPT, Perplexity, Copilot and Google's AI Overviews for recommendations. Most South African business websites cannot be read by those tools at all.
Search used to mean ten blue links and a scroll. Increasingly it means a single written answer, assembled by an AI tool that has already decided which businesses are worth mentioning. If your website cannot be read, understood and attributed by those systems, you are not in the running — however well you rank in ordinary search.
This is new, it is moving quickly, and very few South African SMEs have addressed it. That is exactly why it is worth doing now rather than in two years.
AEO is about being the source a search engine lifts a direct answer from — the featured snippet at the top of Google, the People Also Ask boxes, the AI Overview, the response a voice assistant reads aloud. It rewards content that answers a specific question clearly and immediately, marked up so the engine can recognise it as an answer. Most business websites describe services. Very few answer questions in the form people actually ask them.
GEO is about being retrieved and cited by ChatGPT, Claude, Perplexity, Gemini and Copilot when someone asks for a recommendation. These systems do not rank pages the way Google does. They retrieve information, judge what is credible, and write an answer that may name a small handful of businesses. Being named requires that your site is machine-readable, that your identity is unambiguous, and that other sources on the web corroborate what you say about yourself.
The most common cause is JavaScript. Many websites build their navigation, contact details and sometimes their body content with scripts that run in the visitor's browser. Google renders JavaScript, so the site looks fine in ordinary search results. But most AI crawlers — GPTBot, ClaudeBot, PerplexityBot, CCBot — fetch the raw HTML and do not execute scripts. They see a page with no navigation, no contact block, and no links to your other services.
The second cause is missing structured data. Without it there is nothing stating, in machine-readable form, that you are a business, of this type, in this town, offering these services. The third is entity ambiguity: when your business name is shared with a larger or better-documented organisation, AI tools will confidently describe the other one instead.
We fetch your site the way an AI crawler does rather than the way a browser does, and compare the two. We check what structured data exists and whether it validates. We test your business identity against the major assistants and record what they currently say. We look at whether your content answers real questions or only describes services. And we check what the rest of the web says about you, because these systems weight third-party corroboration far above self-description.
Usually: moving critical content into raw HTML, adding and validating structured data, building genuine FAQ content and marking it up, tightening your entity signals so the business is unambiguous, setting a deliberate robots.txt policy for AI crawlers, publishing an llms.txt summary, and beginning the slower work of third-party citations. Then we re-test, because none of it counts unless the answer actually changes.
| Signal | Why it matters | What we do |
|---|---|---|
| Raw-HTML content | Most AI crawlers do not run JavaScript | Move navigation, contact details and body copy into the served HTML |
| Structured data | States what you are in machine-readable form | Organization, Service, FAQPage and BreadcrumbList markup, validated |
| Entity clarity | Shared names get attributed to the bigger organisation | Consistent naming, location anchoring, sameAs links to real profiles |
| Answer-shaped content | Engines match question headings to question queries | FAQs and question-format headings drawn from what clients actually ask |
| Third-party corroboration | These systems trust others far above your own claims | Directory listings, profiles, reviews, consistent business details |
| Crawler access policy | A copied robots.txt can silently exclude you | Deliberate, explicit rules for GPTBot, ClaudeBot, PerplexityBot and CCBot |
We record what the major AI assistants say about your business today, and fetch your site as a non-JavaScript crawler sees it. This is the before picture, and it is often uncomfortable reading.
We identify what is blocking retrieval and attribution: content locked behind scripts, missing structured data, entity confusion, or simply nothing on the web corroborating you.
We implement the technical changes and build the answer content, working alongside any SEO or website work rather than duplicating it.
We ask the same questions again and report what changed. Some of this moves within weeks; corroboration takes longer.
Nobody can guarantee that an AI assistant will recommend your business. Anyone who tells you otherwise is selling something they cannot deliver. These systems are opaque, they change without notice, and there is no equivalent of a rank tracker that reliably measures citation.
What we can do is remove the reasons you are currently excluded, make your business unambiguous and machine-readable, and measure the change in what the assistants actually say. That is a real, checkable outcome — and for most South African SMEs it is the difference between being invisible and being eligible.
We also will not recommend this work in isolation. If your website loses enquiries, or you have no reliable way of following up the ones you get, that comes first. Being recommended by an AI assistant is worth nothing if the enquiry then disappears.
The AI Visibility Check is R1 500 including VAT. That covers the baseline test across the major assistants, fetching your site the way an AI crawler sees it, a structured data and entity check, and a short written finding listing the specific blockers in priority order. If you go ahead with a Foundation Assessment within 30 days, the R1 500 is credited in full against it. Larger implementation work is quoted after the assessment, because scoping it before we understand the business would be guesswork.
There is overlap, and sound technical SEO helps. But the two diverge in ways that matter. Traditional SEO optimises for ranking a page within a list. AEO and GEO optimise for being extracted and cited inside an answer, which depends far more on machine-readable structure, unambiguous identity, and what other sources say about you. A site can rank respectably in Google and still be completely invisible to ChatGPT.
Ask the assistants directly. Ask ChatGPT, Perplexity or Copilot what your business does and who it serves. If the answer is wrong, generic, about a different company with a similar name, or simply that it has no information, you have your answer. That test takes two minutes and it is where we begin every engagement.
That is a legitimate strategic choice, and it is worth making deliberately rather than by accident. If your value lies in unique published content you would rather not see reproduced, blocking may be right. If you are a service business that wants to be recommended, blocking removes you from consideration entirely. Most SMEs want to be found. The mistake is copying a restrictive robots.txt template without knowing what it excludes.
The technical fixes take effect as soon as the crawlers next visit, which is typically weeks rather than months. Entity recognition and third-party corroboration are slower, because they depend on other websites and profiles catching up. We re-test at intervals and show you what actually moved, rather than reporting on activity.
Yes. Stackwise commissioned a full technical, AEO and GEO audit of its own site, found genuine problems, and fixed them. This site now serves its navigation in raw HTML, carries structured data on every page, marks up its FAQs, sets an explicit AI crawler policy and publishes an llms.txt file. We would not sell this work if we had not done it on ourselves first.