Why 98% of South African businesses are invisible to ChatGPT — and how to fix it

Ask ChatGPT to recommend the best accountant in Sandton. Ask Perplexity to find a physiotherapist in Cape Town. Ask Gemini which law firm handles SARS disputes in Johannesburg. In almost every case, you will not see most of the businesses that dominate those Google searches. They are simply invisible.

This is not a theoretical future problem. It is happening right now, today, to the vast majority of South African businesses. And the reason is not that AI is biased against small businesses — it is that most SA websites were built for a fundamentally different set of signals than AI engines require.

Here are the five core reasons — and exactly how to fix each one.

1. AI crawlers are not explicitly welcomed

AI engines like ChatGPT, Perplexity, Claude and Gemini use their own specialised crawlers to index content. GPTBot crawls for OpenAI. ClaudeBot crawls for Anthropic. PerplexityBot for Perplexity. These are different from Google's Googlebot — and they check your robots.txt file for permission before crawling.

The vast majority of South African websites have no instructions for these crawlers in their robots.txt file. They are not blocked — but they are not explicitly welcomed either. In practice, many AI crawlers are cautious: they skip sites that don't explicitly permit access.

The fix: Add explicit Allow directives to your robots.txt for all 14 major AI crawlers: GPTBot, ClaudeBot, PerplexityBot, Google-Extended, OAI-SearchBot, anthropic-ai, and eight others. This is a 5-minute change that costs nothing and can immediately improve your crawl coverage. Our Quick Wins Pack includes a correctly configured robots.txt as standard.

2. No Schema.org markup

Schema markup is structured data — invisible code in your website's HTML that tells AI engines exactly what your business is, what it does, who it serves, and where it is located. Without it, AI has to guess at the relationships between the words on your page. With it, your business becomes machine-readable: a clearly defined entity in the AI's understanding of the world.

In South Africa, fewer than 5% of SME websites have any Schema markup at all, according to our analysis of over 200 local business websites in 2026. The most impactful schema types for AI citation are: Organization (who you are), LocalBusiness (your sector and location), Service (what you offer), and FAQPage (questions your customers ask).

The fix: Implement JSON-LD schema in your website's HTML head. A correctly configured Organisation + LocalBusiness + Service schema can add 25–30 points to your AI Score and dramatically improve your visibility in Gemini and Bing Copilot, which weight structured data most heavily. This is included in the Quick Wins Pack.

3. No llms.txt file

llms.txt is an emerging standard — similar to robots.txt, but specifically for AI language models. It lives at your domain root (yourcompany.co.za/llms.txt) and tells AI engines your business name, what you do, who you serve, what your key service pages are, and how your content may be used.

Think of it as a structured introduction to your business, written specifically for AI. Most AI engines that support it will read llms.txt before crawling your site — meaning it shapes how they understand you before they've seen a single page. Perplexity and newer versions of ChatGPT are actively weighting this file.

The fix: Create a plain text llms.txt file and upload it to your domain root. It takes under 30 minutes to write and upload. The format is simple: a few headers describing your business, your core services, and your key pages. Our Quick Wins Pack generates and delivers this file ready to upload.

4. Content is not structured for AI citation

AI engines extract answers from content that is self-contained, answer-first, and structured around the exact questions customers ask. Most SA business websites are structured for humans — they use narrative prose, emotional storytelling, and sales language. None of these work for AI citation.

AI engines prefer content that reads like a reference document: a direct answer to a specific question, written in plain language, containing specific facts or numbers, and requiring no surrounding context to make sense. FAQ pages are the single most cited content type across all five major AI engines — because they are literally structured as questions and answers.

The fix: Build a FAQ hub on your website — a dedicated page with 20–30 questions your customers actually ask, each answered in 80–150 words of self-contained, answer-first text. Structure each answer so it can be extracted by an AI without the question for context. This is the highest-leverage content investment you can make for AI visibility in 2026. Our Foundation Launch includes a 25-question FAQ hub written in your brand voice.

5. Missing E-E-A-T authority signals

E-E-A-T stands for Experience, Expertise, Authoritativeness and Trustworthiness. It is Google's quality framework — but AI engines have adopted it as a core citation filter. Content that lacks clear authority signals is deprioritised for citation, even if it is technically well-structured.

For South African SMEs, the most commonly missing E-E-A-T signals are: author bylines with credentials, a populated About page with team information, a physical address or contact details, a privacy policy, and external links to authoritative sources. These signals tell AI engines that a real, credible business is behind the content — not a content farm.

The fix: Add an About page with real team information, add your physical address to your footer, include a privacy policy (required under POPIA anyway), and add author bylines to any articles or blog posts. None of these require technical expertise — they are content changes. Our Foundation Launch includes an E-E-A-T audit with prioritised recommendations.