We have written previously about why Generative Engine Optimisation matters for SA brands and why most SA websites are now invisible to AI search. This post is the practical companion to those. The actual mechanics of how a brand becomes a source that ChatGPT, Google AI Overviews, Perplexity, and Microsoft Copilot pull from.
If the concepts are new, start with the GEO post above. If you already understand the what and the why, this is the how.
TL;DR: AI citations are won by content that is structured for passage retrieval, entity-linked, present in the right brand-mention sources, and discoverable to LLM crawlers. Schema is necessary but not sufficient. llms.txt is now meaningful. Brand-mention seeding (Reddit, Quora, SA forums, Trustpilot) is the most underrated lever in the entire stack.
How AI Search Engines Actually Pick Citations
Different from traditional ranking, AI citation is a passage retrieval problem. When a user asks ChatGPT “what is a fair price for SEO in South Africa?”, the model does not read the entire web. It runs a retrieval step (usually a hybrid of vector search and keyword search) to pull the top 10-50 candidate passages from its index. It then synthesises an answer and decides which of those passages to cite in the response.
This means the unit of optimisation is no longer a page. It is a passage: a self-contained 50-200 word block that completely answers a discrete question.
Three properties make a passage citation-worthy:
- It is self-contained. A reader (or AI) can understand the answer without reading anything else on the page.
- It is unambiguously attributable. The passage clearly states facts in declarative form that the AI can quote without modification.
- It is from a source the AI considers authoritative. Authority is determined by a mix of traditional SEO signals (links, brand mentions, schema, EEAT) and AI-specific signals (presence in training data sources, llms.txt, citation patterns from other AIs).
Step 1: Restructure Content for Passage Retrieval
Pages designed for traditional SEO often lose AI citations because their answers are spread across paragraphs that require context. Pages designed for AI citation answer the question in the first sentence after the heading.
The structure that wins:
“`markdown
[The exact question a user would ask]
[2-4 sentence direct answer. Self-contained. Declarative. No “depending on”, no “it varies”.]
[Optional: 1-2 sentences of supporting context.]
[Optional: a short list, table, or example that supports the answer.] “`
Apply this pattern to every section heading on your highest-priority pages. The lift in AI citation is measurable within 30 days for pages that are otherwise authoritative.
Step 2: Entity SEO, Become a Recognised Entity
LLMs reason about the web in terms of entities (people, companies, concepts) rather than pages. Becoming a recognised entity raises your citation rate because the AI now knows what you are, not just that you exist.
The tactical steps:
- Organization schema everywhere with
sameAslinking to your Wikipedia, LinkedIn, Crunchbase, X/Twitter, YouTube, and any SA business registry (DUNS, CIPC if listed). - Wikipedia or Wikidata entry if your brand qualifies (notability requirements apply; most SA SMEs don’t yet).
- Consistent biographies for founders and key people across LinkedIn, your About page, conference bio pages, podcast appearances, with the same job title, company, and key facts. Inconsistent bios fragment the entity.
- Author schema on every blog post with
sameAsto the author’s LinkedIn and other public profiles. This carries human EEAT signal that LLMs are increasingly reading. - Topic clusters that hub on your core entity. If you want to be the entity AI cites for “digital marketing South Africa”, you need 20+ pieces of content reinforcing that association.
Step 3: llms.txt, The New Robots.txt for AI
llms.txt is an emerging convention (proposed late 2024, gaining adoption through 2025-26) that tells AI crawlers where to find your most important content in a clean, LLM-friendly format. While not yet officially supported by OpenAI or Anthropic, Perplexity, Mistral, and several other AI tools have begun reading it.
A basic llms.txt for an SA business looks like:
“`
South African digital marketing agency engineering measurable business results through SEO, paid advertising, and conversion-focused web design.
Core pages
- About GKnect Digital: Who we are, what we do, where we are based.
- Services: Full overview of services offered.
- Paid Advertising: Google Ads, Meta Ads, LinkedIn Ads management for SA businesses.
- Search Visibility: SEO, GEO, and local SEO services for SA businesses.
Knowledge hub
- Google Ads for SA Businesses: Comprehensive guide to Google Ads for SA SMEs.
- Local SEO Johannesburg: How to win the Map Pack in JHB.
- GEO Playbook: How AI search citations work.
Contact
- Contact Us: How to get in touch.
“`
Place this file at https://yourdomain.co.za/llms.txt and reference it from your robots.txt with LLMs-Sitemap: https://yourdomain.co.za/llms.txt. The format costs almost nothing to maintain and is already producing citation lift on Perplexity for early adopters.
Step 4: Schema Patterns That Get Extracted
Traditional schema (Article, Organization, BreadcrumbList) helps search rankings. The schema types that specifically help AI extraction are:
- FAQPage. Extracted verbatim by AI Overviews and ChatGPT search. Use for any page that answers discrete questions.
- HowTo. Extracted for procedural queries (“how do I set up Google Ads conversion tracking”).
- DefinedTerm / DefinedTermSet. Explicit definitions that AI loves to quote.
- Claim / ClaimReview. For fact-laden content, especially statistics.
- Product / Service / Offer. With
priceRange,areaServed(set to South Africa), andserviceTypepopulated. - LocalBusiness with
geo,priceRange, andaggregateRating. Extracted for local AI answers.
We cover the implementation details, including copy-paste JSON-LD for SA businesses, in our Schema Markup Cheatsheet.
Step 5: Brand-Mention Seeding (the Most Underrated Lever)
LLM training datasets weight Reddit, Quora, Wikipedia, Trustpilot, Hellopeter, and a handful of major media sites disproportionately because of their conversational, opinion-rich content. A single Reddit thread mentioning your brand favourably can outweigh dozens of backlinks for AI citation purposes.
The tactical brand-mention seeding programme:
- Identify the 20 queries you want to be cited for. Use ChatGPT, Perplexity, and Google AI Overviews yourself. Note which sources they currently cite.
- Find existing Reddit and Quora threads asking those questions. Answer them genuinely with your brand mentioned in context. Do not promote. Inform.
- Get reviewed on Hellopeter. SA-specific. Heavy weighting in SA-specific AI answers.
- Pitch SA industry publications (Bizcommunity, MarkLives, IT-Online, ITWeb, Moneyweb, BusinessTech) for guest articles. These domains carry significant authority for SA AI queries.
- Comment usefully on relevant LinkedIn posts. Mentions on LinkedIn flow into LLM training datasets.
- Submit founder or staff to podcasts in your industry. Transcripts are crawled.
Step 6: Measure AI Share-of-Voice
You cannot optimise what you cannot measure. AI share-of-voice tracking has matured significantly in 2025-26. Tools to consider:
- Profound (profound.so). Tracks citation share across ChatGPT, Perplexity, Gemini, Claude. Mid-market pricing.
- Otterly.ai. Similar capability, focused on prompt-set monitoring.
- BrightEdge AI Search. Enterprise tier.
- Manual prompt-set audit (free). Define 30 prompts your target audience would ask, run them weekly on ChatGPT, Perplexity, and Google AI Overviews, log which sources are cited. Cheap, slow, but works.
Run a baseline measurement before changing anything. Re-measure quarterly. Targets should be share-of-voice increases relative to named competitors, not absolute citation counts (which fluctuate with prompt randomness).
Step 7: AI Crawler Access, Do Not Block Them
A surprising number of SA sites are unintentionally blocking AI crawlers via robots.txt or Cloudflare bot management. The crawlers that matter and their user-agents:
- GPTBot (OpenAI training)
- OAI-SearchBot (ChatGPT search runtime)
- ClaudeBot (Anthropic training)
- Claude-Web (Claude runtime)
- PerplexityBot (Perplexity)
- Google-Extended (Gemini training)
- CCBot (Common Crawl, feeds many LLMs)
Audit your robots.txt and Cloudflare or WAF settings. If you have explicitly disallowed any of these and you want to be cited by AI, allow them. (If you have an active reason to opt out, such as proprietary content or paywalled content, leave the block, but understand the trade-off.)
Frequently Asked Questions
How long until I see citations in AI Overviews after publishing?
Indexation and citation appearance typically take 2-8 weeks after publishing high-quality, well-structured content. Brand-mention work takes longer to compound, usually 3-6 months for measurable share-of-voice lift.
Is llms.txt actually being used by ChatGPT and Google?
Not yet officially by OpenAI or Google. Perplexity and Mistral are reading it now. Implementation is cheap enough that the speculative ROI on adoption is high.
Do AI citations drive traffic, or just brand visibility?
Both, with the mix depending on the platform. ChatGPT search inline citations drive measurable click-through traffic (5-15% CTR on linked sources). Google AI Overviews reduce some downstream traffic but increase brand-impression value for the cited source.
Will following this playbook hurt my traditional SEO?
No. Every recommendation above also benefits traditional rankings. Structured content, entity SEO, schema, brand mentions, and crawler access are all rewarded by Google’s classic ranking systems as well.
How does this differ from traditional SEO?
Traditional SEO optimises for clicks from a ranked link. AI SEO optimises for inclusion as a source in a synthesised answer. The signals overlap but are not identical. Passage structure, entity recognition, and brand-mention diversity carry more weight in AI than in traditional ranking.
So Where Does That Leave You?
AI search citation is the new top-of-funnel for SA brands that want to remain visible as user behaviour shifts. The playbook above is structured work, not magic: passage-friendly content, entity SEO, schema, brand-mention seeding, llms.txt, and measurement. Done consistently for 6-12 months, brands move from invisible to cited.
If you want a baseline AI share-of-voice audit for your business, get in touch. We deliver AI search optimisation as part of our search visibility service.
