A few months ago, I started receiving enquiries from people who told me that an AI assistant had recommended Nutmeg Studio. They'd asked ChatGPT or Claude something like "who's a good Shopify designer in South Africa?" and my studio came up in the answer.
If you run an eCommerce brand, this matters to you, because the same shift is happening in your category. More and more buying research now starts with a question typed into an AI assistant, and the brands that get named in those answers getting the customer's attention and click-through.
So how does it actually work? How do you make your business one of the ones that gets cited in AI search? This is a deep dive into what's going on, what helps, and what's just noise. I'll be honest about the parts nobody can promise, because there's a lot of hype and you deserve the grounded version.
First, the honest framing
There's a new acronym doing the rounds: GEO, or Generative Engine Optimisation, which is the practice of structuring your content so AI engines cite it when they answer questions. You'll also see it called AEO, LLMO, or "AI SEO." The industry hasn't settled on one name yet, but the name matters less than the idea behind it.
Here's the most reassuring thing I can tell you, and it comes straight from Google's own official guidance published in May 2026: optimising for AI search is, for the most part, still good SEO and good content. Google was unusually blunt about this. They explicitly told site owners they do not need special "AI text files," content broken into tiny AI-only chunks, AI-specific rewrites of their pages, or exotic schema markup to show up in AI answers. A whole cottage industry had sprung up selling those things, and Google essentially said: save your money.
That's worth holding onto, because the moment something feels mysterious and technical, someone will try to sell you a shortcut. The reality is calmer and more durable than that.
How an AI actually decides who to cite
To get cited, it helps to understand what's happening under the hood. There are really two separate mechanisms, and they reward slightly different things.
When the AI answers from memory. Large language models are trained on an enormous section of the public web. When a model recommends a business "from memory," it's drawing on patterns it absorbed during training: which businesses showed up, repeatedly and across many independent sources, alongside a particular topic. For a model to "know" you as a Shopify expert in South Africa, your name needs to appear in that context again and again, in places it has read: your own site, client sites that credit you, directories, testimonials, articles, your partner listings. No single mention does it. The repetition across independent sources is what builds the association.
When the AI searches live. Increasingly, AI assistants like Claude and ChatGPT don't rely only on memory, they also search the web in real time and synthesise an answer from what they find. This is where things get interesting, and where you have more direct influence over the output. One detail worth knowing: these systems often break your question into several smaller searches rather than searching your exact words. Someone asking "best Shopify designer for skincare brands in South Africa" might trigger separate searches for "Shopify designer South Africa," "Shopify skincare brands," and "conversion-focused Shopify expert." The AI then assembles an answer from whatever surfaces across all of those. A page that cleanly answers several of those sub-questions gets pulled into a whole cluster of related queries, not just one.
The takeaway: live AI search is grounded in ordinary search results. If you don't rank for a topic in normal search, you're unlikely to be cited by an AI answering about it. The two are linked, not separate.
What genuinely helps (the durable list)
Here's where I'll be specific. These are the things that move the needle, rather than hyped-up trends. Notice how many of them are simply good business practices that happen to feed the AI systems too.
1. Be unmistakably specific about what you do and who you serve. Vagueness is your enemy. "We build websites" tells an AI nothing useful. "Shopify design and development for beauty and wellness brands in South Africa" gives it a clear, citable association. The more precisely you define your niche, the more confidently a model can recommend you for it. This is positioning work first and SEO work second, but it pays off in both.
2. Earn third-party corroboration. This is the single biggest multiplier, and the one most worth your energy. AI systems weight independent agreement heavily. Other credible sites confirming who you are matters far more than your own site claiming it. Research on AI citations has found a clear bias toward earned media (mentions on sites you don't own) over brand-owned content. What this means for you: get listed in relevant directories, pitch a guest article to an industry publication, appear on a podcast, and earn a press feature or mention. Every independent source that ties your name to your niche compounds your credibility in the eyes of both search engines and AI.
3. Publish content that answers the actual questions people ask. Think about what your customers actually type into an AI: "How much does a Shopify store cost?" "Should I move from WooCommerce to Shopify?" "Is Shopify good for a small skincare brand?" Each of those is a chance to publish a genuinely useful answer that an AI can draw from and attribute to your business. A practical trick: type your target question into ChatGPT or Claude yourself, look at the shape of the answer it gives - is it a definition, a comparison, a set of steps - and then write a page or blog post that delivers that shape, but more thoroughly and more honestly than whatever it's currently pulling from.
4. Lead with first-hand experience, not commodity content. This is the thing Google singled out as the highest-leverage move of all. They drew a sharp line between generic, anyone-could-write-it content ("7 tips for choosing a web designer") and content built on real, lived expertise and storytelling ("what I learned migrating a 200-product candle brand from WordPress to Shopify without losing their SEO"). The second kind is far more likely to be surfaced and cited, because it contains something the AI can't find in a hundred other places.
5. Get the technical foundation right. None of the above works if an AI can't read your site. This is the boring-but-essential technical layer: webpages that are crawlable and indexed, clean semantic structure, fast load times, clear headings, a named author with a real bio, and visible publish and update dates. One genuinely important and often-missed point: some sites accidentally block AI crawlers without realising it, sometimes through a security setting like Cloudflare that changed its defaults. It's worth checking that the assistants are actually allowed to visit your site, because if they're blocked, none of your other effort will work.
6. Keep your content fresh. Recently updated content appears markedly more often in AI answers. One analysis found that updated webpages were cited several times more frequently than stale ones. AI citations also expire fast; a large share of what gets cited is only a few months old. A blog you tend to, rather than abandon, is a blog that keeps earning citations.
What to ignore (so you don't waste money)
Because there's so much noise, here's what Google explicitly said you can skip, at least for its own AI features:
You don't need a special llms.txt file. You don't need to "chunk" your content into tiny AI-only fragments. You don't need AI-specific rewrites of pages you've already written well for humans. And inauthentic mention-building — paying for fake reviews or spammy mentions — doesn't fool these systems; the same spam filters that protect search already block it.
One honest caveat: Google's guidance speaks for Google's own AI. Other systems like ChatGPT, Claude and Perplexity may weight some signals differently, and the picture is still evolving. But the foundations - expertise, clarity, third-party credibility and technical foundations - help everywhere. When in doubt, invest in those rather than in tactics that only make sense for one platform.
The thing I most want you to take away
I'll be straight with you, because it's the most useful thing in this whole piece: you cannot game your way into being recommended by AI, and you shouldn't try.
The brands that get cited are, overwhelmingly, the ones that built a genuine reputation that the web happened to record. The AI is reflecting that reputation back — it isn't inventing it. When my studio gets recommended, it's not because I cracked a secret formula; it's because years of real client work, public credit, verified partner status, and useful writing about my niche left a consistent trail across the internet. The AI simply read that trail.
So the strategy that works is almost reassuringly old-fashioned. Do excellent work. Launch excellent products. Make sure it gets credited publicly. Be crystal clear about who you help and how. Write honestly and usefully about the things your customers are trying to figure out. Keep your website in technical order. Then let the citations arrive as a byproduct of an earned reputation, which is a far sturdier position than anything built on tricks that the next algorithm update could erase.
That's the whole game, and the good news is it's a game worth playing whether or not the AI is watching.
Nutmeg Studio is a Verified Shopify Partner building conversion-focused stores for product brands across South Africa and beyond. If you're thinking about your store's next chapter — a rebuild, a migration, or simply getting found — book a complimentary clarity call.
