Generative engine optimisation (GEO): a guide for South African brands

Generative engine optimisation (GEO) is the work of making your brand more likely to be named, recommended and cited when people ask AI assistants such as ChatGPT, Gemini and Claude, and when Google shows an AI Overview above its results. SEO earns a click from a list of links. GEO earns a mention inside the answer itself.

In short

• An AI answer names a handful of brands, not ten links. Being one of them is the new first page.
• Assistants name brands from two places: what the model learned in training, and the pages it finds when it searches the web.
• In the research that coined the term, adding quotations, statistics and source citations to a page raised its visibility in AI answers by up to 40%. Keyword stuffing did not help.
• You cannot improve what you do not measure: put the questions your buyers ask to several engines, more than once, and count.

Why GEO matters in South Africa now

On 11 October 2026, in our study of 96 everyday South African searches across 12 sectors, Google showed an AI Overview on 92 of them: see the state of AI Overviews in South Africa. In a separate, smaller check the same day, we ran 12 marketing searches on Google South Africa, and 10 of them showed an AI Overview above the ordinary results. For searches such as "GEO South Africa" and "AI SEO South Africa", the top ordinary results were mostly South African SEO agencies' own sites: few local pages compete to answer these questions.

Search demand for the term itself is still small here. Google's keyword data shows no measurable monthly volume in South Africa for most GEO terms. That is the opportunity: few South African brands are competing to be the source yet.

How AI assistants decide which brands to name

From what the model learned

A model's training data is a snapshot of the web and other text up to a cut-off date. Brands that were widely and consistently described before that date are easier for the model to recall. That memory changes only when a new model version is released.

From what it finds when it searches

When an assistant searches the web, as ChatGPT search, Perplexity and Google's AI Overviews do, it reads a few pages and cites them. Ordinary search signals still count here: in 9 of the 10 AI Overviews in our 12-search check, at least one cited site also ranked in the top five ordinary results for the same search.

The two routes explain a common surprise: one assistant names a brand confidently while another misses it entirely, because one answered from memory and the other from a search.

What the research says

The term comes from GEO: Generative Engine Optimization, a paper by Pranjal Aggarwal and colleagues published at the KDD 2024 conference. The authors rewrote web pages in different ways and measured how much of each AI answer drew on them, weighted by where in the answer it appeared.

Change made to the page	Visibility score (unchanged page: 19.5)
Adding quotations from relevant sources	27.8
Adding statistics	25.9
Making the text more fluent	25.1
Citing sources	24.9
Keyword stuffing	17.8 (worse than doing nothing)

Source: Aggarwal et al., Table 1, position-adjusted word count on their GEO-bench queries. The authors note that results varied by subject area, so treat these as directions, not guarantees.

How to measure GEO

• Start from buyer questions, not keywords. "Which bank has the lowest fees for a small business?" is the kind of question where a brand has to earn its place.
• Ask every question on more than one engine, more than once. Answers vary between models and between runs of the same model.
• Count four things: whether you are named at all, whether you are named first, your share of all brand mentions (AI share of voice), and which sources the answer cites.
• Separate branded from unbranded questions. Being named when someone asks about you by name proves little; unbranded questions are where share is won.
• Do not read a percentage from a small sample. Below 30 answers, a single answer moves the figure by more than three points.

What South African brands can do

• Answer the question on the page. Put a one-sentence definition or answer first, then specifics: comparison tables, real numbers with dates, and a short FAQ.
• Make the page readable without JavaScript. Many AI crawlers fetch a page's HTML and do not run its scripts, so a site that builds its content in the browser can look empty to them. We found and fixed exactly this on our own site.
• Decide which AI crawlers may read your site. robots.txt controls OpenAI's GPTBot and OAI-SearchBot, Anthropic's crawlers and PerplexityBot; blocking a search crawler can keep you out of that assistant's live answers. Google's AI Overviews use the normal Googlebot. The separate Google-Extended setting governs use of your content for Gemini, not whether you appear in Google Search.
• Be present where the answers come from. Find which publications, directories, review sites and marketplaces your category's answers already cite, before you pitch anyone.
• Keep your facts consistent. Your name, what you sell and where you operate should read the same on your site, social profiles, Google Business Profile and directory listings, so models do not confuse you with a similarly named business.
• Measure before and after every change, on the same questions, so you can tell what worked.

For how this differs from the SEO work your team already does, read SEO vs GEO.