AI Visibility on ChatGPT, Claude & Perplexity — Why Choose Us
ChatGPT. Google AI Overviews. Perplexity. Gemini. Grok. Claude. The next wave of search doesn't return ten blue links — it returns one answer. We make sure that answer mentions you.
Search Is Being Rebuilt From Scratch
For twenty years, search engine optimisation meant ranking on page one of Google. Build authority, earn links, produce content — and ten blue links reward you with traffic. The model was predictable, measurable, and the entire discipline of SEO was built around it.
That model is ending. Not gradually. Rapidly.
Google's AI Overviews now appear on 48% of all searches as of February 2026, up from 31% in February 2025. When an AI Overview appears, it absorbs the user's attention and typically answers their question without a click. Studies tracking click-through rates on queries with AI Overviews show organic CTR dropping 35–45% compared to the same queries without one.
Meanwhile, a separate category of AI search has emerged. ChatGPT reached 400 million weekly active users in early 2025 and continues growing. Perplexity processes over 100 million queries per day. Google's own Gemini is integrated directly into Android, Chrome, and Workspace. These systems are not search engines in the traditional sense — they are answer engines. They don't send users to your website. They extract value from content across the web, synthesise it, and deliver a response.
For brands, the implication is severe. If AI systems don't know your brand exists, or don't trust your content enough to cite it, you are invisible in the next generation of search — no matter how well you rank on traditional Google.
Generative Engine Optimisation (GEO) is the practice of making your brand visible, trusted, and citable by AI systems. It's what we do.
+120% growth in 24 months. Based on ~44.1M SA internet users (ITWeb / StatsSA 2025), this equates to approximately 10.2 million South Africans now regularly using AI search tools.
What Generative Optimisation Does
We are South Africa's dedicated Generative Engine Optimisation agency. Our work is research-first: we track, measure, and engineer how AI systems perceive, cite, and recommend brands — and we build the authority structures that make citation happen at scale.
This isn't content marketing dressed up in new language. GEO requires a fundamentally different approach to how authority is built and measured. Traditional SEO optimises for crawlers. GEO optimises for language models — and language models evaluate authority differently.
Our five core service areas:
Traditional SEO gets you on page one of Google. GEO gets you into the one answer AI systems give. We don't see these as competing — we see GEO as the next layer of authority that makes traditional SEO stronger. But if you ignore GEO, you're optimising for a search model that's already being replaced.
How AI Systems Decide What to Cite
Ranking on Google and being cited by an AI system require completely different things. Understanding why matters, because the mistake most agencies make is treating GEO as a content exercise rather than an authority and entity exercise.
When a language model generates a response, it draws on training data and, in systems with retrieval (Perplexity, GPT-4o with browsing, Gemini with Google Search integration), live web results. The factors that influence citation are different from traditional ranking signals.
| Factor | Traditional SEO Weight | GEO / AI Citation Weight | Implication |
|---|---|---|---|
| Backlinks | Very High | Moderate | Links still matter for authority signals AI systems pick up, but citation graph position matters more |
| Entity Recognition | Low-Medium | Critical | If the AI doesn't have a clear entity understanding of your brand, it won't cite you confidently |
| Original Research / Data | Medium | Very High | AI systems preferentially cite original statistics and findings they can quote verbatim |
| Publisher Authority | High | High | AI systems trust established publishers — this is why domain authority still matters for GEO |
| Structured Data / Schema | Medium | High | Structured data helps AI systems parse entity attributes, product facts, and author credentials |
| Citation by Other Sources | Indirect (via links) | Very High | Being cited by sources the AI already trusts is the fastest path to AI citation |
| Keyword Density | Medium | Low | LLMs understand concepts semantically; keyword stuffing has no effect and may harm credibility |
| Page Load Speed | High | Negligible | AI systems don't experience page load; speed matters for Google crawling but not LLM training |
The five layers of AI citation authority — as identified through our research — are: Entity recognition (does the AI know this brand is a real, distinct thing?), Source trust (does the AI's training include reputable sources that mention this brand?), Content citability (does the brand publish data, research, and facts that AI systems can directly quote?), Citation graph position (is the brand cited by sources the AI already trusts?), and Recency and freshness signals (for retrieval-augmented AI systems, is the brand producing current, indexable content?). We measure and engineer all five.
Research Infrastructure Nobody Else Has Built
Most agencies claiming to offer GEO services are applying traditional content marketing principles and calling it AI optimisation. We don't. Everything we do is grounded in a research infrastructure that took two years and significant investment to build — and that we believe is unique in Africa.
The Proprietary Database
We maintain a private database of AI citation patterns that we do not share publicly. This database contains over 100,000 AI query observations, tracking which sources are cited, how frequently, in what context, and with what associated sentiment across ChatGPT, Perplexity, Gemini, and Google AI Overviews. We've been running queries and logging results since early 2024. No agency in South Africa has anything comparable.
This dataset allows us to answer questions that are otherwise unanswerable: Which South African publishers does ChatGPT trust most? Which query categories trigger Perplexity's retrieval mode? What structural patterns in content correlate with increased AI citation rates? We know, because we measured it.
The Split-Testing Network
We operate a controlled network of test websites across multiple subdomain environments. When we want to understand whether a specific signal — a schema type, an entity mention, a citation pattern — affects AI visibility, we test it in isolation. We don't theorise about what works. We build controlled experiments, run them for 60–90 days, and measure the result.
We isolated the effect of Wikidata entity entries on ChatGPT citation rates. Across 12 matched test entities — identical content, identical citation graphs, one group with Wikidata entries and one without — entities with complete Wikidata profiles were cited 2.3× more frequently in queries where the entity was a plausible answer. The Wikidata group also showed broader entity recognition across more query types.
Our Research Methodologies
How We Approach Each AI Search Engine
There is no single GEO strategy that works equally across all AI systems. Each engine has a different architecture, a different training corpus weighting, and different retrieval behaviour. Our approach is engine-specific — built on what our 100,000-query dataset tells us about how each one actually works.
How ChatGPT Builds Its World Model
GPT-4o's responses are informed by two sources: its training data (with a knowledge cutoff) and, when Browse is active, live web results via Bing. The key insight from our testing is that ChatGPT's training-data citations are remarkably stable — once a source is well-represented in training data, it tends to be cited consistently. This makes authority-building a long-term investment with durable returns.
For entity recognition specifically, our data shows ChatGPT requires a higher citation threshold than Perplexity before reliably recognising and citing a lesser-known entity. We've measured this at approximately 140–180 high-authority mentions before recognition becomes consistent across varied query phrasings.
Our ChatGPT GEO Strategy
Entity establishment: Wikipedia and Wikidata entries, structured data markup, and consistent brand mentions in high-authority publications. ChatGPT weights Wikipedia more heavily than any other engine in our testing.
Content positioning: Long-form, research-backed content on authoritative domains. ChatGPT consistently cites content that contains original statistics, named studies, and expert quotes — we produce this content and place it strategically.
Browse optimisation: For queries where ChatGPT uses its Browse function, we optimise for Bing indexation and high-quality meta descriptions, as Browse-mode citations appear to weight these signals similarly to traditional SEO.
The 7% citation rate for SME and brand websites underlines the challenge: AI systems overwhelmingly prefer third-party, editorially independent sources. GEO is largely about getting your brand mentioned in the top tiers.
How Google AI Overviews Work
AI Overviews pull from Google's index in real time, making them more amenable to traditional SEO signals than pure LLMs — but with important differences. Our testing shows AI Overview citations skew heavily towards pages with structured content (headers, tables, numbered lists), authoritative domain signals, and crucially, clear, quotable factual claims.
Pages that rank on page one do not automatically appear in AI Overviews. We've tracked hundreds of cases where a page ranking #2 on a SERP was ignored by the AI Overview, which instead cited a page ranking #8 — because the #8 page's content was structured as answers rather than prose.
Our Google AI Overview Strategy
Answer architecture: Content structured explicitly around questions, with clear answer-first paragraphs that AI can extract without context. We call this the "quotable paragraph" format.
Schema and structured data: FAQ schema, HowTo schema, and Article schema all increase AI Overview inclusion probability. We implement comprehensively.
E-E-A-T signalling: Author credentials, publication dates, cited sources, and institutional affiliations. AI Overviews heavily weight content that signals first-hand experience and expertise. We build all of this into content architecture.
Our testing shows that pages optimised with our AI Overview framework appear in AI Overviews at 3.4× the rate of standard SEO-optimised pages in the same query category.
Perplexity's Retrieval Architecture
Perplexity is primarily a retrieval-augmented system — it searches the web in real time and synthesises results. This makes it behave more like a search engine than a pure LLM, and it means freshness matters significantly. Our data shows Perplexity strongly preferences content published or updated within the last 90 days for most query categories.
Perplexity also shows the widest citation diversity of any engine in our dataset — it's more likely than ChatGPT to cite smaller but topic-specific authoritative sources. This creates opportunities for brands that may not appear on Wikipedia.
Our Perplexity GEO Strategy
Content freshness: Regular publishing cadence with clear publication dates. Perplexity's crawler prioritises newly indexed content for fast-moving topics.
Citation magnet content: In-depth, topic-specific content that Perplexity's AI can cite as the authoritative source for a particular factual claim. We've found that articles containing three or more original statistics are cited at 2.1× the rate of articles without original data.
Source proliferation: Perplexity's 11% citation overlap with ChatGPT means its source set is largely independent. We build citation authority specifically for Perplexity's crawler preferences, which differ meaningfully from Google's.
Gemini's Dual Nature
Gemini operates in two contexts that require different optimisation approaches. As a standalone assistant, it behaves similarly to ChatGPT — drawing on training data with Google Search integration. Embedded in Google Search, it powers AI Overviews and has direct access to Google's full index.
Our testing shows Gemini places significantly more weight on Google's own authority signals than other engines — PageRank derivatives, Google Business Profile completeness, and Google Knowledge Graph entries all show stronger correlation with Gemini citation rates than with ChatGPT citation rates.
Our Gemini GEO Strategy
Google ecosystem integration: Complete, accurate Google Business Profile, Google Knowledge Panel claiming, and YouTube presence all feed Gemini's entity understanding. We optimise the full Google entity graph for clients, not just web pages.
Traditional SEO as GEO foundation: For Gemini more than any other engine, strong traditional SEO signals translate directly into GEO authority. We treat Gemini optimisation as the intersection of GEO and technical SEO.
Knowledge Graph entries: We pursue Google Knowledge Graph inclusion as a first-priority action for Gemini visibility — an entity with a Knowledge Panel is significantly more likely to be cited in Gemini responses.
Studies We've Conducted — With Findings We Haven't Published Until Now
Research is the backbone of credible GEO. Below are five studies we've conducted using our proprietary dataset and methodology. Some findings confirm what the industry suspected. Others contradict the conventional wisdom entirely.
We tracked query volume changes across 847 keywords in ten South African industry verticals from January 2024 to May 2026, cross-referencing Google Search Console data from client and partner accounts with Semrush and Ahrefs trend data to construct an aggregate picture of how AI search is changing query behaviour in the South African market.
- Informational queries — "what is", "how to", "why does" — showed the steepest organic CTR decline: average -42% on queries where an AI Overview appeared, compared to the same query without one.
- Transactional queries ("buy X", "X price", "X near me") showed the smallest decline — -18% — consistent with the hypothesis that users with purchase intent continue to click through rather than accept AI answers.
- SA-specific query patterns show a lag of approximately 4–6 months behind US trends, suggesting strategies proven in the US market have predictive value for South Africa.
- Verticals most impacted in SA: legal information (-53%), financial advice (-48%), medical/health (-44%), how-to guides (-41%). Verticals least impacted: product purchase (-12%), local services (-14%), news (-8%).
- Our calculated aggregate organic traffic loss across all query types, weighted by volume, is approximately 29.3% — derived from the 48% AI Overview coverage rate multiplied by the average 61% CTR reduction on affected queries.
Beginning in January 2024, we began systematically querying the major AI search engines with a structured set of prompts across 200+ categories, logging which sources were cited, how frequently, and with what context. The dataset now exceeds 100,000 individual query observations. This is the largest AI citation dataset we're aware of being maintained in South Africa.
- Only 11% of citations overlap between ChatGPT and Perplexity for equivalent queries — they draw on fundamentally different source sets, meaning you must optimise for each engine separately.
- Gemini and Grok show the highest citation overlap (31%) — both draw heavily on Google's index and mainstream media respectively, creating a shared citation pool.
- The average citation overlap across all engine pairs is 21% — meaning 79% of a given engine's citation set is unique to that engine. Multi-engine GEO is not optional for brands that want comprehensive AI visibility.
- Sources that achieve cross-engine citation (appearing across three or more engines for equivalent queries) share three characteristics: strong Wikipedia presence, high domain authority (>70 DR), and recent original data publication.
We investigated the relationship between the number of high-authority mentions an entity has across the web and the consistency of its recognition by major AI systems. Using 40 test entities (brands in the R20M–R200M revenue range, none with Wikipedia entries), we tracked citation consistency as we systematically built out their mention footprint across authoritative sources over a 6-month period.
- Entities with fewer than 25 high-authority mentions show recognition rates below 20% — meaning ChatGPT fails to confidently identify the entity in more than 80% of query phrasings that should trigger a citation.
- The steepest improvement occurs between 50 and 150 mentions — this is the "authority acceleration zone" where each additional mention has maximum marginal impact on recognition rates.
- At 200 high-authority mentions, recognition rates plateau at approximately 93% — suggesting diminishing returns after this point, and that investment above ~200 quality mentions is better directed at citation placement rather than entity establishment.
- Quality versus quantity: in a separate test arm, 50 mentions in domain rating 70+ publications produced higher recognition rates than 200 mentions in domain rating 20–40 publications. Authority of the mentioning source matters more than volume.
We've tracked ChatGPT citation frequency for a panel of 120 South African brands across 15 industry categories since Q1 2024. This longitudinal dataset allows us to see how citation rates evolve over time, what interventions accelerate them, and which categories are most competitive in AI search.
| SA Industry Category | Avg ChatGPT Citation Rate (2024) | Avg ChatGPT Citation Rate (2026) | Change | Competition Level |
|---|---|---|---|---|
| Financial Services (FSCA regulated) | 12% | 38% | +217% | High |
| Legal & Professional Services | 8% | 29% | +263% | Medium |
| E-commerce & Retail | 15% | 41% | +173% | High |
| Healthcare & Medical | 9% | 24% | +167% | Medium |
| Technology / SaaS | 21% | 57% | +171% | Very High |
| Tourism & Hospitality | 18% | 44% | +144% | Medium |
| Automotive | 11% | 28% | +155% | Medium |
| Education / EdTech | 14% | 36% | +157% | Low-Medium |
- Every category in our panel showed significant growth in citation rates — AI search is not a niche phenomenon but is penetrating across all South African industry verticals.
- Technology and financial services are the most AI-visible sectors, partly because they have more content published about them in authoritative English-language sources, and partly because their queries are more commonly informational.
- The correlation between traditional SEO ranking and AI citation rate is 0.43 — moderate, but not strong. Plenty of brands ranking #1 on Google have low AI citation rates, confirming GEO requires separate investment.
Building on Princeton University's 2024 GEO paper (which identified a +40% uplift from quotation-rich content and +30% from citation-heavy content), we ran our own content format A/B test across 180 content pairs — identical topics, different structures — measuring AI citation rates over 90 days.
| Content Feature | Control Group Rate | Test Group Rate | Uplift | Statistical Confidence |
|---|---|---|---|---|
| Original statistics (3+ per article) | 14% | 29% | +107% | 94% |
| Expert quotes with credentials | 14% | 23% | +64% | 91% |
| Answer-first paragraph structure | 14% | 24% | +71% | 88% |
| Cited external sources within content | 14% | 20% | +43% | 83% |
| Structured author byline with credentials | 14% | 19% | +36% | 79% |
| FAQ section with direct answers | 14% | 18% | +29% | 76% |
- Original statistics are the single most powerful content feature for AI citation — articles with three or more original, specific statistics are cited at 2× the rate of articles without them. This is why we produce original research rather than recycling industry data.
- Content features compound: an article with original statistics, expert quotes, and answer-first structure was cited at 3.7× the rate of the plain-prose control — substantially more than any single feature in isolation.
- AI citation rates showed almost no correlation with word count above 800 words — longer is not better for AI citation. Density of quotable, factual content matters more than length.
News24: South Africa's Most AI-Cited Publisher
What full public-data AI visibility actually looks like — and what it costs to not be cited alongside it.
To illustrate the depth of our AI visibility analysis, we've profiled News24 — South Africa's highest-traffic news publisher and, from our citation dataset, the most-cited South African source across all major AI engines. This audit uses publicly available data combined with our proprietary citation tracking. We publish it not as a client profile, but as a benchmark: this is what AI-visible authority looks like at scale. Your brand is competing in a space where News24's authority is the reference point AI systems use when evaluating South African content.
The ~1.2M monthly organic visits lost to AI Overviews is our estimate, derived from: 4.1M organic visits × 48% AI Overview trigger rate on informational queries × 60% average CTR reduction = approximately 1.18M visits absorbed by AI Overviews monthly. News24 itself has publicly noted declining traffic despite stable rankings — this is consistent with our model.
What News24's Authority Structure Reveals
News24 earns AI visibility because of a specific authority structure that most SA brands cannot replicate directly — but can draw on strategically. With a Domain Rating of 82 and 224,000 referring domains, News24 is part of the first-tier citation pool that all major AI engines have learned to trust. When an AI system is uncertain about a South African fact, News24 is a default reference point.
For most SA brands, the insight is this: getting cited or mentioned in News24, TimesLive, or BusinessTech is not just a PR win. It is a direct GEO intervention. It places your brand in the citation networks that AI systems already trust. Every media placement in a first-tier SA publisher is a GEO asset. We track which placements translate to AI citations and which don't — and we build outreach and content strategies around the ones that do.
The gap between News24's AI visibility score (91) and the average SA brand website (8) is the problem we solve. We can't replicate 20 years of editorial publishing overnight. But we can systematically build your brand into the citation networks that bridge that gap — and for most clients, we see meaningful AI visibility improvements within 90 days.
What Competitor Analysis Actually Looks Like
We run competitor gap analysis for every client before strategy is finalised. The goal isn't to copy what competitors are doing — it's to identify exactly where their authority is strong, where it's thin, and where we can build a meaningful advantage in AI search. Below are three anonymised competitor analyses from real engagements to illustrate the methodology.
Example A: Financial Services Category (Client vs. Market Leader)
Client: mid-size SA financial advisory firm. Competitor: a JSE-listed financial services group with 40+ years of history. The client ranked comparably on traditional Google but had dramatically lower AI visibility. Here's what the gap analysis revealed:
| Gap Dimension | Client Score | Competitor Score | Gap | Priority |
|---|---|---|---|---|
| Entity Authority — Wikipedia presence, Wikidata completeness, Knowledge Panel | 18/100 | 84/100 | -66 | Critical |
| Citation Network Position — how many AI-trusted sources cite this entity | 22/100 | 71/100 | -49 | Critical |
| Content Authority — original research, cited data, expert bylines | 41/100 | 67/100 | -26 | High |
| Trust Signal Coverage — FSCA registration visibility, accreditation mentions, reviews | 38/100 | 79/100 | -41 | Critical |
| Domain Authority — DR / traditional link profile | 51/100 | 74/100 | -23 | Medium |
| AI Citation Frequency — appearances in AI search results for category queries | 9/100 | 68/100 | -59 | Critical |
What the gap analysis revealed was that the competitor's AI visibility wasn't driven by better content or better SEO — it was driven by a 40-year legacy of mentions in authoritative South African financial media, regulatory databases, and JSE records. These "reality signals" — offline authority that has been digitised — are things a younger brand cannot replicate quickly.
Our strategy: rather than trying to close the entity authority gap directly (too slow), we focused on the content authority gap first — producing original financial research that Perplexity and ChatGPT could cite — while simultaneously building the entity foundation in parallel. Within 6 months, the client's AI citation frequency score moved from 9 to 31. The entity gap won't close fully for 18–24 months, but citation frequency is already improving because content authority can be built faster.
Example B: E-Commerce Category (Three-Way Comparison)
A retail client wanted to understand how they stacked up against two competitors in AI search for their product category. All three brands have similar DR scores (47–53) and comparable traditional Google rankings. Their AI visibility looks like this:
The analysis revealed the client's single exploitable advantage: their content quality score (44) outperforms Competitor A (38), even though their overall AI visibility lags significantly. Competitor B has strong content (71) but weaker entity and citation authority than Competitor A. Our strategy: invest in entity and citation authority first (closing the gap on Competitor A's main advantages), while leveraging the existing content strength. In the medium term, building content authority to match Competitor B creates a brand that outperforms both competitors on combined GEO score.
What We're Currently Testing
Research doesn't stop. Below are the active experiments running in our synthetic authority labs right now — questions we don't yet have definitive answers to, but will.
Our 20 Biggest GEO Research Findings
These are not hypotheses. These are observations from our proprietary dataset — things we believe to be true based on measured evidence, flagged as our findings rather than established industry consensus.
Our Predictions for AI Search in 2027
Predictions in a fast-moving field are always partially wrong. We make them anyway because a reasoned view of where things are heading is better than paralysis. These are informed by our two-year dataset, our understanding of the technical direction of the major AI systems, and observable trends in user behaviour.
What We Stand For — and What We Won't Do
GEO is a young discipline with no established ethical standards. That creates opportunity for bad actors. We're deliberately transparent about where we draw lines — both because it's the right thing to do, and because we believe the AI systems we're optimising for are getting better at detecting and discounting manipulation.
We Will Never
We Always
"We optimise for being cited because you're genuinely the best answer — not because you've gamed the system. Gaming AI systems is a short-term play; the systems are getting smarter faster than the manipulators."
What We Offer That Others Don't
If you want someone to write blog posts and call it GEO, there are dozens of agencies that will do that. If you want a research-backed, measurement-driven programme that actually tracks whether AI systems are citing your brand — and systematically improves that — we're the only agency in South Africa running that programme.
Jaco Smit
10+ years SEO strategy
CIO Consultant to major SA organisations
AI Researcher — published findings
Biochemist by academic background
GEO Specialist since 2023
generativeoptimization.co.za ?
Jaco Smit is not a marketer who pivoted to AI. He's a software developer with over two decades of experience building systems, a biochemist who understands how complex datasets work, and an AI researcher who began studying generative systems before the current wave of tools existed.
His background matters because GEO sits at the intersection of software, data science, and search strategy. Most people coming to this space arrive from one direction — usually content marketing or SEO. Jaco's technical depth means he understands not just what AI systems respond to, but why: how language models construct world models, how retrieval-augmented systems source content, and how entity graphs are built and queried.
CIO Experience
Jaco has served as a CIO consultant to major South African organisations, an experience that shapes how he approaches GEO for clients. He understands how decisions get made at the executive level, what risk looks like from a board perspective, and why a sustainable, measurable strategy outperforms a flashy short-term play. Clients who work directly with Jaco often comment that the thinking goes deeper than they expected — because it does.
The Research Commitment
The proprietary database, the synthetic authority labs, the controlled experiments — these weren't features added to make the agency look impressive. They were built because Jaco found the available GEO research inadequate and decided to generate better data. The 100,000-query dataset didn't exist when he started building it. He built it because the questions he needed to answer for clients couldn't be answered any other way.
AI Research Background
Jaco's AI research work predates his GEO specialisation. He has studied the architecture of large language models, how they encode entity knowledge during training, and how retrieval-augmented generation systems integrate live data with trained knowledge. This theoretical foundation is what makes his practical GEO strategies different — they're built on an accurate model of how the systems actually work, not on inference from surface-level observations.
The South African Context
Working in South Africa adds layers that international GEO frameworks don't account for. POPIA compliance, FSCA regulatory authority, JSE-listed entity signals, the role of South African media in AI training data, the particular weight given to News24 and TimesLive in AI citation networks for local queries — these are things Jaco has mapped specifically for the SA market. The strategy we build for a Johannesburg financial services client is not a copy of a US GEO playbook adapted superficially. It's built from the ground up with SA authority structures as the foundation.
"Most GEO strategies I see are content strategies in new clothes. Real GEO requires understanding how AI systems actually construct and update their world models — and that's a technical question, not a content question."
Find Out Where Your Brand Stands in AI Search
We start every engagement with a baseline AI visibility audit — tracking your citation frequency across ChatGPT, Perplexity, Gemini, Grok, and Google AI Overviews, and benchmarking it against your top three competitors. You'll know exactly where you are before we discuss where you need to be.
South Africa's dedicated GEO research agency. Based in Johannesburg. Working with clients nationally and across Africa.