Skip to main content
AEOField notes

How a Johannesburg business ranks in ChatGPT, Perplexity, and Google AI Overviews in 2026

A practical playbook for South African service businesses who want to be cited by AI engines, not just indexed by Google. Built from the RankSmith field test we run every month.

RankSmith editorial11 minute read

RankSmith runs the same field test every month on a growing list of Johannesburg service businesses. We ask the same twelve queries across Google AI Overviews, ChatGPT web search, Perplexity, Claude, and Gemini. We log which engines cite our clients, which cite competitors, and which give a generic answer with no citation. Two years into the test, the pattern that produces citations is consistent. Here is what works.

Why is ranking on ChatGPT and Perplexity a different problem than ranking on Google?

Ranking on classic Google has always been a page contest. Ten blue links, one position per result, the whole page either wins or loses. A Johannesburg plumber ranking third for "emergency plumber Johannesburg" gets roughly seven percent of the click share on that query. Standard playbook, standard SEO.

Ranking on AI search is a passage contest. The user asks a question, the engine fetches a handful of candidate pages, extracts the passages that best answer the question, and cites one to three sources inside a synthesised answer. The passage wins or loses, and whether the user ever sees the underlying page depends on whether they click the citation chip.

Two consequences for a South African service business.

First, you need to win at the passage level. A page that ranks fifth for a query on classic Google but has the clearest answer first heading and first sentence can win the AI citation while the top ranked page loses it. Specific H2s, specific numbers, specific places.

Second, trust signals that Google already rewards are now more important, not less. A verified Google Business Profile, consistent NAP (name, address, phone) across directories, and LocalBusiness schema are the signals AI engines use to decide which of three passages to trust. If you have not done that work, the engine will find the passage and cite a larger competitor's page instead.

GOOGLE AI OVERVIEWSCHATGPTPERPLEXITYCLAUDEGEMINIJOHANNESBURGService businessMONTHLY RANKSMITH CITATION TEST, 5 ENGINES, SAME QUERY SET
FigureThe RankSmith monthly citation test. A Johannesburg service business at the centre, five AI engines around the edge. We run the same twelve queries every month and log which engines cite the client. Google AI Overviews is the primary target for South African reach.

What does a Johannesburg service business need in place to be cited?

There is no trick. The five layers below, in priority order, are what we install on every RankSmith engagement. Every layer feeds the next.

Layer one, the site itself. A fast Next.js site, served over HTTPS, with stable URLs, clean semantic HTML, and no client side rendering on content that matters. WordPress sites running nine plugins render differently for the crawler than they do for a user, and AI engines rely on what the crawler sees.

Layer two, schema. Organization and LocalBusiness on every page. Service on every service page. Article on every blog post. FAQPage where there are questions. BreadcrumbList on non home pages. Each schema document uses a canonical @id so the engine can bind every mention to the same entity across the site. This is the layer where most SA sites fail completely.

Layer three, answer first content. Every H2 is a question. Every first sentence is an answer with at least one number or place. We cover this pattern in detail in how to write an answer first H2 that ChatGPT will actually cite. The short version: if the engine has to rewrite your passage, the citation goes away.

Layer four, llms.txt and llms-full.txt. Short, annotated, at the root of the domain. This tells a future crawler where to start. Today it is low impact, tomorrow it might be the difference. We cover it in llms.txt in 2026, what we actually put in the file.

Layer five, the trust signals Google still controls. A verified Google Business Profile. A claim on Bing Places. Listings on South African directories (SnupIt, Brabys, Yellow Pages). Reviews from real clients. Inbound links from local trade press or industry bodies. These were SEO basics in 2015 and they have never been more important, because AI engines defer to Google's trust graph when weighting passages.

TRUST SIGNAL STACK, JOHANNESBURG SERVICE BUSINESSLocalBusiness schemaJSON LD on every page footerREAD BYGoogle, AI Overviews, ChatGPT webGoogle Business ProfileVerified listing with photos and hoursREAD BYGoogle AI Overviews, GeminiAnswer first contentH2 questions with one sentence answersREAD BYChatGPT, Perplexity, ClaudeThird party mentionsCited on local directories, trade pressREAD BYAll five, trust weightOrganization schemaOne object bound by @id across pagesREAD BYEvery engine, entity binding
FigureTrust signal stack for a Johannesburg service business. Each layer carries a different artefact and feeds different engines. Third party mentions (amber) is the layer most SA businesses under invest in, and the one that most reliably lifts citation rate.

How much of the work is schema, and how much is content?

A site with perfect schema and vague prose gets indexed but not cited. The engine knows the business exists, knows where it is located, knows what it sells, but finds no passage worth quoting. It cites a larger competitor.

A site with sharp prose and no schema gets quoted inconsistently. The engine extracts a passage, has no verified entity to bind it to, and sometimes gets the business name wrong or conflates the business with a similar one elsewhere. We have seen Johannesburg plumbers cited as Cape Town plumbers in AI answers because the schema was missing.

Both failure modes fix themselves when the two layers ship together. Schema on every page, first, because it is the cheapest change and unlocks rich results as well as AI citation. Content rewrite next, because it is slower and needs editorial judgement.

Our order on every RankSmith engagement:

  1. Week one, install schema. Organization, LocalBusiness, Service on relevant pages. Validate with Google Rich Results and Schema.org Validator.
  2. Weeks two to three, rewrite the highest value pages to answer first structure. Usually three to five pages account for eighty percent of the citation opportunity.
  3. Week four, add llms.txt and llms-full.txt.
  4. Week five and beyond, work the directory and Google Business Profile layer.

Rankings in AI engines typically start moving between weeks four and six. Classic Google rankings follow by week eight or ten.

What does the RankSmith field test look like each month?

QUERYUser asks AI enginehow much does a Next.jssite cost in Joburg?CITATIONEngine cites RankSmithpassage lifted from/pricing, with domainchipCLICKUser follows citationdirect to /pricing or/auditBOOKStrategy call bookedrevenue momentAI ANSWER TO BUYER, FOUR STAGES, DIRECT ATTRIBUTIONEach stage is a discrete measurement point. Citation is the earliest moment we can see in analytics.
FigureThe conversion path from an AI citation to a booked strategy call. Four stages, each a measurement point. Citation is the earliest we can see in analytics, book is the revenue moment.

The field test runs on the last Friday of every month. Same twelve queries, same browser, same South African geolocation. The queries are chosen per client but follow a shape:

  • Three head queries in the client's core service category. "Emergency plumber Johannesburg", "commercial cleaning Sandton", "tax consultant Rosebank".
  • Three long tail buyer intent queries. "How much does an emergency plumber cost in Joburg", "best commercial cleaner in Sandton for offices over 1000 square metres".
  • Three research queries where AI citations are heavy. "What should I look for in a Johannesburg accountant", "how often should an office be cleaned in South Africa".
  • Three negative control queries unrelated to the client. If the engine cites the client on these, something is wrong (typically over aggressive entity binding that will backfire).

For every query we log five things: was the client cited, was the competitor cited, what passage was lifted, what URL carried the citation chip, and was the business name attributed correctly. The spreadsheet takes ninety minutes.

A month later we repeat. The month over month delta is the signal. A business moving from two citations to seven citations in sixty days is the same pattern we have seen across every RankSmith engagement. A business stuck on zero citations for ninety days is a sign that one of the five layers above is missing. We audit the layers in order and fix whichever is weakest.

What should a Johannesburg business do first if they are not being cited at all?

The single fastest move we can give a business is a thirty minute self audit.

First, view source on your homepage. Search for application/ld+json. There should be at least two script tags, one for Organization and one for LocalBusiness. Each should include your business name, your physical address in Johannesburg, your phone number, and your URL. If either is missing or the content is generic, schema is your first fix.

Second, look at the H2 on your highest value page. Usually that is the services or pricing page. Is the H2 a question a buyer would type? Is the first sentence under it a complete answer in under twenty words with a specific number? If not, rewrite it. That one change moves more AI citations than any other on a typical site.

Third, visit yourdomain.co.za/llms.txt. If the file 404s, create one. Twenty lines, short blockquote summary, a services section, a company section. If the file exists but points to vague pages or uses generic annotations, rewrite the annotations to start with a verb and the brand name.

If all three pass and you are still not being cited, the fourth fix is outreach. One inbound link from a trusted South African site in the last ninety days is enough to lift the trust weighting. Trade press, industry bodies, or co-authored content with another SA business all work.

We run the full audit across all five layers on every free audit we deliver. The report comes back in forty eight hours with a prioritised fix list and a quote for the work. Or book a strategy call and we will walk through your current citations live, engine by engine, on a screen share.

RankSmith is based in Johannesburg. We serve clients across Gauteng and the rest of South Africa. Pricing in ZAR is on our pricing page, with setup fees from ZAR 48,000 and monthly retainers from ZAR 1,500. Every site we ship has all five layers above installed from day one.

Frequently asked questions

Why is AI search ranking different from classic Google ranking?
Classic Google ranks whole pages and shows ten blue links. AI search retrieves passages, synthesises an answer, and cites one to three sources. The work shifts from page level authority to passage level clarity and trust signals a model can attach to a claim.
Which AI engines matter for a South African business in 2026?
Google AI Overviews has the widest SA reach because it sits inside the existing Google search experience. ChatGPT web search and Perplexity matter for desktop heavy professional buyers. Claude and Gemini are smaller slices but growing. We optimise for all five in parallel because the underlying requirements overlap.
How much of AEO is schema versus content?
Roughly half and half. Schema gives the engine a verified model of the business. Content gives it the passages to quote. Neither works alone. A site with perfect schema and vague prose gets indexed but not cited. A site with sharp prose and no schema gets quoted inconsistently or not at all.
What should a Johannesburg business do first if they are not being cited?
Audit three things. Do you have Organization, LocalBusiness, and Service schema shipping on every page. Do your H2s answer a question in one sentence. Do you have an entry in your llms.txt that points to your most important page. Fix those in order.

Ready when you are

Want this level of work on your site?

Book a thirty minute strategy call. We will audit your current rankings in Google and in AI engines, and map the fastest wins we can ship in the next sixty days.

Chat on WhatsApp