All articles
How-to

Does AI know your business exists? Check it in 5 minutes

A free, do-it-today test: ask the AI engines what your customers ask — in English and in isiZulu, Afrikaans, isiXhosa and Sepedi — and learn to read the answer.

By LoopCite23 June 20267 min read

More of your customers are asking an AI — “who’s the best [your service] near me?” — before they ever open Google. So here is a question worth five minutes of your time: if someone asked ChatGPT that today, would it name you? You don’t need a tool to find out. You need four chatbots, a few honest questions, and a way to read the answers. Here’s how.

Why you can’t just check Google

Two things have changed. First, discovery has moved: a rising share of people start in a chatbot, not a search box (we covered the research in how search is changing). Second, even when AI does send you a customer, it usually arrives later through branded search — so your analytics quietly mis-files it (that’s the downstream impact of AI visibility). The upshot: your Google ranking tells you almost nothing about whether AI recommends you. You have to ask AI directly.

The 5-minute test

Work through these four steps. A notebook or a spreadsheet is all the tooling you need.

  1. Write the questions a customer would ask — not your name. The whole point is to test discovery, so don’t prompt with your brand. Use the questions a stranger would: “best electrician in Durban”, “affordable solar installers near Pretoria”, “which accounting firm is good for small businesses in Cape Town”, “[your product] vs [a competitor]”.
  2. Ask each engine the same questions. Run every question through the engines your customers actually use — ChatGPT, Google’s Gemini (and the AI Overview at the top of Google’s own results), Microsoft Copilot, and Perplexity. They read different sources, so they give different answers.
  3. Ask in more than one language (see the next section).
  4. Record four things for each answer: were you named? Was your website cited as a source? Was what it said accurate? And which competitors were named instead of, or alongside, you?
Don’t prompt with your own name. If you have to tell the AI you exist, you haven’t tested whether it would recommend you — you’ve tested whether it can read.

Ask in the languages your customers use

This is where most South African businesses have a blind spot — and an opportunity. AI engines answer in isiZulu, Afrikaans, isiXhosa and Sepedi, and almost nobody is optimising for those answers. Ask the same question in the language your customers actually use. In Afrikaans, for example: “Wie is die beste elektrisiën in Pretoria?” Then do the same in isiZulu, isiXhosa or Sepedi for your market.

You will often find you are visible in English and invisible the moment the question switches language. Being the cited answer in isiZulu is, for now, a wide-open lane.

How to read the answer

There are really only four outcomes, and each tells you what to fix:

  • Not mentioned at all. The AI either can’t find you or doesn’t consider you a credible answer. Usually a readiness problem (the site isn’t machine-readable) or an authority problem (you’re absent from the sources AI trusts).
  • Mentioned, but your site isn’t cited. The AI knows of you from elsewhere but isn’t using your own content. You’re leaving the story of your business to be told by third parties.
  • Named and cited. The goal. Now the work is defending it and widening it to more questions, engines and languages.
  • Mentioned, but wrong. The most urgent. If the AI states the wrong price, location, services or hours, it will keep repeating that until the sources it reads say otherwise.

What to do with what you find

Whatever the result, the levers are the same. Make your site genuinely readable by machines (real text, structured data, crawlers allowed). State the facts AI keeps getting wrong, plainly, on your own pages. Get listed and accurate on the third-party sources AI cites for your category. And publish answer-shaped content for the exact questions you tested — in the languages your customers ask them.

Doing it properly, at scale

One honest caveat about the five-minute test: it’s a snapshot. AI answers vary with phrasing, between runs, across engines and across languages — so a single check can flatter or mislead you. To track it properly you’d run many phrasings of many questions, across every engine and language, repeatedly, and score the results consistently.

Snapshot vs system
A manual check tells you where you stand today. Measuring it consistently — across engines, languages and phrasings, over time — is what turns a hunch into something you can manage.

That is exactly what LoopCite’s scan automates: it asks the questions your customers ask, across the engines and South African languages they use, then tells you where you’re named, where you’re cited, where you’re misrepresented, and which rivals are winning the answer — and hands you the fixes to close the gap.

Sources

This is a practical guide; the supporting context is drawn from the research behind our other two articles. We did not find published South-Africa-specific statistics on AI-search adoption — a genuine gap, which is part of why a first-party check is worth doing.

  1. SimilarWeb — “The Downstream Impact of AI Visibility” (why AI traffic hides in branded search)
  2. Bain & Company — “How customers are using AI search” (share starting in a chatbot)
  3. Gartner — search-volume decline prediction (19 Feb 2024)