Short answer: To check if ChatGPT recommends your business in Africa, ask it the questions your customers ask, naming your city, in a clean session: logged out, or in a temporary chat set to Unpersonalized, with web search on. Ask each question three times, because answers change between runs, and record who is named and which sources are cited. Repeat the test in Perplexity and Gemini. The share of answers that name you is your result. For a business in Lagos, Nairobi, Accra or Douala, also test the way customers in other cities and abroad would ask.
Ifeoma runs a physiotherapy clinic in Lekki, Lagos. On a quiet Tuesday she opened ChatGPT on her phone and typed “best physiotherapy clinic in Lekki”. Her clinic came up first. She took a screenshot and sent it to the family WhatsApp group, where it collected eleven thumbs-up and one uncle asking whether ChatGPT was a new bank.
A week later a patient mentioned she had found the clinic through a friend, not through ChatGPT. Ifeoma asked her niece to try the same question. The niece, on her own phone, got three clinics. None of them was Ifeoma’s.
Nothing was broken. Ifeoma had asked once, logged in to an account that had spent months helping her write clinic newsletters. The assistant knew exactly whose clinic she might be hoping to see. Her customers’ phones did not.
Illustrative composite — not a client, not a testimonial
We have written about what makes a business worth recommending in how to rank on ChatGPT. This guide covers the step before that: finding out, accurately, whether the assistant names you today. It is written for businesses in Nigeria, Kenya, Ghana, Cameroon and South Africa, where the usual advice of “just ask ChatGPT” misses how location, language and a crowded field of directories change the answer.
Why one question proves nothing
One answer from ChatGPT does not tell you whether it recommends your business, because the same question can produce a different list on the next run. ChatGPT writes each answer fresh. When it searches the web, OpenAI says it typically rewrites your question into one or more targeted search queries, reviews the results, and may send further queries before answering. Different queries find different pages, and different pages name different businesses.
Personalisation adds a second layer. OpenAI says that if memory is enabled, ChatGPT may use saved memories when rewriting a search query. An owner who has spent months discussing her own clinic with the assistant is the least reliable tester that clinic has. Checking whether ChatGPT recommends you from your own account is like asking your mother whether your cooking is good. The answer is warm, consistent and of limited commercial use.
Set up a clean test session
A clean test session is ChatGPT used logged out, or in a temporary chat set to Unpersonalized, with web search switched on. OpenAI says web search is available to people who are not signed in, and that an unpersonalized temporary chat does not use memory, custom instructions or plugins. Either option removes most of what the assistant knows about you.
To start a temporary chat, open a new chat, select Temporary, and choose Unpersonalized before you send the first message; the choice cannot be changed once the conversation starts. ChatGPT may search the web on its own when a question needs current information, and you can force it by selecting Search from the tools menu. Use search for this test. A recommendation question answered from the model’s training data alone tells you about the internet of a year or two ago, which is a fine thing to know and the wrong thing to measure.
Then turn off anything that pretends you are elsewhere. A VPN changes the location ChatGPT infers, which leads to the next point.
Test from where your customers are
ChatGPT uses your approximate location to answer local questions, so a test run from your office only shows what customers near your office see. OpenAI says ChatGPT may estimate your general location, such as your country, state or city, from your IP address, and may rewrite a question such as “good restaurants near me” into a search for that area. Device location sharing is optional and off by default.
For an African business this matters more than the usual guides admit, because many customers are not where the business is. A Lagos clinic is searched for from Abuja and Port Harcourt. A Nairobi safari operator is searched for from Munich and Mumbai. A Douala law firm is searched for by Cameroonians in Paris and Montreal, in French. Three practical adjustments cover most of it:
- Always name the place in the question. “Best physiotherapy clinic in Lekki” tests the same thing from any location. “Best physiotherapy clinic near me” tests only where you happen to be standing.
- Ask someone in a key customer city to run a few questions. A relative in London or a supplier in Accra running five questions once a month is a better diaspora test than any tool that guesses.
- Test in the language customers use. For Cameroon, ask in French: “meilleur cabinet d’avocats à Douala” can return a different list from the English version. We cover the French-language side in notre guide du référencement à Douala.
Ask the way African customers ask
The questions in your test should be the questions customers actually ask, naming the neighbourhood, the service, the price range and the urgency. Owners tend to test with their business name. Customers rarely know it yet; they ask for a category in a place. A useful test mixes four kinds of question:
- Category: “Best dental clinic in Westlands, Nairobi for a family with young children.”
- Problem: “Who can fix a leaking roof in East Legon this week?”
- Comparison: “Alternatives to [your best-known competitor] for company registration in Abuja.”
- Brand: “Is [your business name] in Lekki reliable?” This one tests accuracy, not discovery.
Use local detail where customers would. Nigerians ask about prices in naira, Kenyans in shillings, and a question that mentions a budget gets a different answer from one that does not. Avoid testing with “best business in Africa”; nobody searches for a plumber at continental scale, however ambitious the plumber.
Run the test and record a score
Run each question three times in separate clean sessions, then record your score as the share of answers that name your business. Ten questions asked three times gives thirty answers. Being named in twelve of thirty is a very different position from being named in two, and both are more useful than a single screenshot.
- Write ten customer questions across the four kinds above, each naming a place.
- Open a clean session: logged out, or temporary and Unpersonalized, with Search on.
- Ask each question three times in separate sessions, and save each full answer and its sources.
- Score the answers: named or not, position, details correct or wrong, and who else appeared.
- Repeat in Perplexity and Gemini, and note whether Google shows an AI Overview for your main search.
- Repeat monthly, on roughly the same day, changing one thing at a time.
A spreadsheet with one row per answer is enough. The columns that matter are the question, the assistant, the run number, whether you were named, your position, whether your details were correct, the competitors named, and the sources cited. There are paid dashboards that will report an AI visibility score to two decimal places. Thirty honest rows in a spreadsheet will tell you more about a single business in Accra.
Read the sources, not just the names
The sources ChatGPT cites tell you why it named the businesses it named, and they are the most useful part of the test. OpenAI says responses that use web search may include citations, and that selecting Sources shows the cited pages and other relevant links. Open them. Each one is a page the assistant read before deciding.
In African markets those pages are often directories, listicles and aggregators rather than businesses’ own sites. In our study of nine searches in five African cities, a business’s own website was 23 percent of page-one organic results in travel and hospitality, against 68 percent in professional services, and Google served an AI-generated answer on all nine searches. Our October re-run found the hospitality share had risen to 28 percent. If the assistant keeps citing the same two directories, getting listed accurately on those two directories is the most direct fix available to you.
Check the dates as well. OpenAI itself warns that search results and citations can be incomplete, outdated or incorrect. An assistant quoting your old address is usually quoting an old page somewhere, and the page is what needs correcting.
What your result means and what to fix
Every test result falls into one of three outcomes, absent, named with wrong details, or named accurately, and each one points to a different fix. Being absent is the normal starting point for a small African business, not a verdict on its quality. The table below maps each outcome to its usual cause and its first fix.
| What the test shows | The usual cause | The first fix |
|---|---|---|
| Never named, directories named instead | The assistant finds no clear page about your service in your city | One page per service and location, plus accurate listings on the directories it cites |
| Never named, competitors named | Competitors have more reviews and more outside mentions | Detailed reviews and a complete Google Business Profile |
| Named, with wrong address, phone or prices | An outdated page or listing the assistant is reading | Correct the source it cites, then your other listings |
| Named only from your own account | Personalisation, not visibility | Re-test in a clean session before celebrating in any group chat |
| Named accurately in most answers | Your sources are clear and consistent | Keep reviews coming and re-test monthly |
For a business that is absent, the work starts with the basics in how to appear on Google Maps in Lagos and how to get more Google reviews in Nigeria, which apply well beyond Lagos. Also confirm your site lets the assistants in: OpenAI says sites that are opted out of its OAI-SearchBot crawler will not be shown in ChatGPT search answers. If you would rather start from a baseline someone else has run, our AI visibility check runs a version of this test for your business, and our GEO services page sets out the work that follows it. City-specific detail is on AI SEO in Lagos, Nairobi, Accra and Douala.
When this test is not worth your time yet
The test is not worth running monthly yet if your business has no website, no Google Business Profile and fewer than a handful of reviews, because the result will be “absent” every time and the fixes are already obvious. Run it once to have a baseline, then spend the hour on the profile and the reviews instead.
It is also not worth much if almost all your customers come through referrals and WhatsApp, and you have more work than you can take. An AI assistant is one more front door; it is not compulsory to paint every door at once. We sell AI visibility work, so this section is against our interest, and we would add one further honest note: when we ran our own check in August, AfriShield was named in none of the four African regions we tested. We run the same test on ourselves every month now, from a clean session, with no screenshots sent anywhere until the number moves.
Ifeoma ran the full test in September: ten questions, three runs each, from her niece’s phone. Her clinic was named in four answers out of thirty, and two of those listed an old phone number from a directory she had forgotten existed. She fixed the directory, asked recent patients for reviews that named the treatment and the neighbourhood, and put the next test in her calendar. The screenshot stays off the family group until the number is worth a thumbs-up from the uncle.