I Tracked My Own Website in ChatGPT for 90 Days: What 247 Checks Taught Me

Why did I test this on my own website first?

I only recommend what I have tested on my own brand. Advice that has not been tested is only a theory. Before I tell a business in Kigali that it can be cited by ChatGPT, I want to see what happens on my own site, with my own numbers, including the ones that are not flattering.

So I measured fokwasiaka.com with Amaso, the WordPress plugin I built at FOSIA to check whether AI assistants cite a website. Over 90 days, it asked ChatGPT the questions my clients ask, again and again, and recorded each time whether my site was among the sources.

What did the 247 checks show?

• Overall: ChatGPT cited fokwasiaka.com in 22 checks out of 247, about 9%.
• “best ai search visibility strategies in Kigali”: cited 6 times out of 20, 30%.
• “best ai tools for african entrepreneurs in Kigali”: cited 8 times out of 57, about 14%.
• GEO score of the site: 53 out of 100, which tells me how easy my pages are for an AI to quote, and that there is room to improve.

One limit matters: these numbers measure ChatGPT only. They say nothing about Gemini, Claude, Perplexity or Google’s AI features.

How should you read numbers like these?

A rate, not a screenshot. The same question asked twice can return different sources. One answer that cites you, posted on LinkedIn, proves very little. What counts is the share of checks that cite you over weeks.

Specific beats general. My site was cited three times more often for a precise local question than on average. Questions that combine a topic and a place (“in Kigali”) have fewer good sources to choose from, so a clear local page has a real chance. Broad questions face the whole internet.

9% is a starting point, not a verdict. It means ChatGPT knows the site and uses it for some topics. It also means it leaves it out most of the time. That is useful: it tells me which questions to work on.

A citation is not a client. Visibility without conversion is vanity. The only question that matters is whether your presence online brings you real customers, so I look at enquiries as well as citations.

How can you run the same test on your business?

1. Write ten real questions your customers ask, in their words, with your city, and without your brand name.
2. Ask them to ChatGPT with web search on, in a fresh conversation, and note whether your site is among the sources and who is cited instead.
3. Repeat every week for at least a month, with the same questions, and keep a log.
4. Calculate your rate per question: checks that cite you, divided by all checks.
5. Work on the questions where you are close: rewrite the page that should answer, put the answer in its first lines, add the facts (place, prices, services), and check again.

Doing this by hand is possible but slow. Amaso does it from WordPress, and its guide How to check whether ChatGPT cites your WordPress site explains both methods. Amaso Lite is free on WordPress.org.

What will I watch next on my own site?

The same list of questions, every week, and the same discipline: diagnosis first, then changes, then a new measure. If you want the background on why this work matters for African businesses, read my guide to Generative Engine Optimization for African businesses and why I specialise in SEO and GEO for Africa.

Frequently asked questions

Is a 9% citation rate good?

There is no universal benchmark. It depends on the questions you track and on your competition. Compare your rate with itself over time, question by question, rather than with someone else’s number.

Why only ChatGPT?

Because that is what this measurement covered. I prefer to publish a narrow number I can prove than a broad claim I cannot.

Can I guarantee that ChatGPT will cite my site?

No one can. The AI companies decide what their assistants say. You can make your pages easy to understand and verify, and measure what happens.

Can you look at my numbers with me?

Yes. Book a free 30-minute intro call: I will tell you honestly what I see, and the right next step, or none.