What This Article Covers
AI visibility — GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization) — is how often your brand is mentioned when AI systems answer questions about your industry. We covered the definitions and why they matter now in a separate article: GEO and AEO: Why ChatGPT and Google’s AI Don’t Recommend You.
This article is about one question only: how do you measure it? Because “am I visible in ChatGPT” is far easier to measure wrongly than most people assume — and a wrong measurement sends the budget to the wrong work.
Why Google Rankings Don’t Answer It
The most common objection we hear: “We already rank first on Google, so AI will mention us too.” In every case we have measured, that assumption turned out to be wrong.
The reason is simple but important. Google offers a list; an AI constructs an answer. A list can be extended — page two, page three. An answer cannot. There are three to five names the model can fit into a few sentences, and ranking is not how it picks them.
To build its answer, the AI reads a handful of sources and relies on what those sources say. You may have an excellent page about yourself, but if the model never reads it, you cannot enter the answer. Meanwhile a weaker competitor listed in a directory the model reads often will make it in.
That is why this measurement is a separate job from rank tracking. Not a different tab of the same tool — a different tool.
“Mentioned” and “Known” Are Different Things
Two separate situations need to be distinguished when measuring:
- The model’s memory. Without searching, does the AI recognise your brand from its training data? This reflects your general footprint on the internet and is very hard to change in the short term.
- The search-grounded answer. When the model looks at the live web, does it find you? This is the part you can work on — because you can change the live web.
In practice the second matters far more. For questions like these, today’s assistants mostly search. The answer is built from yesterday’s web, which means work done today can show up tomorrow.
One Query Is Not Measurement
The common mistake: someone opens ChatGPT, asks about their industry once, and concludes “I’m invisible” or “I’m fine.” Both conclusions are unreliable, because AI answers vary from run to run.
In our own measurements, running the same question repeatedly produced a mention in one round and no mention in the next. Judging from a single query is like flipping a coin once and calling it statistics.
Five Conditions for a Proper Measurement
- A frozen question panel. Thirty questions people actually ask in your industry, fixed in place. If the questions change, comparison breaks — and the next measurement cannot answer “did we improve?”
- Repetition. Each question runs multiple times. The result is reported as a rate, not a yes/no.
- Separate out the brand question. If your panel includes “what do you know about this company?”, the brand name sits inside the question, so it is almost guaranteed to appear in the answer. That question inflates the score. When we removed it from our own measurement, two of our four sites dropped to zero. Keep the brand question if you like — but report it separately.
- Multiple engines. ChatGPT, Gemini and Claude answer the same question differently. In a small test of ours the same business scored 100% on two engines and 67% on a third — with 3 questions and no repeats, that is a signal rather than a measurement. But it is enough to show that measuring one engine gives you partial information.
- Source tracking. Record which sites the answer was built from. Your action list comes from here.
The Four Numbers a Measurement Gives You
- Visibility score: the share of answers in which you were mentioned.
- Share of voice: your share among all brands mentioned. This is usually the number that stings — in most sectors marketplaces and directories dominate local businesses.
- Source ranking: where your own site sits among the sources the AI reads, if it appears at all.
- What is said about you: we have seen models surface an old complaint record inside an otherwise neutral recommendation.
After Measuring: How to Improve Visibility
There is no magic switch, but the work is not vague either:
1. Write pages that answer the question. AI quotes the page that answers the question, not your services page. “What we do” does not get cited; “how to choose X and what to watch for” does.
2. Exist in the sources the model reads. Measurement shows you which sites those are — usually a few directories, a few review sites, and a few strong competitor blogs. A brand missing from that list is missing from the answer.
3. Make your identity machine-readable. Business name, address, phone and service descriptions identical everywhere, plus structured data (schema) and machine-readable files such as llms.txt.
4. Correct wrong information. If a model says something inaccurate about you, the source has to be found and fixed. This usually comes down to a single page.
5. Measure again. Because AI reads the live web, work shows up in days or weeks rather than months, unlike SEO. But as competitors start doing the same, position has to be defended.
Where to Start
Start with measurement. What to do only makes sense once you know where you stand — and for most businesses the first measurement is a surprise.
If you would like to see your own situation, you can use our free AI visibility scan: enter your domain, let the questions run, and see the actual answer along with whether your brand appears — in about 30 seconds.
We applied this method to our own four sites and published the results as they came out: We Asked AI About Our Own 4 Websites.
The full 30-question repeated measurement and the report card are what we do: our AI visibility service.
