Research: why ChatGPT names your competitor and not you

The question you are asking yourself now

You asked ChatGPT who is the best in your trade, in your town. You got three or four names. One of them is a competitor. Yours was not there.

The question that follows is simple: is that chance, or does the other business have something you do not?

We have been measuring this since April for businesses in and around Eindhoven. This piece sets out what the numbers show and, just as plainly, what they do not.

What we measured

Two kinds of measurement, on the same businesses.

Who the AI names. For each business we put the question a customer would ask to several AI assistants and record which businesses appear in the answer. That tells us, per assistant, whether a business is named.

How the website holds up. For 238 small business websites in the region we measured, between 7 April and 18 October 2026, what an AI system can read on the site: structured data such as LocalBusiness, FAQ and review markup, clear business details and citable facts. We sum that up in a GEO score from 0 to 100. We also collect the number of Google reviews and the average rating.

Twelve estate agents in Uden, one question

The cleanest comparison we have is a market where we asked every business the same question. On 23 July 2026 we asked ChatGPT, Claude and Gemini for the best estate agent in Uden, for twelve agencies at once. The question was literally “Beste Real Estate in Uden?”, because our system still took the trade label from Google at the time. That is not how a customer would put it, but it was exactly the same question for all twelve.

Six agencies were named by at least one assistant, six by none.

	Named (6)	Not named (6)
GEO score of the site, average	51	47
LocalBusiness markup on the site	5 of 6	4 of 6
Google reviews, median	16.5	6
Average rating	4.4	3.7

The biggest difference is in the reviews: more of them, and rated higher. The difference in the sites is small.

But look at the two agencies that stand out:

• An agency with only 9 reviews was named by all three assistants. It had the strongest site of the twelve: a GEO score of 62, with LocalBusiness markup.
• An agency with 15 reviews and a 4.7 average was named by none. It had better reviews than most of the named agencies, but a site without LocalBusiness markup and a GEO score of 41.

A third case fits the same picture: an agency with 108 reviews and a site without markup was named by two of the three. That many reviews carry a weak site most of the way, but not all of it.

What we read in this: the AI does not choose on one thing. Site and reviews count together. Being strong on one can partly make up for the other. Being strong on one and weak on the other still risks being passed over.

When the AI names everyone

The pattern does not hold everywhere. In early October we measured eleven golf courses in Brabant. Their sites were weak, with GEO scores between 5 and 41. Yet all eleven were named by at least one assistant, nine of them by all three.

The explanation is obvious: there are only a few golf courses per town. Ask for a golf course nearby and the AI has little to choose from, so it names what there is.

For estate agents, painters, accountants and SEO agencies it is different. There are dozens per town and the AI has to choose. That is exactly where your site and your reviews start to count.

How sites in the region hold up

Of the 238 small business websites we measured:

• 75% have no LocalBusiness markup, the structured data that says who you are, where you are and what you do;
• 87% have no FAQ markup;
• 96% have no review markup;
• the average GEO score is 32 out of 100.

That means most businesses make it needlessly hard for an AI system to recognise and cite them. It also means that in most markets it takes little to get ahead on this point.

What this research does not show

We measure so we can be honest about what works, so also about the limits of these numbers:

• Twelve businesses is a pattern, not proof. The Uden comparison is small. With other agencies or another town the gap could be larger or smaller.
• The 238 sites are not a random sample. Many are businesses we approached because their site could be improved. The average small business probably scores somewhat better than these percentages.
• Every measurement is a snapshot. AI assistants do not always give the same answer to the same question. A business that is not named once may be named the next time.
• Correlation is not cause. Agencies with many reviews may also be larger, older and mentioned more often elsewhere. Then reviews are more a sign of being known than the switch that makes the difference. Our numbers cannot separate the two.

What you can do yourself

If you heard ChatGPT name a competitor, this is what the numbers give you:

1. Check whether it was chance. Logged out, ask ChatGPT, Claude and Gemini the same question and repeat it a few times. If all three keep skipping you, it is not chance.
2. Look at both sides at once. How many Google reviews do you have against your competitor, and with what rating? And does your site say, in structured data, who you are, where you are and what you do?
3. Ask your supplier to sort out both. Not just “make my site better” and not just “ask for reviews”, but both, because each one only works fully with the other.
4. Measure again. After a few weeks, ask the same questions again, to the same assistants. Then you can see whether anything has moved.