We Grew a B2C Business on AI Visibility, from Scratch. How Does It Work in B2B?

Then we restructured the business around a different idea. We stopped working with intermediaries altogether and made the internet our only intermediary. At the centre of that decision were search engine optimisation (SEO) and generative engine optimisation (GEO), supported by our own paid campaigns on Google and Facebook.

In May 2026 we launched a new website built to the canons of both disciplines, with its own blog and the full technical groundwork behind it.

The 2026 season is now over, and it closed at roughly twelve times the level of 2025.

The moment I understood what was really happening came at the end of June. A man phoned to book and told me that ChatGPT had recommended us as the best. I had not expected that. From that day I kept a log of every enquiry and noted how each customer had found us. By our own count, around 80% of the calls came after a ChatGPT recommendation.

It is important to be exact about what this proves. This is a consumer business, and we were not relying on AI alone: we ran paid campaigns on Google and Facebook, and many customers left five-star reviews on Google Maps, exactly the kind of public signal these systems pick up. The result was real, but several forces worked together, and we cannot say how much belongs to which.

That is why what came next matters more to us. China GEO EU is a business built on AI visibility. It is B2B: no paid campaign and no crowd of consumers writing reviews. And it carries an obligation the yacht business never had. A company that sells AI visibility has to be visible itself, and it has to get there the way it would advise a client to. So we treated ourselves as our own first client and applied the same method we apply to anyone: audit, fix the foundations, audit again.

Step one: the baseline audit

Every engagement starts the same way. Before you change anything, you find out exactly what the AI assistants say today.

In mid-September, before the main website was live and before the first article was published, we ran a structured AI visibility audit on ourselves. Six standard questions, five AI assistants, three independent runs each, every answer recorded. The questions cover what the company is, who the founder is, who provides this service to European companies entering China, a how-to question, and two that a buyer would ask when choosing a provider.

The first question was the simplest: What is China GEO EU?

One assistant answered with a diplomatic overview of how China and the EU cooperate on earth observation satellites. Another explained a trade agreement on geographical indications. A third politely asked what we meant. Not one of them knew that China GEO EU is a company, and that it exists to help European businesses get represented accurately in AI answers.

Then the buyer question: Who provides AI visibility services for European companies entering China? Over three rounds, across all five assistants, the AI offered dozens of names. Ours was not among them. Not once.

We sell AI visibility, and the AI could not see us. For an audit, that is a perfectly normal first finding, and a valuable one: it is the baseline against which everything that follows is measured.

Step two: build the foundations, organically

We did not publish and hope. The work that followed was done to professional standard, in the order an audit prescribes: the facts first, then the content.

The website. chinageo.eu went live in mid-September, built to the full requirements of both SEO and GEO: the complete technical groundwork, the structure and the machine-readable information that lets search engines and AI systems understand who we are and what we do.

The articles. Over the following weeks we published six long-form articles, the last on 14 October. Each was held to the same standard as the site: written for human readers and for machine retrieval, sources checked, and no sentence stronger than the evidence behind it.

Sitemap, indexing and search bots: monitored every week. A site that is built correctly can still quietly break, so indexing and crawler access are a routine, not an afterthought. Every Monday an automated, read-only check opens the sitemap, confirms it is valid and lists every page, inspects the key pages, and reviews seven days of server logs for genuine Googlebot visits, ignoring look-alikes. It expects at least one visit and no errors. On the first of each month a fuller review compares indexed and excluded pages, and their reasons, with the previous month. Every new article is checked within two to three days, and any change to the infrastructure is checked the same day. If anything goes wrong, whether an invalid sitemap, errors served to Googlebot, a key page dropping out or indexed pages falling by more than a fifth, we react immediately.

One story, everywhere. GeoSynapse, China GEO EU, my LinkedIn profile, the company pages and every publication describe the company in the same terms. AI systems cross-check what they find, and when sources contradict each other, the answers suffer. We saw it first-hand.

The most instructive fix was also the smallest. When we asked the assistants who I am, several of them confidently described me as working for an earlier employer. It was not a hallucination. It was my own LinkedIn profile: the most recent entry in my experience section was that old job, and the machines leaned on it. On 13 September I added a new entry, Founder, GeoSynapse & China GEO EU, and rewrote the headline and summary. It was the cheapest fix on the list.

And there was one decision, taken consciously from the start: organic only. No paid advertising. No paid placements or sponsored publications. And no self-published ranking with ourselves at number one, the shortcut every founder in this field has been tempted by. We looked at how such rankings reach AI answers and wrote about it in How Self-Published Rankings Reach AI Answers. A company that sells accurate representation cannot begin by manufacturing its own endorsement. If the assistants name us, it should be because our facts are clear, consistent and verifiable, and we want to be able to say so without an asterisk.

The articles were shared on LinkedIn from the company pages and from my own profile. At the start, both company pages had zero followers.

Step three: the re-audit

On 15 October we ran the same six questions again, on the same five assistants. Because a single snapshot proves little, we did it three times in one day: morning, afternoon and late evening.

The picture had changed, though not evenly.

Assistant	September	15 October
ChatGPT	Treated "China GEO EU" as an ambiguous acronym	Names the company, and puts it first when asked who provides AI visibility services for European companies entering China, in every one of its answers that day
Perplexity	Earth observation and a trade agreement	Describes the company and its founder correctly; names it among the providers in the later sessions
Gemini	Recognised the founder's venture, but mostly not the company	Names the company in every session, but twice under a label merged with another brand's name
Claude	Earth observation, geopolitics	Could not find a company by that name; suggests other people when asked about me
Grok	Earth observation leadership	Unchanged

Two things stand out. The first is how differently the assistants behave. ChatGPT, Perplexity and Gemini picked us up within weeks. Claude and Grok, which also search the web, did not, and in Claude's case the answer about me got worse rather than better. We do not know why. The second is that the places where we appeared were mostly the questions about us. When someone asks the open buyer question, which providers should I consider?, the answers moved around a great deal, and we showed up in some of them and not in others.

An answer is not a ranking

The finding that changed how I think about this business is not in the table above.

We asked the same question, in a fresh chat, on the same account, at different times of the same day. The answers did not hold still. A question that named us in the morning did not name us in the afternoon, and another one did the reverse. The answer about me flipped between describing my current companies and describing only an earlier job. Between the morning and the afternoon sessions, seven of the thirty-six answers we tracked changed from naming us to not naming us, or the reverse. As far as we know, nothing on our side had changed in those hours.

If you come from search engine optimisation, this is the uncomfortable part. A Google ranking is a position: you are third, or you are not. An AI answer is closer to a probability. Each time someone asks, the assistant assembles a fresh response, and your company has a certain chance of being in it. A screenshot of one good answer is an anecdote, not a result. So is a screenshot of a bad one.

That is why a real audit repeats its questions and tracks how often, on which questions, and in which direction. It is closer to following a sales pipeline than checking a ranking.

Checking the checker

A sceptical reader will already have spotted a weakness. Our working account had a history, and these assistants remember. The answers might partly reflect an account that had been talking about this company for weeks.

So two days later we repeated the whole audit from a brand-new account with no history, and then a second time that same evening.

The result was reassuring and inconclusive at the same time. The new account saw broadly the same picture: ChatGPT, Perplexity and Gemini knew the company and the founder, Claude and Grok did not. The details moved between the two runs, as expected. In one, ChatGPT described me only through my earlier employer. In the next, it correctly named my current company.

There was one difference I do not want to explain away. On the open buyer questions, which providers should I consider? and who are the leading providers?, our working account named us in seven of thirty-six answers across the three sessions. The new account named us in none of twenty-four. That could be the effect of the account's history. It could just as easily be ordinary variation. Two runs cannot tell the two apart, and we are not going to pretend otherwise.

What I would tell any business

I cannot prove what caused the change. But the month confirmed four rules I would apply to any client.

Audit before you build. The baseline is what makes everything after it readable. Without it, "the assistants know us now" is an impression. With it, it is a measurement you can repeat.

Be unmistakable before you try to be prominent. Our first problem was not that the assistants disliked us. It was that they did not know what the name meant. "GEO" turned out to be an earth observation programme, and in some questions an employment-services term. If your name collides with something bigger, the machines will pick the bigger thing. Say plainly, on your own pages, what you are.

There is no single "AI". In one month the same company was known to three assistants and unknown to two. Treat each as its own channel, with its own behaviour, and measure them separately.

Being named is not the same as being described correctly. In two Gemini answers our name was fused with another company's. A mention you do not control can quietly become a mis-statement of what you do. Reading the actual answers, not just counting appearances, is part of the job.

What the audit cannot tell us

We cannot say the website, the articles or the LinkedIn work caused the change. They all happened in the same thirty days, and we had one audit before and one after. We cannot say how large the change is, because the answers move from hour to hour. We cannot say it will last, because models and search indexes keep changing underneath us. And there is one factor we can neither control nor see: whether any assistant updated its model or search behaviour in the same period.

What we can say is narrower and still worth knowing. In September, when you asked the leading AI assistants about our company, the answer was empty. A month later, three of the five knew who we are, and the picture was different on each of them. We reached that point without a paid campaign, without paid placements and without ranking ourselves.

The audit itself is routine, and it repeats on a schedule, alongside the weekly technical checks. What matters is what follows each round: an analysis of what moved and what did not, and adjustments to our content, sources and structure accordingly. That cycle of audit, analysis and adjustment is the methodology, and we work it continuously.

Related insights

All insights & research →

About the author

Marin Tintchev is the founder of GeoSynapse and China GEO EU. He has 30+ years in ICT and international business across China and Europe, was educated at Tsinghua University, and is fluent in Mandarin.

Connect on LinkedIn →