AI Visibility in B2B Is Not a Single Number
Tilman Bohn, our Director Technology, has published a guest article on this topic on the expert blog of Germany's Federal Association of Industrial Communication (bvik): AI Visibility Analyse im B2B: Der Kanal ohne Reporting (in German).
Why a single number is not enough
For industrial companies, AI visibility breaks down by market, language, industry, application and model. Four findings from the article show why.
1. Being mentioned is not the same as being cited
In one measurement, a company had no Spanish-language website. Even so, its visibility score for Spanish questions was only slightly below its score in the languages its website covered. Citations, however, dropped to almost zero.
The reason: LLMs answer in almost any language, but rarely link to sources in a language other than the question's. If you want to be cited in a market, you need content in the local language - on your own website and on third-party sites.
2. China needs its own models and its own questions
China has a largely independent AI ecosystem, with models from ByteDance, Tencent, Baidu and Alibaba. It is used mainly through mobile apps, closely integrated with the providers' other services.
Translating Western questions is not enough for a reliable measurement. Real-world measurements showed at times dramatic differences between question sets created by Western models and those generated "natively" by Chinese models. What matters is how users on the ground actually phrase their questions.
3. A measurement only becomes robust with enough questions
A manufacturer can be the gold standard for one application and be completely absent in a closely related niche. Well positioned in the DACH region, invisible in France or the US. A snapshot from a handful of prompts will not show this.
The recommendation from practice: around 20 questions or more per dimension value, meaning per industry, product or application. With a broad portfolio, that quickly adds up to more than 1,000 questions. To keep follow-up measurements comparable, question sets, brands, languages, markets and model versions are fixed and versioned. Why a handful of prompts is not enough is also explained in our article Are 50 Prompts Really Enough to Measure AI Visibility?
Separating brand and non-brand questions is just as important. For neutral questions, what counts is whether your brand is mentioned at all. For questions that already name the brand, what counts is how it is presented: tone, soft USPs, and any gap between the answer and your own messaging.
4. Visibility is not market share
AI answers regularly feature suppliers that are on no one's competitor list. Revenue and brand awareness only have an indirect effect. What has a direct effect is how well structured and accessible publicly available information is.
If a market leader hides its expertise behind logins, forms or marketing jargon, language models struggle to use it. A smaller supplier with freely accessible, well-structured technical articles and product data then appears prominently - while the industry leader is missing. Such shadow competitors slip through the cracks of traditional market monitoring. A sufficiently detailed AI visibility analysis makes them visible early on.
Conclusion: a process, not a metric
In B2B, AI visibility is a measurement system: it has to reflect the markets, languages and models where buying actually happens, and it has to be repeatable in exactly the same way six months later. Getting started can still be lean: one market, one language, a few dimensions and competitors, measured against the most widely used models.
This is exactly what we built aiva for - the AI Visibility Analysis by effective world, covering Western and Chinese models, with customer-specific dimensions and consultant-led evaluation.
➡️ Read the full guest article on the bvik blog (in German)
➡️ Learn more about aiva
➡️ Download the sample aiva analysis (PDF)
➡️ Request your own aiva analysis