Visibility in AI answers
What assistants say about you in a given country and language, and whether you are named at all. It can only be asked from inside that market.
The answer now arrives before the search
When a customer asks an assistant "what should I use for X", if your name is not in the answer, you are not in that comparison. This is not about clicks declining. It is about whether you are a candidate at all.
Why you cannot check this yourself
An assistant's answer changes with where it is asked from and in what language. The same question from Japan in Japanese and from Germany in German returns different companies, different sources, in a different order.
From your own office you can only see how you look in your own language.
And checking by hand means one question, once. What is actually needed is several questions × several languages × several assistants × continuously, which is not work a person does.
What can be measured
| Whether you are named | Per question, language and region |
| How you are described | Whether the features, strengths and intended customer come across correctly |
| Who you are placed beside | Which competitors appear as alternatives |
| What is cited | The URLs the answer draws on |
| Whether it changed | Did the answer move after you changed the site |
What to fix on your own site
An assistant reads the same pages a person does, but differently.
- Structured data (JSON-LD) that is valid *and* matches what the page says
- Headings and Q&A shaped so an answer can be lifted out of them
llms.txtand a sitemap that say what you want read- Whether AI crawlers can reach you at all — country-level IP rules and bot defences often block them without anyone internally knowing
That last one is the commonly missed item. A company blocking a crawler it never meant to block is not unusual.
Who uses it
- Brand owners — to check continuously whether they are a candidate
- Agencies and consultancies — measured per client, before and after
- Companies selling in several countries — because the answer differs by country, it has to be measured by country
One caveat
Assistant answers vary. The same question does not return the same answer every time, so a single observation is not evidence. It has to be measured continuously and read as a trend.
Kansoku records, for each observation, when it was made, from where, of which model, and what was asked. The variation itself is recorded, which is what lets you compare before and after.
Last updated