Half the work of choosing a vendor now happens before anyone fills in a form. A buyer describes their problem to an assistant and gets three names back.
You can find out whether you are one of them in about thirty minutes. Here is how to run the test so the answer means something.
Ask the questions a buyer would ask
Not your brand name. Asking an assistant about your company tells you what it knows about you, which is a different question.
Write ten prompts the way a buyer types them. Who can run outbound sales for a B2B software company in Germany. Best sales outsourcing companies in Europe. Should we hire an SDR or outsource. What does outsourced lead generation cost.
Mix the stages. Two or three at the beginning of a search, four or five comparing options, two or three about price and terms.
Run each prompt more than once
This is the step everyone skips and it is the one that makes the result real.
These systems are not deterministic. Ask the same question five times and you can get five different lists. One run is an anecdote.
Run each prompt three to five times, in a fresh chat each time, with memory and personalisation off. Record how many times out of five you appear, not whether you appeared.
Cited three times out of five is a measurement. Cited once is noise. Tracking the rate month over month is the only way to know whether anything you changed worked.
Record four things
Whether your company is named at all.
Which companies are named instead, and in what order.
Which pages the assistant cites, yours or anyone else's. This is usually the most useful column, because it shows which kind of page the model trusts for that question.
How you are described when you do appear. Being named as the cheap option is a different problem from not being named.
Read the pattern, not the single answer
If you never appear and the same three competitors always do, look at where those competitors are cited from. It is rarely their own website. It is usually a review site, a comparison article someone else wrote, or a forum thread.
If you appear but the description is wrong or dated, the problem is entity data. Your own pages, your profiles and your listings are telling different stories about what you do.
If you appear for the simple questions and vanish on the comparison ones, you are missing the content that answers a comparison. That is a page you have not written yet.
Getting cited and getting recommended are not the same thing. A model can quote your blog post in an answer that recommends somebody else.
What to fix first
Start with the questions where a competitor is named and you are not, and where the cited page is one you could have written. That is the shortest distance between the test and a result.
Then go after the third party surfaces. Review profiles, industry roundups and comparison pages carry more weight in these answers than your own site does, and most B2B companies have none of them.
Run the same ten prompts again in a month with the same method. If the rate moved, you know what moved it.
If you would rather not run this yourself, it is the first thing we do on an SEO, GEO and AI visibility engagement.