You type "best ERP vendor in the Netherlands" into ChatGPT. You get an answer with five names. Is your company one of them?
Usually not. And that's not bad luck. It's measurable, and you can test it yourself today.
In ten minutes you'll know whether AI assistants mention you. Below is how we approach this for IT companies, which mistakes we keep seeing, and what to do with the outcome.
Your buyer's orientation phase has moved. Not entirely, but enough to notice.
An IT manager used to type "compare ERP vendors" into Google, open six tabs and read himself stupid for an afternoon. Now he asks an assistant.
There is no second page. No position 6 that someone still scrolls past. If your company isn't in that answer, you're simply absent in that phase of the search.
We see that many B2B companies don't know whether they appear in those answers. That's the real problem.
A mention is not the same as a ranking. Google puts you at position 3 and you get a link. An assistant names you in running text, with a source underneath.
Three things to remember.
An assistant picks a handful of sources and summarizes them. It doesn't read ten pages and weigh them. It takes what it finds and writes an answer.
AI crawlers often don't execute JavaScript. Content that only loads client-side doesn't exist for them. So your beautiful interactive comparison page can be invisible.
And it's not about one assistant. ChatGPT, Gemini, Claude and Perplexity pull from different sources and give different answers. Being in ChatGPT doesn't automatically mean being in Perplexity.
Sounds logical. Yet it surprises almost everyone the first time they test it.
Start with the questions your customer asks before he knows you exist.
Not "why is [your brand] good". Nobody types that. The questions from the orientation phase.
Five is enough for a first measurement. If you want to measure seriously, you'll end up with twenty questions. Split those across three categories: generic money questions, comparison questions and specific need questions.
Write them down before you start. Otherwise you'll invent prompts along the way and you can't compare next month.
Open the three assistants in separate tabs. Paste the same prompt into all three. Don't just read whether you're mentioned, but what else the answer says.
Watch four things per prompt.
That last one often gets forgotten. A mention isn't automatically a good mention. We see answers where a company is named based on a page from 2021 that was never updated.
Perplexity shows its sources most clearly. Use that as your starting point, then you know which pages carry the answer.
Now comes the boring part that makes the difference. Put it in a spreadsheet.
One row per prompt. Columns for ChatGPT, Gemini and Perplexity. In those you fill in: mentioned (yes/no), position in the answer, and the source URL underneath it.
After fifteen prompts you'll see patterns you'd never have noticed in isolation. You show up in Perplexity for technical questions, not in ChatGPT for commercial ones. Or you're only mentioned when the prompt contains your brand name, which isn't visibility but confirmation.
Save the date and the exact prompt text. Models change. Next month you can then see whether a difference comes from your own work or from a model update.
You only show up when someone already knows your name. That means the assistant knows you from its training data, but never recommends you to someone still searching.
Competitors dominate through listicles. Period.
Your content doesn't load for crawlers. Your site looks great in the browser. The assistant sees nothing, because the text only appears after a script it doesn't run.
Your source material is outdated. The assistant cites a page that's three years old. You've evolved, the answer hasn't.
You're summarized incorrectly. You are in there, but the answer doesn't match your proposition. Often because your site talks to different audiences and the assistant picks one.
Recognize one? That's your starting point.
The baseline tells you where you stand. Generative engine optimization, GEO for short, is the work to get from there to a mention in the answers.
Four things we do, in this order.
Tech first. Making sure AI crawlers can read your site. That means rendering content server-side, allowing the right crawlers in robots.txt, and setting up schema.org. Organization, Service and FAQPage are the types that matter most for IT companies. Without this step, every content effort is water into the sea.
Then the content. Writing pages an assistant can cite. Answer-first, with one clear question per page and the answer in the first paragraph. That's exactly why we build every article on wildbos.com that way.
Then the mentions outside your own domain. Reviews on the right platforms, press releases to national media, and yes, the listicles. An assistant prefers to cite a source that isn't yours.
And then measure. Every month the same prompts, in four assistants, with the source URLs underneath. That's the only number that's honestly trackable.
If you don't want to run the first analysis yourself, the free AI visibility scan is a fine start. You'll see at once where you stand right now.
Every month. Not every week.
Models change faster than your content. Test weekly and you're mostly measuring noise and drawing conclusions from coincidence. Test quarterly and you intervene too late when a competitor has taken over the answer.
Once a month, the same fifteen to twenty prompts, the same four assistants. That costs you an hour and yields a chart your board understands.
Put it in your calendar. That's the whole secret.
Ask ChatGPT, Gemini and Perplexity the questions your customer asks, and read the answers. If your name isn't mentioned, you're absent in that phase of the search. Do this with fixed prompts so you can compare your measurements.
No. Google sends visitors to your site through positions and links. An AI assistant names you in an answer and picks a handful of sources itself. You can rank excellently and still be completely absent from AI answers.
An assistant decides per question which sources it uses, and that changes with every model update.
Today's ten minutes give you a spreadsheet, not a strategy. That's exactly why it works. You now have a starting point you can repeat every month, and a list of the places where things go wrong.
Want us to run the baseline with you, with the source URLs included? You can schedule an audit call. We'd rather show you where you stand first than sell you something.