Ask ChatGPT to name the leading companies in your sector and you will get an answer. It may even include you. But it tells you surprisingly little about whether AI tools will point buyers toward your business when it counts.
In our monitoring work for a B2B technology client, we tracked how often the company was cited across ChatGPT, Gemini and Claude. We ran two kinds of prompts. Broad category questions asked which providers led the market. Specific, outcome-led questions asked how a named organisation had achieved a particular result.
The gap was striking. Specific prompts produced a citation rate of around 47%. Broad category prompts managed around 17%. Same company, same models, same month. Only the question changed.
Why the question changes the answer
Broad prompts invite AI models to summarise a whole market. They lean on the most established names, comparison sites and the companies with the loudest owned content. Unless you already dominate your category, you are competing for a handful of slots against everyone.
Specific prompts work differently. When someone asks how a council cut call volumes, or how a hospital group improved patient messaging, the model goes looking for evidence. It pulls from case studies, trade press coverage and named examples. If your story exists in those places, you become part of the answer.
This mirrors how buyers use these tools. Few decision-makers ask "who are the top vendors?" and stop there. They ask about problems, outcomes and proof.
What this means for measurement
Many companies test their AI visibility by typing their category into ChatGPT and seeing what comes back. It's a reasonable start, but it can mislead in both directions. A poor result may hide strong visibility on the questions buyers ask. A good result may simply reflect brand recall instead of genuine authority.
We now report the two separately. Category prompts show how visible a company is when buyers are browsing. Specific prompts show whether it owns the conversations where decisions get made. Blending them into one score hides both stories.
What this means for PR
This is where earned media earns its place. Outcome-led questions are answered by outcome-led stories. A trade press feature on how a client solved a real problem gives AI models something concrete to cite.
It also explains why we build our work around pillar topics and sub-topics. A pillar topic is the broad territory you want to own. Sub-topics are the specific problems, outcomes and questions beneath it. Each placement, case study and article adds weight to a sub-topic, and together they build authority on the pillar.
One-off coverage fades. A connected body of work on a defined topic compounds.
Three things to try this month
Start by writing ten questions your buyers ask before they buy. Phrase them as problems and outcomes, not categories. Run them across ChatGPT, Gemini and Claude and note who gets cited.
Next, look at the sources behind those answers. Which outlets, case studies and pages keep appearing? Those are the places your story needs to live.
Finally, map your existing coverage against those questions. Most companies find strong stories that never made it into the places AI models read.
The bigger picture
Only around one in five marketers is tracking AI visibility at all. Companies that start now, and measure it properly, have a real chance to claim their topics before competitors notice the shift.
Want to see how your brand shows up? Request a free AI visibility snapshot and we'll show you where you stand.
