How to measure your AI visibility

What you're actually measuring

AI visibility is how often, and where, AI engines name you when someone asks about your category. It sits apart from keyword rankings and traffic. What you care about here is whether a model brings you up when a buyer asks it for a recommendation, and how it talks about you when it does.

Start from the questions buyers ask

Build a fixed set of prompts and write them the way buyers write them, not the way a keyword tool spits them out. Category questions (“best X for a small team”), comparisons (“A vs B”), and the use-case questions people ask when they’re stuck. If you have sales calls or support tickets, steal the wording straight from them.

The reason to fix the set is that you’re going to run it again and again. So once it reflects how your market talks, leave it alone. Changing the prompts every week means you can never tell whether the number moved or the question did.

Run it across engines, more than once

Buyers don’t all use the same tool, so send every prompt through ChatGPT, Gemini, Claude and Perplexity. Then run each one several times. The models sample their answers, which means the same prompt can name you on the first attempt and forget you on the third. That wobble is the whole reason one screenshot settles nothing, and the reason you repeat.

What to write down on every run

• Whether you got named at all. The plain yes or no that everything else hangs off.
• Where you landed. First brand mention out of the gate reads very differently from a name tucked into a closing list.
• The sentiment.A warm recommendation and a grudging “it exists” are both mentions, and you want to tell them apart.
• The sources it cited. The pages a model pulls from are the ones deciding what it says about you, so log the URLs.

Four numbers worth trending

Once you’ve logged enough runs, the raw notes turn into metrics you can put on a chart and follow.

• Mention rate. How often you show up across the whole prompt set. This is your baseline: the share of buyer questions where you make the answer.
• Share of voice.Your mentions against your competitors’ across the same prompts. This is the competitive read, and share of voice can fall while your own mention rate holds, if a rival is climbing faster than you.
• Sentiment. How favorable the mentions are, not just how many. Ten flat listings and ten enthusiastic ones read the same on a count and very differently to a buyer.
• Citation share.Whose pages the engines cite when they answer. If competitor URLs keep turning up and yours don’t, that’s a map of where the work is.

The habit that matters most

If you keep one thing from this, keep this: measure the trend, not the snapshot. Run the same prompt set on a schedule, weekly or monthly, and watch the four numbers move. A single answer carries enough noise that reacting to it will have you chasing ghosts. A line drifting up or down over six weeks is real, and it’s the only thing that tells you whether the work you’re doing is landing.