A growing number of people ask an AI assistant for a recommendation before they open a search engine. “What is the best project management tool for a small team?” gets an answer with a short list of names. If your brand is on that list, you may get the visit. If it is not, you never find out you missed it. This is what people now call AI search visibility, or generative engine optimization, and the first step is simply measuring it.
What to measure
You do not need anything fancy to start. For each question, record these things from the answer:
- Was your brand mentioned at all?
- Where did it appear in the list, if it was a list?
- How was it described, positively, neutrally or negatively?
- How strongly was it recommended?
- Which competitors were named, and how did they rank?
Ask the questions your customers ask
Write prompts the way a buyer would type them, not the way a marketer would. Include comparison questions, “best for” questions and plain how-do-I questions in your category. Then ask several different assistants, because they do not agree with each other.
One run tells you very little
Answers change from one run to the next, and between providers. A single result might be luck. What matters is the trend across many questions and many runs, so keep every raw answer and compare over time. If your mention rate climbs from one month to the next across the same set of prompts, that means something. If one answer flips, it does not.
What I built to do this
So I built MentionScope, a Windows desktop app. You add your brand and your competitors, write your prompts, paste in your own API keys for the assistants you want to test, and run a search. Prompts go out to every enabled provider in parallel, with progress shown live. A failure on one request never stops the rest, and a cancelled run can be resumed later.
One design choice matters for accuracy. Instead of trying to pick answers apart with text patterns, every raw answer goes to a single consistent model that returns structured results, which are then validated again in code. For each brand it records whether it was mentioned, its rank, the sentiment and how strongly it was recommended. Every run is stored, so the dashboard can show visibility, average rank and a leaderboard with trend arrows.
What I would and would not claim
Nobody can guarantee that an assistant will recommend you, and anyone who says otherwise is selling something. What I can say is that you cannot improve what you do not measure. The things I treat as worth trying are ordinary good practice: clear pages that say plainly what you do, consistent descriptions of your brand across the sites that mention you, structured data, and being cited by sources people trust. I treat them as hypotheses and watch the numbers to see whether they move.
The MentionScope preview on this site uses made-up brands and numbers. You can look through it on its walkthrough page.