THE SHORT ANSWER
Build a fixed list of twenty to fifty prompts a real buyer would type, run them across each engine on a set schedule from clean sessions, and record three things per prompt: whether you were named, whether you were linked, and which competitors appeared. The absolute numbers are noisy; the trend across weeks is what you act on.
Build the prompt list from real buying language
Take the questions buyers actually ask, not the keywords you rank for. Mix four kinds: price questions, comparison questions, how-do-I questions, and vendor-selection questions such as who is best for a given need in a given country.
Twenty prompts is enough to see movement. Fifty is better. Beyond that the maintenance cost outweighs the extra signal.
Run them properly or the data is worthless
Clean sessions. Prior conversation history contaminates the answer badly.
Same wording every time. Rephrasing changes retrieval, so a change in result may just be a change in your prompt.
Fixed schedule. Weekly or fortnightly. Comparing an ad-hoc check against a memory of last month is not measurement.
Record the source list, not just the answer. Knowing which page was cited tells you what to write next.
The three metrics worth keeping
Mention rate, the share of prompts where you were named. Citation rate, the share where you were linked. And share of voice, your mentions against the total mentions of you and your named competitors.
Share of voice is the most useful of the three, because it separates your own progress from a general change in how the engine answers that category.
The leading indicator that costs nothing
Branded search volume. When assistants start naming you, people search your name. That shows in Search Console weeks before it shows in referral traffic, and unlike prompt logs it requires no manual work.
Related answers
Frequently asked questions
How often should I check AI visibility?
Weekly or fortnightly, with identical prompt wording each time. Ad-hoc checks against remembered results are not measurement.
Why do I need a clean session?
Prior conversation history steers the answer heavily. Without a fresh session you are measuring your own history, not the engine's default.
What is the single best metric?
Share of voice: your mentions against the total for you plus named competitors. It separates your progress from category-wide shifts in how the engine answers.
Can I see AI referral traffic in analytics?
Partly. Some engines pass a referrer and some do not, and answers that name you without linking generate branded search instead of referral traffic.
NEED A CLEARER PLAN?
Let’s turn your next move into momentum.
Talk to us →