Articles / Measurement
How often should you measure AI visibility? What weekly and monthly can detect
Run as often as each run is big enough to detect the change you care about. At 75 answers a run only a 12-point rise is clear; at 600, about 3 points.
- Published
- Reading time
- 6 min read
- By
- AmpliRank
- Tags
- AI Visibility · Cadence · Sample Size
Measure AI visibility as often as each run is large enough to show the change you care about. With 75 answers a run, only a rise of about 12 points from a 30% rate is a clear change; with 600 answers, about 3 points is. So weekly runs suit large answer counts or watching for big moves, and monthly runs suit small budgets and changes of a few points.
In short
- Cadence and run size trade off: the same monthly answers can make four small weekly runs or one larger monthly one.
- A small run can only call a large change. At 75 answers, the smallest clear rise from 30% is about 12 points; at 300, about 5.
- Weekly runs catch big moves early: an engine update, a competitor launch, a page that suddenly gets cited.
- Monthly runs, or month-on-month comparisons on the same questions, see the smaller changes most content work produces.
- Whatever the cadence, keep the questions and engines fixed, or runs can’t be compared.
Why does cadence depend on run size?
Cadence depends on run size because a change is only clear when the 95% range of the difference between two runs excludes zero, and that range is wider when each run has fewer answers. Is that change real? works through the paired comparison, and the pillar guide on measuring AI visibility sets out why a rate needs its range.
A run’s size is questions × engines × repeats. With a fixed number of answers a month, running more often means smaller runs. Twenty-five questions asked to 3 engines once is 75 answers; four of those runs a month is 300 answers, the same as one monthly run of 100 questions. Which split is better depends on the size of the change you expect. The glossary defines measurement cadence and the other terms used here.
What can a run of a given size detect?
A run can detect a change only when the change is larger than the noise between two runs of that size. The table gives the smallest rise from a 30% rate that is a clear change, for five run sizes and two levels of churn.
The assumption, stated plainly: both runs ask the same questions to the same engines; the earlier run named you in 30% of answers; and a set share of all answers loses you between runs by chance, either 5 in every 100 (a calm week) or 10 in every 100 (a volatile one). The rise comes from answers that gain you on top of that. Each figure is the smallest net rise whose paired 95% range just excludes zero.
| Answers per run | Smallest clear rise, 5 in 100 lose you | Smallest clear rise, 10 in 100 lose you |
|---|---|---|
| 75 | 12.0 points | 14.7 points |
| 150 | 7.3 points | 8.7 points |
| 300 | 4.7 points | 6.0 points |
| 600 | 3.0 points | 4.0 points |
| 1,200 | 2.0 points | 2.8 points |
Three things follow. A run of 75 answers can tell you that something large happened, not that a page you shipped moved you by a few points. Doubling the run size shrinks the smallest clear change to about two thirds, not to half. And churn matters: when more answers flip each way by chance, the same run needs a larger net rise before it stands out.
The two churn levels are assumptions for illustration, not measurements. To see your own, take two runs, count the answers that named you last time and not this time, and divide by all answers.
Weekly or monthly: which should you choose?
Choose weekly runs when you need to know about large changes quickly, and monthly runs when the changes you care about are a few points. Most teams need both kinds of reading, so the question is how to split a fixed number of answers.
| Weekly runs | Monthly runs | |
|---|---|---|
| Answers per run, from 300 a month | 75 | 300 |
| Smallest clear rise from 30% (5 in 100 lose you) | about 12 points | about 5 points |
| Good for | engine updates, competitor launches, a page suddenly cited or dropped | the slower effect of content and third-party work |
| Risk | reading noise as news, or “no clear change” every week | learning about a large move weeks late |
A middle path is to run weekly and compare month against month: the four weekly runs of one month against the four of the next, on the same questions and engines. That gives an early warning each week and a larger comparison each month. Treat the monthly comparison with care, because repeats of the same questions tend to agree with each other: four runs of 75 answers to the same 25 questions carry less information than 300 answers to 100 different questions, and an honest range for them is wider than the table’s.
When should you run more often?
Run more often when something has changed that could move your rate quickly, and keep the run large enough to see it:
- After a launch or a big page change. Weekly runs for the following month show whether the engines picked it up, as long as the change is large enough for the run size.
- When a competitor moves. A rival’s launch, pricing change or press coverage can change who is named in category answers.
- When the engines change. Model and search updates can move answers across a whole category at once.
- When you report weekly anyway. A weekly brief needs a weekly run, but it should say “no clear change” most weeks rather than narrate noise.
Outside those moments, a monthly run of the same questions is often enough for a team that acts on content and sources over weeks, not days.
What breaks a cadence?
Changing what is measured breaks a cadence faster than any schedule choice. Each run is comparable with the last only on the questions and engines both asked:
- Changing questions. Add new questions as a new set; keep the old set unchanged so the trend continues.
- Changing engines. A run that adds or skips an engine is compared on the engines both runs share.
- Changing repeats. More repeats per question make a run larger but not more varied; how many answers do you need? explains why distinct questions add more than repeats.
Where to start
Write down the smallest change in your rate that would make your team act. Find the run size that can detect it in the table, or in the free AI visibility calculator, then divide your monthly answers by that size: the result is how many runs a month you can afford at that sensitivity. AmpliRank schedules runs weekly or monthly, and its plans are sold in answers a month, so the same arithmetic sizes a plan.
Sources
- Quinn McNemar, Note on the sampling error of the difference between correlated proportions or percentages, Psychometrika 12 (2), 153–157, 1947, accessed 14 November 2026.
- Edwin B. Wilson, Probable Inference, the Law of Succession, and Statistical Inference, Journal of the American Statistical Association 22 (158), 209–212, 1927, accessed 14 November 2026.