# Noise bands > How much AI visibility moves between runs when nothing changed, how Ranqo measured it, and where a smaller lift reads No detectable change. Source: https://ranqo.ai/docs/methodology/noise-bands AI answers change from run to run even when nothing about your brand has. The same prompt, asked a week later, can name you one time and leave you out the next. A noise band is the size of that ordinary movement for one kind of metric. In the [Action Center](https://ranqo.ai/docs/guides/action-center), a recommendation's drawer uses it to decide whether a reading is a change or **No detectable change**. ## The bands In the drawer, a lift must be larger than the band for its metric, in percentage points, to show as a number: | Metric scope | Change needed to count as real | | --- | --- | | Overall visibility | 15 pts | | One engine | 20 pts | | One prompt category | 27 pts | Each recommendation is measured on one metric, named under **Measured** in its drawer: **Overall visibility**, one engine's mention rate (such as *Gemini mention rate*) or one prompt category's visibility (such as *Discovery visibility*). The band for that kind of metric applies. ## How the bands were measured The bands are measured, not chosen. In August 2026 Ranqo took every pair of consecutive completed runs for brands with at least three runs: 81 brands and 851 runs. For each pair it computed the three kinds of metric with the same calculation the Action Center uses to take a reading, and recorded how far each one moved. That gave 770 run-to-run comparisons of overall visibility, 2,290 of a single engine and 3,659 of a single prompt category. Each band is the 95th percentile of those movements, rounded to a whole point: about 95 of every 100 measured run-to-run changes were smaller than the band. Using the product's own calculation matters. A band worked out any other way would describe a different number from the one you see. The bands describe the change between two consecutive runs. A reading compares the run a recommendation was written from with a later run, usually several runs apart. ## Why the bands are wide - **Each answer is a large step.** A brand tracking 20 prompts on an engine moves that engine's rate by 5 points with every single answer that names it or stops naming it. - **Engines vary their answers.** A mention seen for the first time is still there in the next run only about half the time. - **A category is the noisiest.** It holds only some of your prompts, so each answer in it is a bigger share of the total. That is why the category band is the widest, even though it feels like it should be the most precise. - **Overall visibility is the steadiest.** It is combined across every prompt and engine, so single answers move it least. ## What happens inside the band In the recommendation's drawer, a reading inside the band is shown as **No detectable change** in place of a number. After the verdict window has passed, the drawer adds that the metric moved less than its normal week-to-week swing; see [Measurement windows](https://ranqo.ai/docs/methodology/measurement-windows) for how that label is worked out. When the reading compared two runs of answers, the drawer shows answers won instead of points, whatever the size of the change in points. See [Confirmed wins](https://ranqo.ai/docs/methodology/confirmed-wins). Other places show the reading without the band: - The chip on a Done card, and the lift column in the List and Matrix views, print the reading in points at any size, green for zero or more and red below zero. - The notice after a reading reports the change in points at any size. When a chip and the drawer disagree, the drawer's verdict is the one that accounts for the band. **Verified Lift**, the fourth card on the page, adds up the positive readings larger than the overall visibility band of 15 points. It applies that band to every card whatever its metric, although its tooltip refers to each metric's own band, so an engine or category reading can count there while the drawer reads **No detectable change**. The [Action Center](https://ranqo.ai/docs/guides/action-center#verified-lift) guide lists what else it includes and leaves out. Without a band, a total like Verified Lift would become a running total of chance. A set of recommendations nobody acted on would still produce a positive total, because some runs happen to move up. The expected outcome in each recommendation's drawer uses the same band: when there are enough answers left to win, it says how many you would need before the change stands out from normal movement. ## A small real change can still be found Inside the band does not mean nothing happened. It means the overall number cannot tell a real change from an ordinary week. When it writes a recommendation, Ranqo also records which specific answers you were missing and later checks whether you won them and kept them. That can confirm wins that are too small to clear the band. See [Confirmed wins](https://ranqo.ai/docs/methodology/confirmed-wins). ## Where the bands apply In the dashboard, the bands are used only by the Action Center drawer's **Lift** step, the **Verified Lift** card (overall band only), each recommendation's expected outcome, and the **Rising** marker on the Competitors page's Watchlist (overall band only), which a brand earns only by gaining more than the band. The Action Center's lift chips and the notice after a reading, and every other figure on the Visibility, Competitors and Sources pages, show every change as measured, so read a small move there with the same caution. A change of a few points between two runs is usually noise. ## Related - [Measurement windows](https://ranqo.ai/docs/methodology/measurement-windows): How many runs a reading and a verdict need. - [Confirmed wins](https://ranqo.ai/docs/methodology/confirmed-wins): Counting the answers you won, one prompt and engine at a time.