A score dropped

A lower number is not always a lower standing. Most drops come from how you are comparing, from ordinary variation between runs, or from a change to what you track. Work through the checks below in order: the early ones take seconds, and only the last one is about AI answering differently.

Check how you are comparing#

• The change is relative. The change beside Visibility on the Visibility page is a percentage of the previous value, not a difference in points. Going from 20% to 15% shows as a 25% drop.
• Windows, not runs. Each card compares the date range you picked with the range of the same length before it, for example vs previous 30 days. Changing the range changes both sides of the comparison.
• Home compares runs. The chip under the visibility score on Home shows a change in points, from the most recent run that finished more than a week before today (or your oldest run, when none did) to the latest run. It reads "vs last week" whatever the gap, and it can differ from the Visibility page for the same brand.
• Filters make numbers jumpier. One engine, one category or one theme holds fewer answers than your whole set, so it moves more from run to run.
• A carried value is not a drop. When the range holds no run, the cards show the last completed run before it with no change beside them, and the trend chart names that run's date. See Carried values.
• Position moves the other way. A lower average position is better, so its arrow points up when the number falls. See Average position.

Check for noise#

AI engines do not give the same answer twice, so your figures move from run to run even when nothing has changed. Ranqo measured how far they move across the brands it tracks. For 95% of pairs of consecutive runs, the change was no larger than:

• 15 points for overall visibility,
• 20 points for visibility on one engine,
• 27 points for visibility in one prompt category.

A drop inside those bands from one run to the next is usually noise. The Action Center uses the same bands: it does not show a measured lift smaller than them as a number. Open Snapshots and look at the runs one by one: a dip that recovers on the next run is noise, a drop that holds for several runs is worth investigating. See Noise bands.

Check whether what you track changed#

Visibility is a share of the answers to the prompts you track, on the engines you track. Change either and the share can move without any answer changing.

• Prompts added. New prompts on topics where AI rarely names you lower the average. This is common right after you generate more prompts.
• Prompts deleted. A deleted prompt's answers go with it, from past runs too, so earlier figures can change as well.
• Engines added or removed. Visibility is weighted by engine. Adding an engine where you are rarely named lowers it, and so does losing an engine where you were strong, for example after a plan change. See Visibility.
• Brand settings. Removing an alias means fewer mentions are matched, and a new language means every prompt is answered in that language. Both apply from the next run. A new market applies to prompts written after the change: existing prompts keep the market they were saved with. See Brand settings.

To see what changed, turn on Changes above the trend chart on the Visibility page, which appears when the range holds a change: each change to your prompts or engines is marked on the chart. A Tracking config changed banner appears when the latest run differs from the one before it, and a run's page in Snapshots shows Tracking configuration changed since last run with how many prompts were added or removed and which platforms.

Check the run's coverage#

Open the run in Snapshots:

• Partial coverage means some queries could not be answered. The scores are measured over the answers that did return, so they describe less of your coverage than a full run and move more.
• N of M queries used historical data means failed queries were filled with each prompt's most recent successful answer on that engine.

Google AI Overviews can also lower a score without a failure. When Google shows no AI Overview for a prompt, that answer counts as one that does not name you, so a week in which Google shows fewer overviews for your prompts reads as a drop on that engine.

Check how competitors are counted#

Share of Voice and your rank depend on how competitor mentions are counted, not only on yours:

• Merged or separated names. Ranqo groups the spellings of one company under one competitor. Separating a spelling, or grouping it back, changes that competitor's mention count and can move your rank and your share. See Competitors.
• Your watchlist. A competitor you track always counts toward the Share of Voice denominator, even if AI named it once. Adding competitors to the watchlist can lower your share a little.

See Share of Voice and Competitor identity.

Find a real change#

If none of the above explains it, AI is answering differently. Narrow it down:

1. Find the engine or category

   On the Visibility page, Platform Breakdown and Category Visibility show which engine or category fell. Each bar has a tick at the card's average.

2. Find the prompts

   On Prompts, open the Tracking tab. Each prompt's Visibility, its Mentions and the Engines logos, dimmed where the latest answer did not name you, show which questions you lost.

3. Read what changed

   Open a prompt's insights page. Brand Rankings shows who is named instead of you, Performance by Platform shows where, and AI Response History lets you read the answers from before and after the drop. An answer opened there is the reformatted version when there is one, which covers only the first 8,000 characters of a long answer; a brand named after that point is not counted. See Why numbers change.

4. See what AI read

   On the same page, Citation Sources lists the domains AI cited for the prompt, and whether each one mentions your brand. A new comparison site or forum that leaves you out can explain a lost mention. Sources shows the same across all your prompts.

Then act on it: the Action Center turns your latest results into prioritized recommendations, and measures each one you complete against the same noise bands.

Related#