Why does the overview look empty?

Either no model key, so nothing has run, or the first scheduled batch has not fired yet. The page says which rather than drawing a flat line at zero.

What does the date range change?

Which runs are averaged, not what was measured. A narrow range on a weekly cadence leaves one batch per prompt, which is why short ranges look volatile.

There is a banner about run health. Should I worry?

Read it. It appears when probes are failing, which is nearly always the model key: no credit, a revoked token, or a provider rate limit.

Reading the overview

What each panel on the presence overview is for, the order to read them in, and which numbers move fast enough to be worth checking weekly.

Updated October 20, 2026

The overview answers one question: are models recommending us, and is it improving. Everything on it is computed from stored runs, so any figure can be opened back to the answer that produced it.

The order to read it in

Any banner at the top. A run-health warning or a trial notice changes how you should read everything below it, so it comes first.
Visibility Score. The headline rate with its band and its change against the previous period.
Visibility Score by Platform. The same rate on each assistant. The spread here is usually wider than any weekly move.
Share of voice. Your position against the competitors you registered.
Visibility by topic. Where the headline figure is actually coming from.
Counts. Prompts, probes, batches, fanouts, sources. Context for every rate above.

What moves fast and what does not

Fan-out searches change constantly: they are generated per answer, so a new set appears with every batch.

Cited sources change over days as models re-fetch pages and as third-party pages update.

Visibility Score moves over weeks, and genuinely moves when the sources that get cited start carrying you. A score that jumped overnight is almost always sampling noise or a change you made to the prompt set.

Training-mode figures move over months. Do not check them weekly.

The counts are diagnostics

Easy to skim past, and they answer the three most common confusions.

Probes against batches. A batch is one measurement; a probe is one answer inside it. Rates are per probe, and the band depends on the probe count.
Errored and degraded runs. Probes that failed or came back incomplete. A handful is normal. A lot means the key or the provider.
Fanouts: total, optimizing, ranked. How much of the work queue you have actually committed to.

Clicking through

The overview is a set of entry points rather than a destination. Clicking a point on the trend opens the batch behind it; clicking a platform filters to it; clicking a topic opens its prompts. When a number looks wrong, two clicks usually reach the answer text that explains it, and reading that beats reasoning about the chart.

Note

Every rate on this page is a grounded rate by default. The mode selector changes that, and a figure read in another mode should never be compared against one read in this one. See grounded and training.

Questions about reading the overview

Why does the overview look empty?: Either no model key, so nothing has run, or the first scheduled batch has not fired yet. The page says which rather than drawing a flat line at zero.
What does the date range change?: Which runs are averaged, not what was measured. A narrow range on a weekly cadence leaves one batch per prompt, which is why short ranges look volatile.
There is a banner about run health. Should I worry?: Read it. It appears when probes are failing, which is nearly always the model key: no credit, a revoked token, or a provider rate limit.

Was this page helpful?